Organizational Design for the GenAI Age Reflex Area
2026-09-29
Table of Contents
- Establishing the Case and Requirements for Organizational Redesign
- Reconstructing the Architecture of Work
- Redesigning Around New Bottlenecks and Organizational Constraints
- Redesigning Role and Job Architecture
- Repositioning Specialist Expertise
- Designing Human-AI and Human-Agent Work Configurations
- Redesigning Team Composition and Support
- Redesigning Managerial Work
- Reconsidering Management Layers and Spans of Control
- Redistributing Decision Rights and Authority
- Designing Human-AI Decision and Accountability Boundaries
- Designing Independent Challenge and Escalation
- Reducing Handoffs and Coordination Burden
- Redrawing Functional and Organizational Boundaries
- Choosing Structural Forms for GenAI-Enabled Work
- Creating End-to-End Outcome Ownership
- Balancing Centralization, Federation, and Local Autonomy
- Organizing Shared GenAI Capabilities and Platforms
- Redesigning Shared Services, Staff Functions, and Centers of Excellence
- Rebalancing Workforce Composition
- Redesigning Entry-Level Work and Career Architecture
- Preserving Expertise, Learning, and Succession
- Allocating Productivity Gains and Released Capacity
- Reconsidering Organizational Scale, Sourcing, and Firm Boundaries
- Redesigning Performance Measures and Incentives
- Designing Organizational Resilience Around Critical AI Dependencies
- Testing, Transitioning, and Reversing Structural Change
Establishing the Case and Requirements for Organizational Redesign
Recognizing When Individual GenAI Gains Have Become a Structural Issue
- Which GenAI-enabled productivity gains are recurring across enough people or teams to challenge the current organizational design?
- Have these gains changed who needs to participate in the work, or have they only made existing tasks faster?
- Which handoffs, approvals, specialist dependencies, or managerial activities have become less necessary because of these gains?
- Are individual gains translating into better organizational outcomes, or are they being absorbed by unchanged workflows and structures?
- Which assumptions about scarce expertise, information processing, coordination, or staffing no longer appear to hold?
- What new bottlenecks or dependencies have emerged as individual productive capacity has increased?
- What evidence would justify moving from local work improvement to a broader organizational-design review?
Distinguishing Organizational Redesign from Local Workflow Improvement
- Does the problem require changing organizational responsibilities, authority, reporting relationships, team composition, or capability placement?
- Could the desired improvement be achieved by changing the workflow while leaving the surrounding organization intact?
- Which organizational dependencies would remain unchanged even if the workflow were redesigned successfully?
- Are we trying to solve a structural problem with process changes, or a process problem with reorganization?
- What would actually become impossible or unnecessarily costly if the current organizational structure stayed in place?
- Which proposed changes concern how work is performed rather than how the organization itself is arranged?
- Where should we draw the boundary between this redesign and separate GenAI adoption, process, technology, or change-management work?
Testing Whether Existing Structures Still Match How Work Gets Done
- Which parts of the formal organization no longer reflect how GenAI-enabled work actually flows?
- Where are people routinely working across role, team, or functional boundaries that the formal structure still treats as separate?
- Which reporting relationships or organizational units exist for coordination needs that have materially changed?
- Where have responsibilities shifted informally without corresponding changes to authority or accountability?
- Which teams or functions are maintaining structures around work that has already disappeared, shrunk, or moved elsewhere?
- Where is the formal structure creating avoidable handoffs, delays, duplication, or escalation in the new work system?
- Which mismatches between formal design and actual work are large enough to justify structural correction?
Separating Temporary GenAI Effects from Durable Design Requirements
- Which observed changes are likely to persist even as specific models, tools, or vendors change?
- Are current productivity gains stable across ordinary users and operating conditions, or concentrated in pilots and early adopters?
- Which work changes depend on capabilities that remain unreliable, immature, or rapidly changing?
- Have new responsibilities and dependencies stabilized enough to support permanent organizational redesign?
- Which proposed structural changes could become obsolete if GenAI capability develops differently than expected?
- Where would a temporary arrangement preserve flexibility better than a permanent organizational change?
- What evidence over what period would make us confident that a GenAI-driven change is durable enough to design around?
Translating Strategic Priorities into Organizational Design Requirements
- Which strategic outcomes should the organization become structurally better able to deliver because of GenAI?
- Does the strategy require greater speed, innovation, scale, personalization, efficiency, resilience, or some different organizational capability?
- Which parts of the organization must become more autonomous, integrated, specialized, or coordinated to support those priorities?
- What kinds of human judgment or expertise become more strategically important as GenAI handles more routine cognitive work?
- Where would structural efficiency conflict with strategic needs for resilience, customer intimacy, learning, or differentiation?
- Which GenAI-enabled organizational possibilities are interesting but irrelevant to the strategy we are actually pursuing?
- What design requirements follow directly from our strategic priorities before we consider any particular organizational form?
Identifying Constraints the New Organization Must Continue to Respect
- Which legal, regulatory, professional, contractual, or policy constraints limit how responsibilities can be redistributed?
- Which activities require formal independence, segregation of duties, qualified human judgment, or specific accountability?
- What technical, data, security, or access limitations constrain where GenAI-enabled work can realistically sit?
- Which customer, market, geographic, or operational requirements make local presence or local authority important?
- What critical capabilities cannot be disrupted while the organization is being redesigned?
- Which financial, workforce, labor, or contractual constraints limit how quickly roles or units can change?
- Which constraints are genuinely fixed, and which are inherited assumptions that should themselves be challenged?
Deciding Which Organizational Level Actually Needs Redesign
- Is the design problem primarily within a role, a team, a function, a business unit, or the enterprise as a whole?
- At what level does the GenAI-driven change first alter a persistent organizational dependency?
- Which problems can be solved locally without forcing unnecessary changes elsewhere?
- Which local changes would fail unless authority, platforms, incentives, or structures also changed at a higher level?
- Are similar problems appearing across several units strongly enough to justify an enterprise-level response?
- Could redesign at one level create new misalignment with adjacent levels of the organization?
- What is the smallest organizational level at which the underlying problem can be solved completely?
Establishing What Must Be Preserved Through Structural Change
- Which capabilities, relationships, expertise, and responsibilities remain valuable regardless of how the structure changes?
- Which existing organizational boundaries protect independence, accountability, or professional judgment that should not be lost?
- What institutional knowledge could disappear if roles or units are removed too quickly?
- Which customer or stakeholder relationships depend on continuity with particular teams or roles?
- What learning, apprenticeship, or succession mechanisms are embedded in the current structure even if they are not formally recognized?
- Which redundancies provide resilience rather than unnecessary duplication?
- What should be explicitly protected before we begin simplifying or reallocating the organization?
Setting an Evidence Threshold Before Redesigning the Organization
- What evidence do we currently have that GenAI has changed the organizational mechanism rather than merely task performance?
- Are observed gains based on actual end-to-end outcomes or on activity measures such as time saved and output produced?
- Have we measured hidden review, integration, exception, and coordination work created elsewhere?
- Has the changed way of working been demonstrated under ordinary operating conditions rather than only in controlled pilots?
- How much variation exists across roles, teams, cases, and levels of employee experience?
- Which structural changes would require stronger evidence because they are expensive, difficult to reverse, or damaging if wrong?
- What additional experiment or measurement would most reduce uncertainty before a permanent redesign decision?
Defining Design Criteria Before Choosing a New Structure
- What must the redesigned organization improve relative to the current one?
- Which outcomes should be optimized, and which should be treated as constraints rather than objectives?
- How should we trade off efficiency, speed, quality, autonomy, control, learning, resilience, and employee development?
- What level of accountability clarity should every proposed design have to satisfy?
- Which capabilities must remain close to the work, and which should benefit from enterprise scale?
- How much flexibility should the structure retain as GenAI capabilities continue to change?
- What criteria will we use to compare structural alternatives before becoming attached to a preferred organizational model?
Reconstructing the Architecture of Work
Mapping How Cognitive Work Flows from Need to Outcome
- What are the major stages through which a need becomes a completed organizational outcome?
- Which people, teams, systems, and AI capabilities participate at each stage?
- Where does responsibility transfer from one actor to another along the flow?
- Which stages involve production, judgment, coordination, approval, execution, or accountability?
- Where does work wait because another person, function, or decision is required?
- Which stages have already changed materially because of GenAI even though the formal workflow has not?
- What does the full end-to-end work architecture reveal that individual task analysis would miss?
- Which roles or organizational layers mainly collect, summarize, translate, route, or repackage information for others?
- Which activities exist because decision-makers cannot personally absorb the volume of information available?
- Where are specialist intermediaries needed mainly to make information understandable to another function?
- Which reporting processes are built around manually consolidating information from several sources?
- Where does work move upward primarily because higher levels historically had greater analytical capacity?
- Which information-processing activities can GenAI now perform reliably enough to challenge their organizational rationale?
- What other valuable functions do these roles or layers perform that would remain even if the information-processing work disappeared?
Separating Production, Evaluation, Integration, Judgment, and Accountability
- Which parts of the work create the initial output, and which parts determine whether that output is good enough?
- Who integrates different outputs into a coherent final result?
- Where does consequential judgment enter the work after routine production is complete?
- Who has authority to accept, reject, or modify the result?
- Who remains accountable for the outcome even when another person or AI system performed most of the production?
- Which of these responsibilities are currently bundled in one role mainly because production was historically human?
- Would separating any of these responsibilities improve efficiency, independence, or accountability in a GenAI-enabled workflow?
Locating Tasks That GenAI Has Materially Changed
- Which tasks have experienced durable changes in time, cost, quality, or required expertise because of GenAI?
- Which tasks can now be delegated almost completely, and which remain primarily human with AI assistance?
- Where does GenAI improve work only for routine cases while difficult cases remain largely unchanged?
- Which tasks appear automatable but create substantial verification or correction work afterward?
- Where has GenAI changed the level of employee experience required to perform the task competently?
- Which changed tasks are central enough to the surrounding role or workflow to create organizational consequences?
- Which apparent changes remain too unstable or context-dependent to design around?
Identifying Valuable Work That Becomes Newly Feasible
- What useful work did we previously leave undone because human cognitive capacity was too expensive or scarce?
- Which analyses, experiments, follow-ups, documentation, personalization, or quality improvements are now economically feasible?
- What new services or capabilities become possible once the marginal cost of cognitive production falls?
- Could newly feasible work absorb much of the capacity that GenAI appears to release?
- Which new work creates genuine value rather than simply more artifacts or activity?
- What new review, integration, or decision demands would this additional work create?
- Should the organization redesign around greater scope rather than assuming that GenAI will primarily reduce labor requirements?
Tracing Human Dependencies Across the Existing Workflow
- Which steps cannot proceed until another person or team provides information, approval, expertise, or judgment?
- Which dependencies exist because knowledge is genuinely specialized, and which exist because access to knowledge was historically limited?
- Where does one function depend repeatedly on another for routine cases?
- Which dependencies create waiting time without adding meaningful control or judgment?
- Which dependencies remain necessary because they protect accountability, independence, or coordination?
- Has GenAI reduced the need for any dependency without the organization formally changing the workflow?
- Which dependencies would still constrain throughput even if every individual task became faster?
Identifying Work That Still Requires Direct Human Contribution
- Which parts of the work depend on tacit context, relationships, negotiation, empathy, or physical presence?
- Where is human judgment required because the consequences are significant or difficult to reverse?
- Which responsibilities must remain human because someone needs legitimate organizational or professional authority?
- What work requires recognizing unusual circumstances that are difficult to specify in advance?
- Where does accountability require meaningful human understanding rather than nominal approval?
- Which human contributions remain important even if AI can generate technically plausible outputs?
- Are we preserving direct human involvement because it remains valuable, or simply because the old workflow assumed it?
Separating Activities That No Longer Need to Stay Bundled
- Which activities are currently combined in one role only because the same person historically had to perform them?
- Can routine production be separated from judgment, review, relationship management, or accountability?
- Would separating certain activities allow AI to handle them at scale without weakening ownership of the final outcome?
- Which tasks could move to a shared capability while the role retains its distinctive responsibilities?
- Would unbundling create cleaner responsibilities or merely create additional handoffs?
- What new coordination burden would appear if the activities were separated?
- Which activities should remain bundled because their value depends on one person understanding the whole context?
Recombining Activities That No Longer Need Separate Specialists
- Which activities are separated today because no single role previously had affordable access to all the necessary knowledge?
- Where can GenAI give one role enough adjacent-domain capability to own more of the work end to end?
- Which specialist contributions consist mainly of routine knowledge access or production rather than deep judgment?
- Could recombining activities remove repeated handoffs without making one role unmanageably broad?
- What expertise would the broader role still need in order to recognize when specialist escalation is necessary?
- Which professional, risk, or independence requirements prevent certain activities from being recombined?
- Where would recombination improve ownership and speed without weakening the quality of difficult cases?
Comparing the Existing Work Architecture with a GenAI-Enabled Alternative
- Which stages of the current work architecture would disappear, shrink, expand, or change purpose in a GenAI-enabled version?
- Which responsibilities would move from humans to AI, and which would become more important for humans?
- How would the number and location of handoffs differ between the two architectures?
- Where would expertise, judgment, approval, and accountability sit in the alternative design?
- What new dependencies on data, platforms, agents, reviewers, or specialists would the alternative create?
- Which bottleneck would likely replace the bottleneck removed by GenAI?
- Does the alternative genuinely simplify the work system, or mainly make the existing architecture operate faster?
Redesigning Around New Bottlenecks and Organizational Constraints
Responding When Production Capacity Is No Longer the Main Constraint
- What resource now limits outcomes once producing drafts, analyses, code, or other artifacts becomes cheap?
- Is the organization generating more work than it can evaluate, integrate, decide on, or implement?
- Which roles or functions are still staffed as though production were the primary scarcity?
- Should released production capacity move toward review, judgment, customer work, integration, or another bottleneck?
- Are existing targets encouraging unnecessary output simply because production capacity is available?
- Which organizational units should shrink, grow, or change purpose if production is no longer the main constraint?
- How should the organization measure performance once producing more artifacts no longer indicates greater value?
Responding When Verification Becomes the New Bottleneck
- Which AI-enabled outputs now require more review than the organization has competent reviewers available?
- What types of verification can be automated, sampled, standardized, or handled by exception?
- Which outputs require deep domain expertise rather than routine checking?
- Are reviewers spending time correcting work that should never have been delegated to AI in the first place?
- Should verification capacity be embedded in teams, pooled centrally, or concentrated among specialists?
- How can the organization prevent verification work from becoming invisible and therefore systematically understaffed?
- What change to the work architecture would reduce the verification burden instead of merely adding more reviewers?
Responding When Decision Capacity Cannot Keep Up with Analysis
- Is GenAI generating recommendations and alternatives faster than authorized decision-makers can evaluate them?
- Which decisions could safely move to lower levels or become rule-bound instead of remaining centralized?
- Are decision-makers receiving more information than they can use meaningfully?
- Which analyses should stop being produced because they do not change decisions?
- Could clearer decision rights prevent every recommendation from escalating to the same senior roles?
- Where could AI support prioritization without becoming the de facto decision-maker?
- What organizational change would increase decision throughput without sacrificing accountability or judgment?
Responding When Expert Attention Becomes Scarcer Than Output
- Which experts are becoming the mandatory review point for a rapidly growing volume of AI-enabled work?
- What routine cases can be resolved without expert involvement if standards and escalation rules are improved?
- Can expert knowledge be converted into reusable guidance, evaluations, or systems without eliminating expert judgment?
- Which expert activities create the greatest value and should therefore receive protected attention?
- Should experts be embedded in teams, pooled, or accessed only for exceptions?
- Is demand for expert attention increasing because GenAI is creating more work, not because the underlying problem became more complex?
- How should the organization prevent scarce experts from becoming permanent throughput bottlenecks?
Responding When Integration Becomes Harder as Local Output Expands
- What kinds of locally generated analyses, tools, agents, or decisions now need to fit together at the organizational level?
- Are teams optimizing their own work while creating incompatible outputs or duplicated solutions elsewhere?
- Which integration responsibilities currently have no clear owner?
- Where should common standards replace repeated negotiation between teams?
- Which interfaces between functions or platforms need redesign as local productive capacity grows?
- Would more central coordination solve the problem, or would it create another bottleneck?
- What should remain locally flexible while still integrating into a coherent organizational system?
Responding When Exception Handling Dominates Remaining Human Work
- What proportion of human work now consists of unusual, ambiguous, difficult, or failed cases?
- Do current staffing levels reflect the complexity of exception work rather than the declining volume of routine work?
- Which roles need deeper judgment because AI has removed most of the straightforward cases?
- Are exception cases being routed to people with the right expertise and authority?
- What patterns in recurring exceptions indicate that the AI-enabled workflow itself needs redesign?
- How should workload and performance expectations change when the remaining human cases are systematically harder?
- Should exception handling remain distributed across teams or become a dedicated specialist capability?
Responding When Prioritization Becomes Harder as More Work Becomes Feasible
- Which projects or activities are being pursued only because GenAI has made them cheaper to produce?
- Has the organization created enough additional capacity to pursue all newly feasible opportunities meaningfully?
- What criteria should determine which AI-enabled opportunities deserve scarce human attention?
- Who has authority to stop low-value work when production itself is inexpensive?
- Are teams generating more initiatives than leadership, customers, or operating systems can absorb?
- Could stronger portfolio governance create more value than additional AI production capacity?
- How should the organization distinguish valuable expanded scope from uncontrolled work proliferation?
Responding When Data, Context, or Access Limits GenAI-Enabled Work
- Which potentially valuable GenAI-enabled workflows are constrained primarily by unavailable data or organizational context?
- Is the problem technical access, unclear ownership, poor quality, fragmented systems, or legitimate restrictions?
- Which organizational units control the information needed by other teams or agents?
- Are access decisions being made at a level that understands both enterprise risk and local value?
- Would changing data or knowledge ownership remove repeated dependencies between teams?
- Which context must remain local rather than being centralized into a shared AI system?
- Does the bottleneck require organizational redesign, or should it remain a technology or governance issue?
Responding When Managerial Attention Becomes the Limiting Resource
- Which decisions, escalations, reviews, or coordination tasks are consuming increasing amounts of managerial attention?
- Has GenAI increased team output faster than management capacity to prioritize and integrate it?
- Which managerial activities could be delegated, automated, or moved closer to the work?
- Are managers becoming approval bottlenecks because decision rights have not changed with employee capability?
- Which aspects of management still require scarce human attention and should be protected?
- Would wider spans improve the situation, or would they intensify the attention constraint?
- What organizational redesign would reduce managerial dependency rather than merely asking managers to process more work?
Determining Whether a Constraint Has Disappeared or Merely Moved Elsewhere
- What constraint did GenAI appear to remove from the original work system?
- What new review, coordination, decision, integration, infrastructure, or expertise demand appeared afterward?
- Has total end-to-end effort fallen, or has work simply shifted between roles or functions?
- Which part of the organization now bears costs that were previously visible elsewhere?
- Does the new bottleneck have greater consequences than the constraint that was removed?
- Would structural redesign reduce the new constraint, or merely move it again?
- What end-to-end evidence would demonstrate that the organization has genuinely increased usable capacity?
Redesigning Role and Job Architecture
Redesigning a Role After Routine Production Work Disappears
- Which activities that historically justified this role are now reliably performed by GenAI?
- What responsibilities remain once routine production is removed?
- Do the remaining responsibilities still form a coherent and substantial job?
- Should the role shift toward judgment, customer interaction, orchestration, review, exception handling, or another form of value?
- What new skills and authority would the redesigned role require?
- Could remaining responsibilities be absorbed more coherently by another role instead?
- How should success be measured once output production is no longer the role's main contribution?
Broadening a Role into Adjacent Work Made Accessible by GenAI
- Which adjacent activities can this role now perform competently with GenAI support?
- What handoffs would disappear if the role owned a broader end-to-end outcome?
- Does the person have enough domain understanding to evaluate AI-supported work outside the role's traditional specialty?
- Where would specialist escalation still be required?
- Would broadening the role increase meaningful ownership or simply create overload?
- What decision rights need to expand alongside the broader responsibilities?
- What evidence would show that the broader role performs better than the current specialist handoff model?
Combining Roles Whose Historical Separation No Longer Adds Value
- Why were these roles originally separated?
- Which of those reasons still exist after GenAI changes the work?
- How much duplication, waiting, translation, or handoff effort would disappear if the roles were combined?
- Would one combined role have enough expertise and capacity to carry the full responsibility bundle?
- Which conflicts of interest, independence requirements, or control functions require the roles to remain separate?
- What career or staffing consequences would combining the roles create?
- Would the combined role create clearer end-to-end accountability than the current arrangement?
Splitting Production, Review, and Accountability into Different Responsibilities
- Does one role currently produce work, check its own work, and remain solely accountable for the outcome?
- Which parts of production can be delegated to AI or another role without weakening ownership?
- Where would independent review materially improve reliability or challenge?
- Who should have authority to accept or reject the produced result?
- How can accountability remain clear when several actors contribute different parts of the work?
- Would splitting the responsibilities create excessive handoffs or useful separation of duties?
- What criteria should determine when production, review, and accountability may safely remain bundled?
Preserving a Role Whose Tasks Change but Core Responsibility Does Not
- Which core outcome or responsibility still requires this role to exist?
- Which tasks can change substantially without changing the role's fundamental organizational purpose?
- Is the role being threatened because visible activities disappeared even though accountability remains?
- What new methods should replace obsolete tasks while preserving ownership of the outcome?
- Does the role need fewer people, different skills, or simply different tools?
- How should job expectations change without overstating the structural significance of task automation?
- What evidence would show that preserving the role is still more coherent than redistributing its responsibility elsewhere?
- Which responsibilities are employees already performing that are missing from the formal role?
- Which listed responsibilities have become marginal, automated, or obsolete?
- Has GenAI caused informal responsibility to expand without corresponding authority or recognition?
- Which new review, judgment, orchestration, or AI-supervision duties should be made explicit?
- Are performance measures still tied to activities the role no longer meaningfully performs?
- What skills and development expectations should change with the actual work?
- What organizational problems will persist if the formal role continues to lag behind reality?
Correcting Roles That Gain Responsibility Without Matching Authority
- What new outcomes or decisions has the role become responsible for?
- Which approvals or decision rights remain elsewhere despite the expanded responsibility?
- Where is the role accountable for outcomes it cannot materially control?
- Which authority should move with the work, and which should remain separate for legitimate reasons?
- Are employees compensating through informal influence because formal authority is insufficient?
- What risks arise if responsibility continues to expand without corresponding mandate?
- How should the role be redesigned so authority, information, resources, and accountability become coherent?
Shifting a Role from Production Toward Orchestration and Judgment
- Which production activities can now be delegated reliably to GenAI or agents?
- What orchestration responsibilities become necessary when multiple AI-supported tasks contribute to one outcome?
- Which judgments should remain explicitly human?
- How much underlying technical or domain competence must the role retain to supervise the work credibly?
- What new skill does the role need to direct, evaluate, integrate, and challenge AI-produced outputs?
- How should workload be measured when the role produces less directly but controls more consequential work?
- Does the redesigned role create genuine value, or is "orchestration" merely a new label for unclear responsibilities?
Preserving Hidden Responsibilities When Visible Tasks Are Automated
- What does this role contribute beyond the tasks most easily observed and automated?
- Does it maintain relationships, resolve ambiguity, coach others, remember history, coordinate exceptions, or protect standards?
- Which hidden responsibilities would become ownerless if the role disappeared?
- Can those responsibilities be reassigned deliberately rather than preserved accidentally?
- Which hidden responsibilities remain valuable enough to justify retaining or redesigning the role?
- Are any apparently valuable hidden responsibilities actually consequences of unnecessary complexity elsewhere?
- What must be transferred before visible task automation can safely reduce or eliminate the role?
Retiring or Reconstructing a Role Whose Original Purpose Has Disappeared
- What original organizational problem justified creating this role?
- Has GenAI or another organizational change genuinely removed that problem?
- Which responsibilities would still need an owner if the role disappeared?
- Could the role be reconstructed around a new scarcity rather than simply retained under its old title?
- What knowledge or relationships need to be transferred before retirement?
- Would eliminating the role create fewer dependencies, or would its work simply become dispersed and hidden?
- What evidence would justify retirement rather than continued experimentation with a redesigned role?
Repositioning Specialist Expertise
Moving Routine Specialist Cases into AI-Enabled Generalist Work
- Which specialist cases are sufficiently routine, codified, and verifiable for AI-enabled generalists to handle safely?
- What expertise must generalists retain to recognize when a case is no longer routine?
- Which specialist handoffs would disappear if normal cases moved outward?
- What standards or tools would help generalists perform the work consistently?
- How should specialists remain involved in difficult cases without reviewing every routine case?
- What error patterns would indicate that routine work has been moved too far from specialists?
- How should specialist capacity be redeployed once routine case volume falls?
Shifting Specialists Toward Exceptions and Difficult Cases
- Which routine specialist activities can be handled reliably elsewhere?
- What types of cases genuinely require the specialist's deeper judgment?
- Will concentrating specialists on difficult cases make their workload more cognitively demanding even if case volume falls?
- How should cases be triaged so that specialists receive the right exceptions?
- What additional context or preparation should accompany an escalated case?
- How should performance expectations change when specialists handle fewer but harder cases?
- What mechanisms will allow specialist learning from difficult cases to improve routine AI-enabled work?
Preserving Specialist Independence for Challenge and Assurance
- Which specialist functions need enough organizational independence to challenge producers or decision-makers credibly?
- Would embedding these specialists more deeply in operating teams weaken their ability to provide objective challenge?
- Which decisions require an independent specialist view even when AI makes specialist knowledge widely accessible?
- Who should control the specialist's priorities, performance evaluation, and escalation rights?
- What access to information does the specialist need without becoming operationally responsible for the work being reviewed?
- Where would independence create unnecessary delay rather than meaningful assurance?
- How should the organization distinguish specialist support from independent specialist challenge?
Expanding Specialist Capacity When GenAI Creates More Work to Review
- Has GenAI increased the volume of outputs requiring specialist judgment faster than it has reduced specialist production work?
- Which new review demands are genuinely necessary and which result from weak workflow design?
- Could standardized criteria or automated checks remove some specialist review work?
- Which specialist capabilities are becoming bottlenecks as AI-supported production expands?
- Should the organization hire more specialists, redeploy existing ones, or redesign the escalation model?
- Is review demand temporary while users learn, or likely to remain structurally higher?
- How can additional specialist capacity be added without turning every AI-enabled workflow into a mandatory review queue?
Choosing Between Centralized and Embedded Specialist Expertise
- Which specialist work benefits most from close proximity to a specific business, team, or customer context?
- Which specialist capabilities gain more value from pooling, standardization, and shared learning?
- How frequently does each team genuinely need specialist involvement?
- Would embedded specialists become overloaded with local routine work that could be handled through shared capability?
- Would centralized specialists lose too much context to make useful judgments quickly?
- Could a hybrid model combine embedded relationships with centralized professional development and standards?
- What decision rights and escalation paths would make either model work coherently?
Distinguishing Access to Specialist Knowledge from Genuine Expertise
- Which specialist questions can GenAI answer well enough for nonspecialists to proceed independently?
- Where does the user still need deep expertise to recognize whether the AI's answer is wrong or incomplete?
- What kinds of tacit judgment, contextual interpretation, or professional responsibility are not reproduced by knowledge access alone?
- Are people treating fluent AI explanations as evidence that specialist competence is no longer needed?
- Which decisions require a qualified specialist even when the relevant information is widely accessible?
- How should the organization teach generalists to recognize the boundary of their AI-supported competence?
- What evidence would justify reducing specialist involvement without confusing information access with expertise?
Separating Specialist Ownership from Routine Specialist Participation
- Which outcomes still need a specialist owner even if specialists no longer participate in every case?
- Can specialists define standards and escalation rules while routine execution moves elsewhere?
- Which responsibilities must remain with specialists because they require professional or technical ownership?
- How should routine teams know when the specialist owner must become directly involved?
- What visibility does the specialist owner need into work they no longer perform personally?
- Could separating ownership from routine participation create accountability without unnecessary specialist bottlenecks?
- What would demonstrate that the specialist still has meaningful control over the responsibilities they formally own?
Creating Shared Specialist Pools for Selective Intervention
- Which specialist capabilities are needed across many teams but too intermittently to justify permanent embedding?
- What demand patterns would support a shared specialist pool?
- How should teams access specialists without creating a slow centralized queue?
- What triage criteria should determine which requests receive specialist attention?
- How can pooled specialists maintain enough context to provide useful advice quickly?
- Who should own staffing, prioritization, and professional development within the shared pool?
- When would a shared pool be inferior to embedded specialist capacity or direct self-service?
Redefining Specialist Functions Around Standards, Architecture, and Judgment
- Which routine production activities can the specialist function stop owning?
- What standards should the function define for work increasingly performed by generalists or AI?
- Which architectural decisions require concentrated specialist authority?
- What difficult judgments should remain inside the specialist function?
- How should the function monitor quality without reviewing every case?
- What role should specialists play in developing reusable AI-enabled capabilities for the rest of the organization?
- How should the function's success be measured once its value comes less from production volume and more from system quality?
Deciding Which Specialist Capabilities Need a Dedicated Organizational Home
- Which specialist capabilities require a stable community rather than being dispersed across operating teams?
- Does the capability depend on deep professional development, shared standards, or accumulated specialist knowledge?
- Would dispersing specialists weaken independence, learning, or career progression?
- Could the capability be maintained through a virtual community without a formal organizational unit?
- How much ongoing demand exists for the capability outside exceptional cases?
- What risks would arise if no single organizational home remained accountable for maintaining the expertise?
- What evidence would justify a dedicated specialist unit rather than embedded, pooled, or external expertise?
Designing Human-AI and Human-Agent Work Configurations
Using GenAI as an Assistant Within a Human-Owned Task
- Which parts of the task should GenAI support while the human remains responsible for the full outcome?
- What information, judgment, or context must the human supply before AI assistance becomes useful?
- Which outputs should the human inspect rather than accept directly?
- Where does AI assistance save meaningful effort without changing the underlying ownership of the task?
- What competence must the human retain to recognize weak or inappropriate AI output?
- Which parts of the task should remain unaided because direct human engagement creates important value?
- How should the role be designed so AI assistance strengthens rather than obscures human responsibility?
Delegating a Bounded Task to AI While Humans Integrate the Result
- Which task can be specified clearly enough to delegate without transferring ownership of the larger outcome?
- What inputs, constraints, and completion criteria must be defined before delegation?
- How will the human determine whether the returned result is fit for integration?
- What failures could remain hidden until the AI-produced component interacts with other work?
- Who is responsible for reconciling the AI output with surrounding human-produced work?
- When should the task be returned to human execution rather than repeatedly corrected through AI?
- Does delegating this component actually reduce total work, or mainly move effort into review and integration?
Delegating a Multi-Step Workflow to an Agent with Defined Checkpoints
- Which sequence of steps can an agent perform coherently without human intervention after every action?
- At which points does the workflow become consequential enough to require explicit human review?
- What information should the human receive at each checkpoint to judge whether work should continue?
- Which intermediate actions must remain reversible until the next checkpoint is passed?
- What conditions should cause the agent to stop and escalate rather than continue?
- Who owns failures that occur between checkpoints when no human was directly involved?
- How should checkpoint placement change as evidence about the agent's reliability improves?
Supervising Multiple Agents from a Single Human Role
- What distinct work can the agents perform in parallel without creating unmanageable supervision demands?
- How many simultaneous agent workflows can one person meaningfully understand and oversee?
- Which agent activities require active supervision and which can be monitored by exception?
- How should conflicting outputs or actions from different agents be reconciled?
- What visibility does the human need into agent progress, decisions, and failures?
- When does adding another agent increase throughput, and when does it simply increase coordination complexity?
- How should the role change if supervising digital work becomes more important than performing the work directly?
Dividing Routine Cases and Exceptions Between AI and Humans
- Which cases are sufficiently predictable, bounded, and verifiable for AI to handle routinely?
- What characteristics should cause a case to be classified as an exception?
- Can the system recognize difficult cases reliably enough to route them before damage occurs?
- Which exceptions require a general human operator and which require a specialist?
- How should repeated exceptions feed back into the design of the routine AI workflow?
- Does concentrating humans on exceptions create a workload that is systematically more difficult than current staffing assumes?
- What proportion of cases must remain human-handled before the division of work stops creating meaningful value?
Running Human and AI Work in Parallel Before Integrating the Result
- When is it useful for humans and AI to develop independent analyses or solutions before seeing each other's work?
- What kinds of problems benefit from comparing independent approaches rather than sequential assistance?
- How should conflicting human and AI conclusions be evaluated?
- Who should integrate the two outputs into a final result?
- Does parallel work create useful challenge or unnecessary duplication?
- Which biases or blind spots can this configuration expose that a single workflow might miss?
- When should parallel work be temporary for validation rather than a permanent operating model?
Allowing Agents to Execute Actions Within Predefined Boundaries
- Which actions can an agent take without obtaining case-by-case approval?
- What limits should apply to financial value, data access, customer impact, system access, or other consequences?
- Which actions should remain prohibited regardless of the agent's apparent confidence?
- How should the system respond when an intended action falls close to or outside an authorization boundary?
- Who has authority to define, change, and revoke those execution permissions?
- What monitoring is necessary to detect actions that are individually allowed but collectively problematic?
- How should accountability remain clear when the agent executes correctly within a boundary but the resulting outcome is still harmful?
Extending the Amount of Work Delegated Before Human Intervention
- What evidence supports allowing the AI or agent to operate for longer before human review?
- Which additional steps can be delegated without materially increasing consequence or uncertainty?
- Does longer delegation reduce coordination overhead enough to justify the additional autonomy?
- What signals should trigger earlier intervention even within an otherwise autonomous sequence?
- How does verification become harder as more intermediate reasoning and action occur without human involvement?
- What human competence must remain available when intervention becomes less frequent?
- At what point would extending delegation undermine meaningful human control?
Coordinating Several Agents Within One Human-Owned Workflow
- What distinct roles should different agents perform within the same end-to-end workflow?
- Which dependencies between agents require explicit sequencing or shared context?
- How should one agent's errors be prevented from propagating automatically into the work of others?
- Who decides which agent output prevails when agents disagree?
- What shared standards or state must the agents use to remain coordinated?
- How can the human owner understand the combined workflow without inspecting every interaction between agents?
- When does a multi-agent design create genuine capability beyond what one simpler system could achieve?
Determining Where Direct Human Participation Must Remain Continuous
- Which activities require ongoing human presence rather than periodic review or exception handling?
- Does continuous participation protect judgment, relationships, safety, accountability, or some other essential function?
- Which parts of the work lose value if the human enters only at the end?
- Are there situations where the human must observe the process itself in order to understand the outcome?
- What competence would deteriorate if humans stopped participating continuously?
- Could continuous participation be narrowed to particular cases rather than required universally?
- What evidence would justify reducing direct human involvement without turning oversight into a formality?
Redesigning Team Composition and Support
Reducing Team Size After GenAI Expands Individual Capability
- Which capabilities previously requiring several team members can now be performed by fewer people with GenAI?
- What responsibilities would disappear, shrink, or remain if the team became smaller?
- Which forms of collaboration would be lost along with the removed positions?
- Can the smaller team still handle peaks, absences, difficult cases, and unexpected work?
- Does reduced headcount actually lower coordination cost without creating excessive dependence on individual employees?
- What happens to development, mentoring, and succession when fewer people participate in the work?
- What evidence should be sustained before the smaller team size becomes permanent?
Broadening Team Roles Without Losing Necessary Depth
- Which team members could take on adjacent responsibilities with GenAI support?
- What specialist depth must still exist somewhere in the team or its support network?
- How broad can a role become before cognitive load and accountability become unclear?
- Which responsibilities should remain concentrated in one person because they require sustained expertise?
- How should team members recognize when to stop relying on broad AI-supported capability and seek deeper expertise?
- Could broader roles reduce handoffs while preserving sufficient quality?
- What learning and career model would support broader roles without producing shallow competence?
Moving Specialists from Permanent Team Membership to On-Demand Support
- Which specialist contributions are needed too infrequently to justify permanent team membership?
- Can the team handle routine specialist questions independently with GenAI?
- What cases should trigger on-demand specialist involvement?
- How quickly must specialists be available for the model to work operationally?
- What context must accompany a request so specialists do not repeatedly reconstruct the case from scratch?
- Will specialists still understand the team's work well enough if they participate only selectively?
- At what demand level would permanent specialist membership again become more effective?
Reducing Routine Support Capacity Around High-Value Professionals
- Which support activities around high-value professionals can GenAI now perform reliably?
- What support responsibilities remain because they involve relationships, coordination, discretion, or contextual judgment?
- How much of the professional's own time would be consumed if support roles were reduced too aggressively?
- Does AI actually eliminate the support work, or shift configuration and review work back to the professional?
- Which support activities should become shared rather than individually assigned?
- What hidden organizational knowledge might disappear when experienced support staff are removed?
- What staffing configuration maximizes the professional's valuable time rather than merely minimizing support headcount?
Repurposing Support Roles Toward Coordination, Review, and Exceptions
- Which support tasks are declining enough to free meaningful capacity?
- What coordination, review, quality, or exception work is growing at the same time?
- Do existing support employees have enough context and capability to move into these responsibilities?
- Which new authority or training would the redesigned support role require?
- Could the role become more valuable by preventing downstream problems rather than producing routine materials?
- How should its performance measures change when visible output declines?
- Is repurposing genuinely coherent, or are unrelated leftover tasks simply being bundled into one job?
Choosing Between an AI-Enabled Individual and a Multi-Person Team
- Can one AI-enabled person deliver the required outcome at acceptable quality for routine cases?
- What additional value does a human team provide beyond production capacity?
- Does the work benefit materially from independent perspectives, challenge, creativity, or shared judgment?
- How consequential would a single person's blind spot or absence be?
- Can specialists be added selectively rather than maintained as permanent team members?
- Does one-person execution create unacceptable key-person dependency?
- Should the staffing model vary according to case complexity rather than use one configuration for all work?
Adding Independent Review to a Highly AI-Enabled Team
- What risks arise when the same small team directs AI, evaluates the output, and owns the result?
- Which outcomes are consequential enough to justify independent review?
- What expertise should the reviewer possess that differs from the producing team?
- At what stage should independent review occur to influence the result without recreating unnecessary handoffs?
- Should every case be reviewed or only defined categories and exceptions?
- How can independent review remain genuinely independent if the reviewer depends operationally on the team being reviewed?
- What evidence would justify reducing review once the team's AI-enabled process becomes more mature?
Adding Technical AI Capability to an Existing Domain Team
- What AI-related work requires enough ongoing technical depth to justify embedding capability in the team?
- Which responsibilities should remain with enterprise technology or AI platform groups?
- Does the embedded specialist need to build systems, configure workflows, evaluate performance, or primarily translate between domain and technical teams?
- How can the technical role remain connected to enterprise architecture and standards?
- Would embedding the capability reduce delays enough to justify potential duplication across teams?
- What career and professional community should support embedded technical specialists?
- When should the team rely on shared technical services rather than maintain its own specialist?
Varying Team Composition According to Case Complexity
- Which characteristics reliably distinguish routine, difficult, and exceptional cases?
- What is the smallest viable team for each level of complexity?
- Which specialist or leadership roles should join only as complexity increases?
- Can team composition change dynamically without creating repeated coordination delays?
- Who decides when a case requires a larger or differently skilled team?
- What information should transfer as additional people join the work?
- How can the organization avoid staffing every case for the most difficult scenario?
Preserving Human Collaboration Where AI Does Not Replace Its Value
- Which outcomes improve because several humans think, challenge, negotiate, or create together?
- Are we removing teamwork because AI can reproduce some output rather than because collaboration no longer creates value?
- Which forms of tacit knowledge or social understanding emerge only through direct human interaction?
- Where does peer challenge catch errors that AI-assisted individual work may miss?
- How important is collaboration for development, trust, innovation, and collective ownership?
- Could AI remove low-value coordination while preserving the parts of collaboration that matter?
- What evidence would show that a more individual AI-enabled model performs worse than genuine human teaming?
Redesigning Managerial Work
Automating Routine Reporting and Administrative Management Work
- Which managerial activities consist mainly of collecting, formatting, summarizing, or routing information?
- Which of these activities can be automated without reducing the manager's understanding of the underlying work?
- What information should still require direct managerial engagement rather than passive receipt of AI summaries?
- Where would automation merely shift administrative effort into correcting or monitoring the system?
- Which approvals are administrative habits rather than meaningful exercises of judgment?
- How should managers use the time released from routine administration?
- What responsibilities should be removed from the role entirely rather than simply automated within it?
Redirecting Managerial Capacity Toward Coaching and Development
- How much managerial time can realistically be released from administrative or reporting work?
- Which employees need more coaching because GenAI is changing the skills their roles require?
- What developmental conversations are currently crowded out by operational management?
- How should managers help employees distinguish AI-supported performance from genuine capability growth?
- Which assignments or experiences should managers deliberately create as routine learning work disappears?
- How should coaching quality be measured without reducing it to activity counts?
- Does the manager have the capability and span needed to turn released time into meaningful development?
Managing Increased Output from AI-Enabled Teams
- What additional outputs are teams generating because GenAI lowers the cost of production?
- Which of those outputs actually require managerial review, coordination, or decisions?
- Has the manager become the bottleneck between increased team capacity and organizational action?
- Which work should be stopped or deprioritized rather than simply processed faster?
- What decisions can move downward so increased output does not automatically escalate upward?
- How should quality standards change when producing more material becomes easy?
- What new operating rhythm would help managers integrate greater output without increasing unnecessary control?
- What work is performed by humans, what is performed by agents, and where do responsibilities intersect?
- Which agent outcomes should managers monitor directly and which should be handled through system-level controls?
- How should managers distinguish performance problems caused by people from those caused by tools, data, or agents?
- What authority does the manager need to suspend, change, or constrain agent use?
- How should team workload account for the effort required to supervise digital work?
- What new managerial capabilities are needed to oversee human and agent performance together?
- How should accountability be structured when an agent performs work under a manager who did not inspect every action?
Concentrating Managerial Attention on Exceptions and Difficult Decisions
- Which routine decisions and approvals can be removed from the manager's workload?
- What types of exceptions genuinely benefit from managerial judgment?
- How should difficult cases reach the manager without routine cases being escalated unnecessarily?
- Does concentrating exceptions on managers make their remaining work significantly more demanding?
- What information should accompany an escalation so the manager can decide efficiently?
- Which recurring exceptions reveal flaws in the underlying workflow or decision rights?
- At what point should a recurring exception be redesigned out of the management system rather than continually escalated?
- How much of this manager's value currently comes from transmitting information upward or downward?
- Can teams and leaders now access that information directly through shared systems or GenAI?
- What contextual interpretation does the manager provide beyond the information itself?
- Which decisions, coaching, conflict resolution, or coordination responsibilities would remain if information relay disappeared?
- Does the remaining responsibility bundle justify a distinct management role?
- Could the role be repurposed toward integration or people development instead of removed?
- What organizational dependencies would break if the information-relay function disappeared tomorrow?
Responding When Managerial Judgment Becomes a Bottleneck
- Which decisions repeatedly wait for the same managers despite improved analytical capability elsewhere?
- Does the manager uniquely possess authority, expertise, context, or simply historical control?
- Which decisions could be delegated without weakening accountability?
- Could clearer principles allow teams to resolve recurring cases independently?
- Which decisions genuinely require scarce managerial judgment and should remain protected?
- Should more managerial capacity be added, or should dependence on managerial intervention be reduced?
- What evidence would show that a redesign has increased decision throughput without degrading outcomes?
Setting Quality and Delegation Boundaries for AI-Enabled Work
- Which work can employees or agents perform independently within defined quality expectations?
- What types of decisions or outputs require managerial review before use?
- How should managers distinguish acceptable variation from work that requires intervention?
- Which failures should automatically trigger escalation?
- Are delegation boundaries based on consequence and competence, or simply historical hierarchy?
- How should boundaries change as employees and AI systems demonstrate greater reliability?
- What evidence should managers use to expand or contract delegated authority?
Making Managers Responsible for Future Capability Development
- Which future skills and forms of judgment must this team continue to develop?
- How should managers allocate real work so employees gain those capabilities rather than simply maximize current AI-enabled output?
- Which employees risk losing developmental opportunities because routine tasks are automated?
- What responsibility should managers have for succession and specialist pipelines beyond immediate team performance?
- How should capability development influence staffing and work assignment decisions?
- What measures would reveal whether a manager is building or merely consuming team capability?
- How should short-term productivity expectations be balanced against the manager's responsibility to develop future capacity?
Redesigning Management When Automation Increases Rather Than Reduces Workload
- What new managerial work has appeared because AI increased output, speed, complexity, or exceptions?
- Which activities were expected to disappear but instead shifted into supervision, integration, or review?
- Is the manager now responsible for more workflows, agents, decisions, or coordination than before?
- Which new demands should be automated, delegated, or reassigned rather than absorbed indefinitely?
- Does the management span still make sense under the new complexity?
- Are productivity expectations based on gross automation gains rather than actual managerial workload?
- What structural change would prevent AI-enabled work from simply expanding until managerial capacity is exhausted?
Reconsidering Management Layers and Spans of Control
- What proportion of this layer's work consists of collecting, interpreting, and passing information between levels?
- Can the organization now provide that information directly without losing essential context?
- What decision, coaching, coordination, or accountability functions would remain if information aggregation disappeared?
- Does the layer still reduce complexity for senior leaders or merely reproduce information already available elsewhere?
- Would removing it improve speed while overloading adjacent levels?
- What responsibilities would need explicit reassignment before the layer could disappear?
- What evidence would distinguish an obsolete information layer from a still-valuable management layer?
Widening Spans After Routine Managerial Work Declines
- Which managerial activities have genuinely disappeared or become much less demanding?
- How much additional supervisory capacity has actually been released?
- Are direct reports experienced and autonomous enough for a wider span?
- What coaching, performance, or relationship demands would increase with more direct reports?
- Does the manager also supervise agents or complex workflows that are not reflected in human headcount?
- How would a wider span affect escalation speed, employee access, and managerial judgment?
- What pilot evidence should be required before adopting the wider span permanently?
Preserving Narrower Spans Where Coaching and Judgment Remain Intensive
- Which teams require unusually frequent coaching, judgment, or hands-on managerial involvement?
- Does GenAI increase rather than reduce the complexity of the cases those teams handle?
- Are employees early in their careers or working in roles where development requires close supervision?
- How consequential are the decisions made within the team?
- Would a wider span materially reduce access to meaningful managerial support?
- Can administrative automation free enough time to preserve quality without widening the span?
- What evidence would justify retaining a narrow span despite pressure for structural efficiency?
Removing a Layer and Redistributing Its Remaining Responsibilities
- Which responsibilities performed by the layer disappear when the layer is removed?
- Which responsibilities must move upward, downward, sideways, or into shared systems?
- Who will own coaching, resource allocation, conflict resolution, and escalation afterward?
- Will decision rights move with the redistributed responsibilities?
- Does the layer currently provide institutional context that adjacent levels lack?
- What new coordination load will fall on remaining managers and teams?
- How will the organization know whether responsibilities were truly redistributed rather than simply abandoned?
Combining Management Layers Whose Distinct Purposes Have Weakened
- What different purposes originally justified the two layers?
- Which of those purposes remain distinct after GenAI changes information flow and managerial work?
- Could one management role perform the remaining responsibilities without becoming overloaded?
- What decision rights would need to be consolidated or clarified?
- How would combining the layers affect access to leaders and speed of escalation?
- What development or promotion opportunities would disappear if one layer were removed?
- Would the combined structure simplify accountability or create an overly broad management role?
Preventing Flatter Structures from Producing More Senior Micromanagement
- Does increased visibility allow senior leaders to intervene in decisions previously owned locally?
- Which decisions should remain delegated even when senior leaders can now see more detail?
- Are lower-level managers still genuinely accountable if senior leaders frequently override them?
- What information should senior leaders receive without converting visibility into routine approval authority?
- Which escalation rules would protect local autonomy while preserving enterprise oversight?
- Are leaders using AI-generated monitoring to improve support or to increase direct control?
- How can the organization detect when delayering has produced central micromanagement rather than genuine empowerment?
Repurposing a Layer Instead of Eliminating It
- Which original responsibilities of the layer have genuinely declined?
- What emerging organizational problem could the layer solve better than another part of the organization?
- Could the layer shift toward coaching, integration, portfolio management, or cross-team coordination?
- Would repurposing preserve a role that is no longer needed simply to avoid structural change?
- What new authority and capabilities would the repurposed layer require?
- How would success be measured differently from the old management model?
- What evidence would justify repurposing over full removal or redistribution of responsibilities?
Accounting for Agent Supervision When Setting Management Spans
- How much managerial attention do the team's agents require in addition to human supervision?
- Does agent oversight involve routine monitoring, exception handling, authorization, or consequential judgment?
- How many agent workflows can a manager understand well enough to remain accountable?
- Should digital systems count toward managerial span based on number, autonomy, risk, or complexity?
- Are managers supervising agents directly or supervising people who themselves control agents?
- Could agentic work make a numerically wider human span functionally more complex than before?
- What measure of supervisory load would be more useful than simply counting direct reports?
Correcting Delayering That Creates New Coordination Gaps
- Which coordination responsibilities disappeared when the layer was removed?
- Where are teams now resolving cross-unit dependencies through informal workarounds?
- Have senior leaders or frontline teams absorbed coordination work they are poorly positioned to perform?
- Which decisions now lack a clear forum or owner?
- Would restoring a management layer solve the issue, or could another coordination mechanism work better?
- What evidence shows that the problem comes from delayering rather than unrelated workflow design?
- How can the organization repair the coordination gap without automatically rebuilding the old hierarchy?
Piloting Wider Spans Before Making Permanent Structural Changes
- Which teams provide a sufficiently bounded environment for testing a wider managerial span?
- What baseline measures should be collected before changing the span?
- How long must the pilot run to reveal effects on coaching, decisions, workload, quality, and employee support?
- What responsibilities should remain unchanged so the span itself can be tested meaningfully?
- Which outcomes would indicate that the manager is becoming overloaded?
- What reversal criteria should be agreed before the pilot begins?
- What conditions must be present before results from one team can be generalized to others?
Redistributing Decision Rights and Authority
Moving Decisions Downward When Frontline Capability Increases
- Which decisions are currently escalated mainly because frontline employees historically lacked sufficient information or expertise?
- What new GenAI-supported capabilities make local decision-making more feasible?
- How consequential and reversible are the decisions being considered for delegation?
- What boundaries should define the frontline team's authority?
- What evidence would show that local decisions are at least as good as centrally made ones?
- How should accountability move when authority is delegated downward?
- Which decisions should remain centralized despite greater frontline capability because their effects extend beyond the local context?
Retaining Central Authority Where Local Decisions Create Enterprise-Wide Consequences
- Which local decisions can materially affect shared resources, reputation, risk, customers, or other business units?
- Can frontline teams see the full consequences of these decisions?
- Would common rules be sufficient, or is direct central approval genuinely required?
- How quickly must central decisions be made to avoid becoming an operational bottleneck?
- Which information from local teams is essential for central decision quality?
- Could authority be centralized only for defined thresholds rather than for every case?
- How should the organization prevent enterprise coordination needs from becoming a justification for unnecessary central control?
Reconsidering Decisions That Remain Central Only Because Expertise Was Scarce
- Which decisions were centralized because specialized knowledge was historically concentrated at higher levels?
- Can GenAI now provide lower levels with enough knowledge and analytical support to decide competently?
- What tacit expertise or organizational perspective remains unavailable through AI?
- How easy is it to verify whether a decentralized decision was sound?
- Should central experts move from deciding each case to defining standards and handling exceptions?
- What developmental capability would local teams need before authority could shift?
- What evidence would demonstrate that expertise scarcity no longer justifies the existing decision location?
Correcting Decentralized Authority Where Local Capability Is Still Insufficient
- Which decisions have moved outward faster than the receiving teams' capability to make them well?
- Are poor outcomes caused by weak judgment, missing information, unclear standards, or inadequate AI support?
- Which decisions should temporarily move back toward a more capable level?
- Could targeted training or specialist support fix the capability gap without reversing decentralization entirely?
- Are teams recognizing and escalating cases beyond their competence?
- How should authority expand as local capability matures?
- What evidence would show that decentralization can safely resume?
- Why was this class of decision originally escalated?
- Does the lower level now have direct access to the information or analysis previously available only above?
- What judgment or authority does the higher level still add?
- How much waiting time and coordination does the escalation create?
- Could clear rules allow most cases to remain local while preserving escalation for genuine exceptions?
- Would removing the escalation make accountability clearer or more ambiguous?
- What cases should be monitored after the change to confirm that the old constraint has genuinely disappeared?
Preventing Better Central Visibility from Producing Unnecessary Central Control
- What new operational information can senior leaders now see because of GenAI-enabled monitoring and synthesis?
- Which of those insights justify intervention, and which should remain informational only?
- Are local leaders changing behavior because they expect senior leaders to second-guess visible decisions?
- What decision rights should remain explicit even when central leaders have more detailed information?
- Could monitoring thresholds identify genuine exceptions without inviting routine intervention?
- How should performance accountability work if senior leaders repeatedly override local choices?
- What governance norm would separate enterprise visibility from enterprise control?
Moving Decisions Closer to Customers or Operations
- Which decisions depend heavily on immediate customer, operational, or contextual knowledge?
- How much delay is created by sending those decisions away from the point of action?
- Does GenAI provide enough analytical or policy support for local actors to decide confidently?
- What authority and resource access would need to move with the decision?
- Which customer or operational decisions have externalities that still require central limits?
- How should outcomes be monitored without recreating a central approval process?
- What evidence would show that moving decisions closer to the work improves responsiveness without weakening consistency?
Centralizing Decisions That Require Coordination Across Interdependent Units
- Which decisions create tradeoffs that no single unit can optimize independently?
- Are local decisions producing conflicts over shared resources, priorities, or customer outcomes?
- Can GenAI improve central understanding of the distributed information needed for coordination?
- What authority should the central decision-maker have over affected units?
- How can local context be preserved rather than flattened into centralized analysis?
- Which recurring decisions could be governed by shared rules instead of repeated central intervention?
- Does centralization reduce total coordination effort enough to justify the loss of local autonomy?
- Which roles can now analyze decisions more effectively without being authorized to make them?
- Does better analytical capability actually justify moving authority, or only improve the quality of recommendations?
- What institutional responsibilities require approval to remain elsewhere?
- Are employees confusing access to expert-like AI assistance with legitimate organizational mandate?
- How should recommendation quality be recognized without creating informal decision authority?
- What information should authorized decision-makers receive from newly capable analytical roles?
- Under what conditions should demonstrated capability eventually lead to a formal transfer of authority?
Rebalancing Decision Rights After Workflows and Roles Have Changed
- Which decisions still sit with roles that no longer perform the work closest to them?
- Have responsibilities moved without corresponding decision authority?
- Which new AI-enabled roles now require authority they were never designed to hold?
- Are approvals still aligned with the points at which meaningful judgment actually occurs?
- Which decisions should move upward, downward, or sideways under the redesigned workflow?
- How should escalation paths change once ordinary decision ownership is redistributed?
- What final decision-rights map would make responsibility, authority, information, and accountability coherent again?
Designing Human-AI Decision and Accountability Boundaries
Separating AI-Generated Options from the Authority to Choose Among Them
- Which parts of the decision can AI support without receiving authority to determine the outcome?
- Who should remain authorized to choose among AI-generated options?
- What information should accompany each option so the decision-maker can assess it independently?
- How should uncertainty, assumptions, and limitations in the AI analysis be made visible?
- What would prevent the human decision-maker from merely selecting the AI's preferred option by default?
- Which decisions are consequential enough that AI should not rank the options at all?
- How should responsibility be recorded when AI materially shapes the available choices but a human makes the decision?
Preventing AI Recommendations from Becoming Unexamined Decisions
- Where are people likely to accept an AI recommendation because reviewing it feels slower than following it?
- What level of independent reasoning should be expected before a recommendation is accepted?
- Which decisions require the human to examine evidence rather than only the AI's conclusion?
- What warning signs should cause an AI recommendation to receive greater scrutiny?
- How can the workflow make disagreement with AI practically easy rather than procedurally burdensome?
- What patterns of repeated acceptance would indicate that formal human decision authority has become nominal?
- How should managers detect whether recommendations are functioning as de facto automated decisions?
Allowing Agents to Execute Actions Without Giving Them Unbounded Authority
- Which categories of action can an agent execute independently?
- What limits should apply to value, scope, duration, data access, affected parties, or system permissions?
- Which actions should always require explicit human authorization before execution?
- How should an agent behave when a requested action falls outside its delegated authority?
- Who has authority to create, modify, suspend, or revoke the agent's permissions?
- How should cumulative actions be monitored when individually permitted actions can create larger combined consequences?
- What evidence would justify expanding or narrowing the agent's execution authority over time?
Designing Human Approval for Work the Approver Did Not Personally Produce
- What must an approver understand before accepting AI-generated or agent-executed work?
- Which evidence should be available so approval does not depend only on a polished final output?
- How much underlying work can reasonably be delegated before meaningful approval becomes impossible?
- What expertise must the approver retain to identify important errors or missing considerations?
- How much time must be available for review if the approval is intended to be substantive?
- When should the approver require independent specialist input before accepting the result?
- How should the organization distinguish genuine approval from a procedural signature applied after the outcome is effectively fixed?
Assigning Accountability When Several AI Systems Contribute to One Outcome
- Which human or organizational role owns the final outcome when several AI systems contribute?
- Who is responsible for understanding how the different systems interact?
- How should responsibility be divided between platform owners, workflow owners, business owners, and operators?
- What happens when the immediate failure cannot be attributed to a single AI component?
- Who has authority to halt the combined workflow while the cause is investigated?
- What records are needed to reconstruct which systems and human decisions contributed to the outcome?
- How can the organization prevent distributed technical responsibility from becoming diffuse business accountability?
Separating Recommendation, Approval, Execution, and Accountability Across Roles
- Which role should generate or evaluate recommendations?
- Which role should possess formal approval authority?
- Who or what should execute the approved action?
- Who remains accountable for the result after execution?
- Where would combining these responsibilities create conflicts of interest or weak challenge?
- Where would separating them create unnecessary delay or handoffs?
- How should the organization document the complete chain so each participant understands both their authority and their limits?
Clarifying Ownership After an Automated Workflow Replaces Human Handoffs
- Which ownership responsibilities were previously carried implicitly by people at each handoff?
- Which of those responsibilities disappear when the handoffs are automated?
- Which responsibilities still require an explicit human or organizational owner?
- Who owns the end-to-end result once intermediate transfers are no longer visible?
- Who monitors whether automated transitions occur correctly?
- Where should exceptions leave the automated flow and return to human ownership?
- How can automation remove handoffs without also removing accountability that those handoffs previously made visible?
Delegating Reversible Low-Consequence Decisions to Automated Systems
- Which decisions are sufficiently low consequence and reversible to automate safely?
- How quickly can a mistaken decision be detected and corrected?
- What cumulative harm could arise from many individually minor automated decisions?
- Which rules or constraints should bound the system's discretion?
- What monitoring is appropriate when humans do not approve every individual case?
- What conditions should cause a decision type to return to human review?
- How should the organization verify that the automated decisions remain within the intended consequence range over time?
Requiring Explicit Human Approval for Difficult-to-Reverse Actions
- Which actions become costly, harmful, or impossible to reverse once executed?
- At what point in the workflow should human approval occur?
- What evidence and alternatives should the approver see before authorizing the action?
- Which expertise is necessary for the approval to be meaningful?
- Should more than one human be involved for particularly consequential or irreversible actions?
- How should time pressure be handled when waiting for approval also creates risk?
- What safeguards should prevent an agent from bypassing approval through a sequence of individually permitted actions?
Resolving Accountability After an Automated Decision Chain Fails
- Who owned the outcome that failed?
- Which human decisions, system behaviors, delegated authorities, and organizational rules contributed to the failure?
- Was any participant accountable for monitoring the full decision chain rather than only one component?
- Did escalation or stop authority exist at the point where the failure could still have been prevented?
- Which responsibilities were ambiguous or missing in the original design?
- How should accountability be corrected without attributing every technical failure to the nearest human operator?
- What structural change would reduce the likelihood of similar accountability gaps in future automated decision chains?
Designing Independent Challenge and Escalation
Setting Escalation Thresholds Based on Consequence, Uncertainty, and AI Autonomy
- Which combinations of consequence, uncertainty, and autonomy should trigger mandatory escalation?
- How should reversibility affect the threshold for human intervention?
- Which signals can be detected automatically and which require human recognition?
- Should thresholds differ for routine work, customer-facing actions, regulated decisions, or critical operations?
- Who has authority to define and update escalation thresholds?
- How can the organization prevent teams from lowering thresholds simply to improve speed?
- What evidence should be reviewed to determine whether current escalation thresholds are too strict or too permissive?
Introducing Independent Review for High-Consequence Decisions
- Which decisions are consequential enough to justify review by someone outside the producing team?
- What expertise and authority should the independent reviewer possess?
- At what stage can independent review still materially influence the outcome?
- What information should the reviewer receive about AI involvement and underlying evidence?
- Should review be universal for the decision type or triggered only by defined conditions?
- How can independence be preserved if the reviewer works closely with the team whose decision is being assessed?
- What evidence would justify reducing, strengthening, or redesigning independent review?
Routing Difficult Exceptions to Qualified Specialists
- What characteristics distinguish a specialist exception from a case a generalist can still handle?
- How should the workflow identify those characteristics early enough to prevent weak decisions?
- Which specialist group should receive each category of exception?
- What context and prior analysis should accompany the escalation?
- How quickly must specialist support be available for the operating model to remain workable?
- What should happen when specialist views conflict with the operating team's preferred course?
- How should recurring exceptions be used to improve the routine workflow rather than permanently increasing specialist demand?
Correcting Oversight Roles That Lack Real Authority to Intervene
- What is the oversight role formally expected to prevent or control?
- Can the overseer actually pause, reject, modify, or escalate the work?
- Does the overseer have access to the information needed to recognize a problem?
- Is the role evaluated or rewarded in ways that discourage intervention?
- Which authority needs to move so oversight becomes operationally meaningful?
- Would strengthening the role create excessive duplication with existing decision owners?
- How can the organization verify that oversight authority is exercised in practice rather than existing only on paper?
Clarifying Override and Stop Authority for Autonomous Work
- Who can interrupt an autonomous system when its behavior becomes unsafe, inappropriate, or unexpected?
- Under what conditions may an operator override an AI-generated decision or action?
- Can people intervene quickly enough before consequential downstream actions occur?
- Does stopping one agent or workflow create unacceptable effects elsewhere?
- Who decides when a stopped system may resume operation?
- How should override actions and reasons be documented and reviewed?
- What happens if responsibility for stopping the system is shared so widely that nobody acts?
Protecting Reviewers from Routine Acceptance of AI Recommendations
- Are reviewers seeing enough independent evidence to challenge the AI recommendation?
- Does workload allow them sufficient time to conduct real review?
- What patterns would indicate automation bias or habitual acceptance?
- Should reviewers form an initial judgment before seeing the AI recommendation in some cases?
- How can disagreement with the AI be recorded without creating unnecessary procedural burden?
- What competence must reviewers maintain to assess recommendations independently?
- How should review design change if acceptance rates become unusually high or almost automatic?
Redesigning Escalation When Review Queues Become Bottlenecks
- Which cases currently enter review even though they rarely require intervention?
- What characteristics distinguish cases that genuinely need specialist or managerial attention?
- Could routine cases move to automated checks, sampling, or retrospective monitoring?
- Which reviewers are overloaded because too many unrelated decision types converge on them?
- Can review authority be distributed without weakening standards?
- What risks would increase if the organization moved from universal to exception-based review?
- How should throughput, quality, and missed exceptions be monitored after redesigning the escalation model?
Preserving Separation of Duties After Work Becomes More Automated
- Which duties are intentionally separated today to prevent conflicts, errors, or abuse?
- Does automation make it technically possible but organizationally inappropriate to combine those duties?
- Could one agent now perform steps that previously required independent human actors?
- Where must independent human or system controls remain despite efficiency gains?
- How should separation be maintained when several duties are embedded inside one automated workflow?
- Which combinations of access and authority would create unacceptable concentration?
- How can the organization preserve the control objective while simplifying unnecessary procedural separation?
Preventing Central Assurance Functions from Becoming Universal Approval Queues
- Which decisions genuinely require central assurance rather than local ownership?
- What low-risk cases can be handled through predefined standards without case-by-case review?
- Are central reviewers evaluating risk or simply repeating work already done locally?
- Which controls could be embedded into platforms or workflows instead of performed manually?
- What exceptions should still reach central assurance?
- How can local teams remain accountable when they operate within centrally defined boundaries?
- What evidence would show that central assurance has become a bottleneck rather than a source of valuable challenge?
Moving from Universal Review to Risk-Based and Exception-Based Review
- Which current reviews rarely change the outcome?
- What risk factors should determine whether a case receives human review?
- Which cases are suitable for sampling rather than universal inspection?
- How should the system detect novel or ambiguous cases that do not match known risk patterns?
- What minimum monitoring should remain even for cases that bypass manual review?
- How can the organization detect deterioration after review volume is reduced?
- What failure rate or emerging pattern should cause broader review to be reinstated?
Reducing Handoffs and Coordination Burden
Distinguishing Faster Handoffs from Truly Removed Handoffs
- Has GenAI eliminated the need for another organizational actor, or only accelerated the transfer between actors?
- Does the downstream role still perform distinct judgment, ownership, or control?
- What waiting time has actually disappeared from the end-to-end process?
- Has work shifted from explicit handoff activity into hidden review or system orchestration?
- Could one role now own both sides of the handoff coherently?
- What would be lost if the separate organizational actors were combined?
- What evidence would demonstrate that the dependency itself has disappeared rather than merely becoming faster?
Allowing One Role to Absorb an Adjacent Workflow Step
- Which adjacent step can the role now perform credibly with GenAI support?
- What knowledge or capability previously justified passing the work to another role?
- Does the role have enough expertise to evaluate the additional step?
- Which approvals or specialist interventions would still remain necessary?
- Would absorbing the step reduce waiting and fragmentation enough to justify broader responsibility?
- What additional workload or cognitive burden would the expanded role carry?
- How should authority and accountability change when the role absorbs the adjacent work?
Removing Routine Approval Transfers That No Longer Add Judgment
- Which approvals are consistently granted without substantive changes to the underlying work?
- What original risk or information gap justified the approval?
- Has GenAI or improved information access removed that justification?
- Could clear thresholds allow routine cases to proceed without approval?
- Which cases should still escalate because they involve unusual consequence or uncertainty?
- Who remains accountable once the routine approval step disappears?
- How will the organization monitor whether removing approval changes error rates, risk, or decision quality?
Reducing Translation Work Between Specialist Functions
- Where do teams spend significant effort translating terminology, formats, or assumptions for one another?
- Which translation activities can GenAI perform reliably?
- Does easier translation reduce the need for specialist intermediaries or only speed their work?
- What tacit differences between functions cannot be solved through language translation alone?
- Which misunderstandings reflect conflicting incentives rather than communication difficulty?
- Could shared AI-supported context reduce repeated explanatory work across functions?
- What organizational boundary should remain even if most routine translation becomes automatic?
Replacing Repetitive Coordination with Shared AI-Enabled Context
- Which meetings, status updates, or reporting routines exist mainly to keep participants informed?
- Can shared AI-enabled context provide the same situational awareness asynchronously?
- What information must remain current for participants to rely on the shared context?
- Which coordination conversations involve negotiation or judgment that cannot be replaced by shared information?
- Who owns the accuracy and maintenance of the shared context?
- Could reducing routine coordination create weaker relationships or fewer opportunities to detect emerging problems?
- What evidence would show that shared context has reduced coordination burden without reducing organizational awareness?
Shortening Cross-Functional Waiting Time Without Losing Ownership
- Which cross-functional waits contribute most to end-to-end cycle time?
- Is the delay caused by workload, unclear ownership, information gaps, or approval requirements?
- Can GenAI prepare the receiving function's work before a formal transfer occurs?
- Which responsibilities must remain with the receiving function despite efforts to shorten the wait?
- Could service-level commitments or embedded capability reduce delay without removing the boundary?
- How should urgent cases bypass ordinary queues without undermining prioritization?
- What change would improve flow while keeping accountability explicit at each stage?
Preventing Removed Handoffs from Reappearing as Review Work
- Which handoff was removed, and what function did it previously perform?
- Has the receiving role become dependent on informal checking by the role that was supposedly removed from the workflow?
- Is new review work consuming the same effort as the old handoff?
- What capability gap causes the new owner to seek repeated reassurance or correction?
- Should more expertise move into the broader role, or should the original boundary be partially restored?
- Could clearer standards reduce the need for informal review?
- How will the organization measure whether the redesigned workflow genuinely reduced total dependency?
Identifying Coordination Hidden Inside Agentic Workflows
- Which human interactions have disappeared from view because agents now coordinate the steps automatically?
- What systems, teams, data sources, permissions, or approvals still support the apparently seamless workflow?
- Where can failure in one hidden dependency propagate across the workflow?
- Who understands the full dependency chain well enough to own it?
- Which coordination burden has moved from people into platform or technical teams?
- Does the agentic workflow simplify the organization or merely conceal complexity behind one interface?
- What dependencies should be eliminated rather than simply automated?
Reducing Repeated Transfers of Ownership Across One Outcome
- How many times does formal or informal ownership change before the outcome is complete?
- Which transfers reflect genuine changes in responsibility and which are historical workflow artifacts?
- Where does each transfer create delay, ambiguity, or loss of context?
- Could one role, team, or product owner retain responsibility across more of the process?
- Which specialist contributions can occur without transferring overall ownership?
- What authority would an end-to-end owner need to coordinate across remaining functions?
- How would fewer ownership transfers affect accountability when something goes wrong?
Rebundling Sequential Work Around End-to-End Responsibility
- Which sequential activities can now be performed by one role or team with GenAI support?
- What value would come from owning the sequence as one outcome rather than several functional outputs?
- Which specialist or control steps must remain distinct within the broader responsibility?
- Would rebundling reduce handoffs enough to justify broader role or team scope?
- What new skills and authority would the end-to-end owner require?
- Could the broader bundle become too complex for one accountable owner to understand?
- What measures should evaluate the end-to-end result rather than the performance of individual stages?
Redrawing Functional and Organizational Boundaries
Reconsidering a Functional Boundary Built Around Scarce Expertise
- What expertise originally justified separating this work into a distinct function?
- How much of that expertise can now be accessed credibly through GenAI by people outside the function?
- Which parts of the function still require deep professional competence?
- What routine work could move outward without eliminating the specialist community?
- Would weakening the boundary improve end-to-end ownership or create inconsistent practice?
- Which standards, development, or assurance responsibilities require the function to remain distinct?
- What evidence would justify changing the boundary rather than simply improving access to the function?
Preserving a Boundary That Provides Independent Challenge
- What conflicts or risks is the boundary intended to protect against?
- Would combining the functions weaken the ability of one side to challenge the other?
- Can GenAI improve information sharing without removing organizational independence?
- Which decisions require visibly separate ownership or approval?
- What incentives could distort challenge if both activities were placed under the same leader?
- Can the boundary be made more permeable operationally while preserving independence structurally?
- What evidence would be required before concluding that separate organizational homes are no longer necessary?
- Which geographic structures exist partly because language, communication, or information transfer was historically difficult?
- How much local work can now be supported effectively from elsewhere using GenAI?
- Which local responsibilities remain dependent on relationships, regulation, culture, or physical operations?
- Could some support capabilities be consolidated while preserving local commercial or operational ownership?
- Does easier translation actually remove the need for local expertise?
- What resilience or time-zone advantages would be lost through geographic consolidation?
- Which geographic boundaries should remain organizational even if cognitive work becomes location-independent?
Redrawing Boundaries That Create Repeated Cross-Functional Dependencies
- Which functions interact so frequently that their boundary creates persistent coordination costs?
- What different expertise, incentives, or accountability currently justify keeping them separate?
- Has GenAI reduced those differences enough to support a new boundary?
- Could work be reorganized around products, customers, journeys, or outcomes instead of functions?
- Which specialist communities would need another organizational mechanism if the functions were integrated?
- Would merging the boundary reduce coordination or simply relocate conflicts inside a larger unit?
- What structural alternative best preserves necessary specialization while reducing repeated dependency?
Reconsidering Boundaries Between Frontline and Back-Office Work
- Which work is sent to the back office because frontline roles historically lacked expertise or processing capacity?
- Can GenAI now enable frontline employees to resolve more cases at the point of need?
- Which back-office activities still benefit from scale, specialization, independence, or concentration?
- How much additional responsibility can frontline roles absorb without harming customer or operational performance?
- Which cases should move directly to specialists rather than remain with the frontline?
- What authority and information access would need to move with redistributed work?
- Would changing the boundary reduce total effort or merely shift support work into the frontline role?
Preserving Specialist Units for Standards, Depth, and Exceptional Cases
- Which specialist unit responsibilities remain valuable even if routine work moves elsewhere?
- Does the unit need enough routine work to maintain expertise, or can difficult cases sustain capability?
- What standards should the unit own across distributed execution?
- How should teams access the unit for exceptional cases?
- Which decisions require formal specialist authority?
- How can the unit remain connected to real operating work rather than becoming detached from practice?
- What size and structure would support its enduring responsibilities without preserving obsolete production capacity?
Clarifying Boundaries That Become Blurred by Shared AI Systems
- Which teams now rely on the same AI systems for work that used to be clearly separated?
- Has shared technology created ambiguity about who owns the resulting output or process?
- Which responsibilities belong to the platform owner and which remain with the business function?
- Can one team's changes to a shared AI capability affect another team's outcomes?
- Where should common standards apply despite separate functional ownership?
- How should exceptions be handled when the cause crosses both technical and business boundaries?
- What organizational boundary needs to become clearer even though the underlying technology is shared?
Reconsidering Units Whose Main Purpose Is Information Transformation
- How much of the unit's work consists of summarizing, formatting, translating, consolidating, or routing information?
- Which of those activities can now be automated reliably?
- What other judgment, relationship, assurance, or coordination functions does the unit perform?
- Could those remaining functions be absorbed more naturally elsewhere?
- Would removing the unit reduce cycle time or create unowned integration work?
- What knowledge or context would be lost if the unit disappeared?
- Should the unit shrink, change purpose, merge with another function, or cease to exist?
Preserving Separate Units Where Local Context Still Matters Materially
- Which decisions or activities depend on context that cannot be captured reliably in shared AI systems?
- What customer, regulatory, cultural, operational, or market knowledge remains genuinely local?
- Would consolidation reduce responsiveness or ownership in ways that exceed its efficiency benefits?
- Can shared platforms reduce duplicated work while separate units preserve contextual authority?
- Which capabilities should be standardized and which should remain locally distinct?
- How should the organization prevent local context from becoming a blanket justification for unnecessary duplication?
- What evidence would show that the value of separate units remains greater than the coordination cost they create?
Testing Whether Organizational Units Should Merge, Split, or Change Purpose
- What organizational problem is the current unit intended to solve?
- Has GenAI changed the scale, scope, or type of work enough to challenge that purpose?
- Are two units increasingly performing overlapping work that could be combined?
- Has one unit developed sufficiently different responsibilities that it should split?
- Could the unit preserve its organizational identity while shifting toward a fundamentally different purpose?
- What dependencies, capabilities, and accountability would change under each alternative?
- Which option best reflects the enduring work rather than the historical organization chart?
Reconsidering a Functional Structure After Expertise Becomes More Accessible
- Which advantages of functional specialization remain strong after GenAI broadens access to expertise?
- Which functional handoffs create coordination costs that can now be avoided?
- Can business or product teams perform more specialist work without weakening professional standards?
- What capabilities still need functional homes for development, depth, or independent authority?
- Would greater end-to-end ownership outperform functional efficiency in the most important workflows?
- Could a hybrid preserve functional communities while organizing delivery around outcomes?
- What evidence would justify moving away from the functional structure rather than merely improving its interfaces?
Reconsidering a Matrix Structure When Coordination Costs Change
- Which competing dimensions originally justified the matrix?
- Does GenAI reduce the information and coordination burden that made dual reporting necessary?
- Are employees still managing genuine conflicting priorities that require two lines of authority?
- Could clearer outcome ownership replace some matrix relationships?
- Does AI-supported transparency simplify the matrix or enable even more central intervention?
- Which specialist or geographic dimensions still deserve formal structural representation?
- Would simplifying the matrix remove ambiguity or eliminate a coordination mechanism the organization still needs?
Moving Toward Product or Outcome-Based Structures
- Which outcomes currently span several functions without one accountable owner?
- Can smaller GenAI-enabled teams now contain enough capability to own those outcomes end to end?
- Which specialists can move from routine participation to shared or on-demand support?
- What authority would product or outcome owners require across functional boundaries?
- Which shared standards and platforms should remain outside the product team?
- How should career development and professional communities work when people are organized primarily around outcomes?
- What evidence would show that product or outcome structures improve end-to-end performance rather than simply reorganize the same dependencies?
- Which capabilities should every outcome team consume from a common platform?
- Which decisions should remain with the outcome team rather than the platform organization?
- How should platform priorities reflect needs across many teams without becoming a central delivery queue?
- What standards should be built into the platform so teams can operate autonomously?
- How should costs and funding be divided between shared infrastructure and local applications?
- What happens when one outcome team needs capability that conflicts with enterprise platform standards?
- How can the organization preserve both local speed and enterprise reuse as the number of teams grows?
Reassessing Geographic Structures as Distance Matters Less for Cognitive Work
- Which geographic units exist because knowledge work historically needed to be performed locally?
- Which activities can now be centralized, distributed globally, or performed remotely with little loss?
- What local regulatory, customer, cultural, or physical responsibilities remain decisive?
- Could geographic units become smaller while retaining local decision authority?
- What follow-the-sun, resilience, or talent benefits might support a distributed model instead of centralization?
- How should local accountability work if more cognitive support comes from elsewhere?
- Which parts of the geographic structure reflect enduring market differences rather than old communication constraints?
Reassessing Customer-Based Structures as Personalization Becomes Cheaper
- Which customer segments currently justify separate teams because tailored service is expensive to deliver?
- Can GenAI provide meaningful personalization without maintaining distinct organizational units for every segment?
- Which customer groups still require dedicated relationships, expertise, or operating models?
- Could one platform support multiple segments while local teams retain customer ownership?
- Does cheaper personalization increase the number of economically viable segments rather than reduce them?
- What new complexity would arise from serving more differentiated customer needs?
- Should structural segmentation follow customer economics, relationship needs, or simply the ability to personalize outputs?
Using Networked Structures for More Autonomous Teams
- Which teams can operate with greater autonomy because they now have broader access to information and expertise?
- What common standards are necessary so autonomous teams remain interoperable?
- How should dependencies between teams be coordinated without recreating hierarchy?
- What authority should remain central over shared resources, risks, and strategic priorities?
- How should knowledge move across the network when teams develop local AI-enabled practices?
- What prevents autonomous teams from duplicating infrastructure or optimizing against one another?
- Which kinds of work are too interdependent or consequential for a networked structure to perform well?
Designing Hybrid Structures for Uneven GenAI Maturity and Risk
- Which parts of the organization are mature enough for more autonomous GenAI-enabled structures?
- Which areas still require stronger central support or control?
- How should structural differences reflect genuine risk and capability rather than political history?
- Can different operating models coexist without creating confusing interfaces?
- What shared architecture, governance, or performance system should connect the different structures?
- How should units move from one structural model to another as maturity changes?
- What conditions would indicate that the hybrid has become unnecessarily complex and should be simplified?
Reconsidering Divisional Structures When Shared GenAI Capabilities Span Business Units
- Which capabilities are currently duplicated across divisions because each historically needed its own resources?
- What GenAI-enabled capabilities now create economies of reuse across the divisions?
- Which divisional responsibilities still require independent ownership and local adaptation?
- Could shared platforms replace duplicated support without weakening divisional accountability?
- What decisions should remain with divisions even when capability becomes common?
- How should common investments be funded and prioritized across divisions with different needs?
- Does the divisional structure still reflect meaningful business differences, or increasingly duplicate the same cognitive capabilities?
- What does the redesigned work architecture require from the formal organization?
- Which boundaries have become more or less important after GenAI changes the work?
- Where must authority, expertise, platforms, and outcome ownership sit?
- Which structural alternatives best fit the actual dependencies rather than a preferred management philosophy?
- What tradeoffs does each alternative create in speed, specialization, coordination, accountability, resilience, and development?
- Which parts of the future structure should remain deliberately flexible because GenAI capability is still changing?
- What evidence or pilot would best test the preferred structural form before full implementation?
Creating End-to-End Outcome Ownership
Assigning Ownership Where One Outcome Crosses Several Functions
- Which outcome currently depends on several functions without one clearly accountable owner?
- Where does responsibility become fragmented as work moves between functions?
- Who is best positioned to own the outcome rather than only one functional contribution?
- What authority would the outcome owner need across the participating functions?
- Which functional responsibilities should remain independent even under end-to-end ownership?
- How should conflicts between functional priorities and outcome priorities be resolved?
- What measures would allow the owner to be held accountable for the complete result?
- Which ownership transitions have disappeared because GenAI now connects previously separate stages of work?
- Does responsibility still change hands formally even though the operational handoff no longer exists?
- Could one person or team now retain ownership across a larger portion of the process?
- Which responsibilities previously performed at intermediate handoffs still need explicit owners?
- What authority must move with the consolidated ownership?
- Could consolidating ownership create excessive concentration of responsibility or weaken independent challenge?
- How should the organization verify that fewer ownership transfers actually improve accountability?
Giving Product Teams Responsibility for End-to-End Results
- What outcome should the product team own beyond delivery of features or outputs?
- Which functional dependencies currently prevent the team from controlling that outcome?
- What decision rights would the team need over priorities, workflow, data, and AI-enabled capability?
- Which specialist functions should support the team without taking ownership away from it?
- How should shared platform or enterprise constraints limit product-team autonomy?
- What measures should replace function-specific output metrics for the team?
- Where should accountability remain outside the product team because the consequences exceed its scope?
Assigning Process Ownership Across Humans, Agents, and Systems
- Who should own the complete process when execution is distributed across people, agents, and conventional systems?
- Which parts of the process can operate autonomously without changing overall ownership?
- Who is responsible for ensuring that handoffs between human and automated actors work correctly?
- Which failures belong to the process owner even when the immediate cause lies in a technical component?
- What visibility does the owner need into agent and system performance?
- Which technical responsibilities should remain with platform or system owners rather than the process owner?
- How should process ownership be expressed so accountability remains clear despite distributed execution?
Establishing Ownership Across an Entire Customer Journey
- Which parts of the customer journey currently have separate owners whose local objectives can conflict?
- What customer outcome should one owner or team be responsible for across the journey?
- Which organizational boundaries create repeated loss of context or responsibility for the customer?
- How can GenAI reduce coordination across journey stages without obscuring who owns the overall experience?
- Which functions must retain independent authority even within a journey-oriented model?
- What data and decision rights would an end-to-end journey owner need?
- How should journey performance be measured when many functions still contribute to the outcome?
- Which outcomes should remain owned locally even though the underlying AI capabilities are shared centrally?
- What platform decisions materially constrain the local owner's ability to deliver the outcome?
- Which responsibilities belong to the local owner and which belong to the platform owner?
- How should local teams influence platform priorities when their outcomes depend on shared capabilities?
- What happens when a platform limitation prevents a local team from meeting its commitments?
- How should accountability be divided when a shared platform failure harms a locally owned outcome?
- What governance would preserve local ownership without encouraging teams to bypass shared platforms?
Resolving Conflicting Functional Metrics Around One Outcome
- Which functions contributing to the same outcome are currently rewarded for different or conflicting measures?
- How do those metrics drive behavior that weakens the end-to-end result?
- Which shared outcome measures should take precedence over local activity measures?
- What functional measures still need to remain because they protect quality, risk, or professional standards?
- Who should arbitrate tradeoffs when local metrics conflict with the shared outcome?
- How should incentives reflect both individual functional contribution and collective outcome performance?
- What evidence would show that metric alignment has improved real ownership rather than only reporting?
Correcting Outcome Ownership That Lacks Matching Decision Rights
- What outcome is a person or team formally accountable for without controlling the key decisions that shape it?
- Which approvals, resources, priorities, or dependencies sit outside the owner's authority?
- Which decision rights should move to make accountability credible?
- Which decisions must remain elsewhere because of legitimate enterprise or control requirements?
- How should unresolved dependencies be governed when the owner cannot control them directly?
- Does the owner have enough authority to stop work that threatens the outcome?
- What structural change would align accountability, authority, resources, and information more coherently?
- Who inside the organization remains accountable when most execution is performed externally or by AI?
- Which responsibilities cannot be delegated even when delivery is outsourced or automated?
- Who should evaluate whether the external or AI-produced work remains fit for purpose?
- What authority does the internal owner need over vendors, platforms, agents, and supporting teams?
- How should failures be handled when the immediate cause lies outside the owner's direct control?
- What internal capability must remain so the organization can challenge or replace the external provider?
- How can ownership remain meaningful rather than becoming nominal contract administration?
Defining Where One End-to-End Outcome Begins and Ends
- What event or need marks the beginning of the outcome being owned?
- What completed state should count as the end rather than an intermediate functional output?
- Which upstream factors materially affect the outcome but belong to a different owner?
- Which downstream consequences should remain part of the owner's responsibility?
- Where would an ownership boundary create the least ambiguity between adjacent outcomes?
- Are current boundaries based on organizational history rather than the actual flow of value?
- How should overlapping outcomes be coordinated when no clean boundary is possible?
Balancing Centralization, Federation, and Local Autonomy
Centralizing Enterprise-Wide Dependencies Without Centralizing Local Delivery
- Which capabilities or dependencies create enterprise-wide consequences if managed inconsistently?
- Which of those should be centralized as infrastructure, standards, or control rather than as delivery work?
- What local decisions can remain autonomous once common dependencies are managed centrally?
- How should central teams provide reusable capability without becoming owners of local outcomes?
- Which interfaces must be explicit between enterprise capability owners and local delivery teams?
- How can local teams influence centralized dependencies that materially affect their performance?
- What would indicate that centralizing a dependency has unintentionally centralized too much operational control?
Granting Greater Autonomy as Domain Capability Matures
- What capabilities must a domain demonstrate before receiving greater autonomy?
- Can the domain manage its own AI-enabled workflows through their full lifecycle?
- Does it understand and manage the relevant business, technical, and risk consequences?
- Which decisions can move outward immediately and which should remain central until further maturity develops?
- What common standards should remain mandatory regardless of local capability?
- How should autonomy be reduced if performance, control, or reliability deteriorates?
- What evidence should trigger the next step from central support toward greater local ownership?
Managing Rapid Local Experimentation Across Business Units
- Which kinds of experimentation can business units conduct without central approval?
- What minimum visibility should the enterprise retain over local GenAI experiments?
- How can units experiment quickly without duplicating common infrastructure unnecessarily?
- Which risks or dependencies should automatically bring an experiment into central review?
- How should useful local discoveries be shared across other units?
- Who decides when a local experiment should become a reusable enterprise capability?
- How can the organization preserve experimentation without allowing a fragmented shadow AI estate to emerge?
Combining Central Standards with Local Delivery Ownership
- Which standards must remain common across all local delivery teams?
- Where should teams have freedom to adapt workflows, models, and operating practices to their domain?
- Can common standards be embedded into platforms rather than enforced through manual approvals?
- Who owns interpretation when a local requirement conflicts with a central standard?
- How should exceptions to enterprise standards be approved and documented?
- What responsibilities remain with central teams after local teams own delivery?
- How can the organization determine whether standards are enabling safe autonomy or constraining legitimate local adaptation?
Varying Local Autonomy According to Risk and Consequence
- Which characteristics of a GenAI-enabled activity should determine how much local autonomy is appropriate?
- How should consequence, reversibility, regulatory exposure, data sensitivity, and AI autonomy affect the model?
- Which low-risk activities can be managed almost entirely within local teams?
- Which high-risk activities require central approval, independent challenge, or shared ownership?
- Can one business unit operate under different autonomy levels for different workflows?
- Who should classify the risk level and resolve disagreements about it?
- How should autonomy change when evidence shows that a workflow is more or less risky than originally assumed?
Reducing Duplicated GenAI Capability Across Autonomous Units
- Which units are independently building capabilities that solve essentially the same problem?
- Is the duplication justified by genuinely different domain requirements?
- Which components could become shared without removing local ownership of the outcome?
- What incentives currently encourage teams to build locally rather than reuse existing capability?
- How should ownership of a shared replacement capability be established?
- What migration burden would consolidation create for teams already operating their own solutions?
- How can the organization reduce waste without suppressing useful local innovation?
Correcting a Central Model That Has Become a Delivery Bottleneck
- Which decisions or development activities are waiting unnecessarily for the central AI team?
- What responsibilities does the center still perform that mature domain teams could own themselves?
- Which central controls could be turned into reusable standards, tools, or automated guardrails?
- What domain capabilities are missing that currently justify continued central involvement?
- How should authority be transferred without creating inconsistent local practices?
- What work should remain central because it genuinely benefits from enterprise scale or independence?
- What evidence would show that the new model has reduced dependence on the center without increasing fragmentation?
Constraining Local Autonomy When Enterprise-Wide Risks Accumulate
- Which local AI-enabled decisions create risks that only become visible in aggregate?
- Are several teams relying on the same providers, data, agents, or technical patterns without realizing the concentration?
- Which enterprise consequences cannot be managed adequately by individual units?
- What central limits should apply without taking over local operating decisions?
- How should the center detect emerging aggregate risks across otherwise compliant local activities?
- Who has authority to suspend or restrict local practices when enterprise exposure becomes unacceptable?
- How should local autonomy be restored once the underlying risk has been reduced?
Clarifying Responsibilities in a Federated Operating Model
- Which responsibilities belong unequivocally to the enterprise center?
- Which responsibilities belong unequivocally to domains or business units?
- Which responsibilities are genuinely shared and therefore need explicit interfaces?
- Who owns business outcomes when technical capability is provided centrally?
- Who owns lifecycle management for locally developed AI-enabled workflows?
- How should disputes over architecture, risk, funding, or priorities be resolved?
- What responsibilities are currently falling between the center and the domains because neither side clearly owns them?
Moving from Centralized Delivery Toward Federation Over Time
- Which centralized activities were necessary because local GenAI capability was initially scarce?
- Which domains now have enough capability to take direct ownership of design and delivery?
- What reusable platforms and standards must exist before responsibility moves outward?
- Which central roles should shrink, change purpose, or disappear as federation matures?
- How should knowledge be transferred from central experts to local teams?
- What sequencing would avoid transferring responsibility before domains can manage it safely?
- What would the mature center still need to own once most delivery becomes federated?
Deciding Which GenAI Capabilities Should Be Reusable Across the Enterprise
- Which capabilities solve substantially similar needs across multiple functions or business units?
- Where would a shared capability create meaningful economies of scale or learning?
- Which domain differences are too important for a common solution?
- What parts of a capability should be standardized and what parts should remain configurable?
- Who should own the roadmap for a capability serving many internal customers?
- How should demand from competing business units be prioritized?
- What evidence would justify turning repeated local solutions into one enterprise capability?
Providing Common Model Access Across Multiple Business Units
- Which model providers or model families should be available through a common enterprise access layer?
- What value would common access create in cost, security, reliability, or portability?
- Which business units require model choices that differ from the enterprise default?
- How should usage limits, costs, permissions, and availability be allocated across units?
- Who decides when a new model becomes approved for enterprise use?
- How can common access avoid locking the organization unnecessarily into one provider?
- What responsibilities remain with local teams when model access itself is centralized?
Organizing Shared Retrieval and Enterprise Knowledge Access
- Which organizational knowledge should be accessible across multiple GenAI-enabled workflows?
- Who owns the underlying sources and decides which uses are permitted?
- How should access respect different permissions even when the retrieval layer is shared?
- What information should remain domain-specific rather than enter an enterprise knowledge layer?
- Who is responsible for correcting stale, conflicting, or misleading knowledge?
- How should local teams contribute useful knowledge without losing ownership of authoritative sources?
- What organizational model would keep shared retrieval useful without creating a new central information bottleneck?
Providing Common Infrastructure for Agentic Work
- Which agent capabilities should be supplied once for reuse across the organization?
- What runtime, orchestration, logging, identity, and monitoring services should be common?
- Which parts of agent behavior should remain configurable by domain teams?
- How should local teams deploy agents without building their own incompatible infrastructure?
- Who owns reliability when a shared agent platform supports many business-critical workflows?
- How should enterprise infrastructure accommodate different autonomy and risk levels?
- What dependencies would become dangerous if too many workflows relied on one shared agent platform?
Centralizing Reusable Evaluation and Monitoring Capability
- Which evaluation methods and monitoring tools can be shared across different GenAI use cases?
- Which quality criteria remain domain-specific even when the infrastructure is common?
- Who should operate the shared evaluation capability?
- How should domain experts contribute to tests without requiring the central team to understand every workflow?
- What performance signals should be visible both centrally and locally?
- When should a common evaluation result prevent a local deployment or trigger intervention?
- How can shared evaluation create leverage without turning the central team into the approver of every use case?
Providing Shared Identity, Authorization, and Security Foundations
- Which identity and permission mechanisms should every GenAI-enabled system use?
- How should agent identities differ from human identities where necessary?
- Who defines what systems, data, and actions different agents may access?
- How should business teams request or change permissions without bypassing enterprise controls?
- What should happen when local workflow needs conflict with standard authorization patterns?
- How can access remain traceable as agents act across several systems?
- Which responsibilities stay with security or platform teams and which remain with the business owner using those permissions?
Building Domain Applications on Common Enterprise Platforms
- Which platform capabilities can domain teams use without relying on central developers?
- What domain-specific logic should remain owned by the local application team?
- How much configuration freedom should the shared platform permit?
- Which enterprise standards should be enforced automatically by the platform?
- What happens when a domain requires a capability the platform does not yet support?
- How should platform improvements created for one domain become reusable by others?
- How can the organization avoid forcing highly different use cases into a platform that no longer fits them?
Consolidating Duplicated Local GenAI Infrastructure
- Which local infrastructure components perform substantially the same enterprise function?
- What legitimate needs led teams to create separate infrastructure in the first place?
- Can a shared replacement meet those needs without reducing local delivery speed?
- What technical and organizational migration costs would consolidation create?
- Who should own the consolidated capability and its future priorities?
- How should local teams retain influence once infrastructure ownership moves away from them?
- What duplication should remain because it provides resilience, independence, or necessary domain specialization?
Funding Shared Capabilities Whose Benefits Span Multiple Units
- Which GenAI capabilities create value across units but lack a natural single business owner?
- Should funding come from enterprise budgets, usage charges, shared contributions, or another mechanism?
- How might internal chargebacks discourage beneficial reuse?
- How should investments be prioritized when different units expect different levels of benefit?
- Who has authority to fund foundational capability before a direct return is visible?
- How should ongoing operating costs be allocated once adoption expands?
- What funding model would prevent shared capabilities from becoming either chronically underfunded or detached from real business demand?
Turning a Successful Local GenAI Solution into an Enterprise Capability
- What evidence shows that the local solution solves a recurring problem beyond its original domain?
- Which parts are reusable and which depend too heavily on local context?
- Who should own the capability after it becomes enterprise-wide?
- What technical, governance, support, and reliability work is required before broader reuse?
- How should the originating team remain involved without becoming the permanent support desk for the enterprise?
- What changes are necessary so other teams can adopt the capability safely without recreating it?
- When should a successful local solution remain local rather than being generalized prematurely?
Redesigning Shared Services, Staff Functions, and Centers of Excellence
Automating Routine Production Within a Shared Service
- Which high-volume shared-service activities can GenAI perform reliably?
- How much human effort is genuinely removed once review, exceptions, and correction are included?
- Which service requests still require human judgment or relationship management?
- How should staffing change as routine case volume declines?
- Does automation allow the shared service to support more scope rather than simply reduce capacity?
- What new monitoring or exception work emerges as production becomes automated?
- How should service performance measures change when routine output becomes much cheaper to produce?
Reconsidering a Staff Function Built Mainly Around Information Intermediation
- How much of the staff function's work consists of collecting, synthesizing, translating, or routing information?
- Can leaders and operating teams now access that information more directly?
- What judgment, integration, standards, or challenge does the staff function provide beyond information transmission?
- Which responsibilities could move into operating units without weakening enterprise coordination?
- Could the function shift toward solving harder cross-organizational problems?
- What would become ownerless if the function were reduced or removed?
- Does the remaining value justify a distinct staff function, a smaller expert team, or no separate unit?
Using a Center of Excellence to Concentrate Scarce Early Expertise
- Which GenAI expertise is currently too scarce to distribute broadly across the organization?
- What early responsibilities should the Center of Excellence own directly?
- Which standards, reusable assets, and learning should it create for others?
- How should business units access the center without turning it into an uncontrolled request queue?
- What work should the center explicitly avoid owning because business context is essential?
- What indicators would show that local capability is becoming strong enough to reduce central delivery?
- What sunset or evolution criteria should be defined when the center is created?
Correcting a Center of Excellence That Has Become an Approval Bottleneck
- Which decisions require the Center of Excellence even though they are routine or low risk?
- How much delay does mandatory central review create?
- Which decisions could be delegated through clear standards or preapproved patterns?
- Which central reviews genuinely add specialist judgment?
- Can platform controls replace some case-by-case approval?
- What capability must local teams develop before approval authority can move outward?
- How should the center's success be redefined so it is rewarded for enabling safe autonomy rather than controlling volume?
Shifting a Center of Excellence from Delivery Toward Enablement
- Which delivery responsibilities can mature business units now own directly?
- What reusable platforms, standards, training, or specialist support should the center provide instead?
- How should central experts transfer knowledge without becoming permanently embedded in every domain?
- Which complex or novel cases should still receive direct central involvement?
- What happens to central staffing as local delivery capacity grows?
- How should the center measure leverage across the organization rather than the volume it delivers itself?
- When should the center become a smaller permanent capability rather than a transformation organization?
Moving Shared Services Toward Exception-Based Human Work
- Which routine service cases can GenAI handle without human intervention?
- What conditions should route a case to a human operator?
- Are remaining human cases likely to be more difficult, emotionally sensitive, or consequential?
- What new skills do shared-service employees need when their work becomes primarily exception handling?
- How should staffing models account for peaks in exception demand?
- Can recurring exceptions be redesigned into the automated path without lowering service quality?
- What service measures should reflect the complexity of remaining human work rather than simple case volume?
Refocusing Staff Functions on Judgment, Standards, and Integration
- Which routine production responsibilities can staff functions stop performing?
- What enterprise standards still require a central owner?
- Where does the organization need independent judgment that operating units should not provide for themselves?
- Which cross-unit integration problems lack another natural organizational owner?
- How can the staff function support business teams without reclaiming operational ownership?
- What expertise must remain concentrated centrally for the refocused role to add value?
- How should performance be measured when the function's contribution becomes system quality rather than output volume?
Reducing Central Service Demand as Embedded Capability Grows
- Which services are local teams increasingly able to perform themselves?
- What evidence shows that embedded capability is reliable enough to replace central delivery?
- Which services should remain centralized because they benefit from scale or independent control?
- How should central staffing adjust as demand shifts outward?
- Could local self-service create duplicated work or inconsistent standards?
- What shared infrastructure would allow teams to self-serve without rebuilding central capability locally?
- How should responsibility transfer so local ownership becomes explicit rather than informal?
- What temporary purpose justified creating the AI transformation office?
- Which responsibilities should now move into permanent business, technology, people, or risk functions?
- Which capabilities still lack a clear long-term organizational home?
- Has the office become dependent on preserving a transformation agenda to justify its existence?
- What knowledge, relationships, and reusable assets need to transfer before closure?
- Which temporary roles should disappear rather than be converted automatically into permanent positions?
- What evidence would show that GenAI has become embedded enough for the transformation office to end?
Preserving Central Functions That Still Provide Scale or Independent Challenge
- Which central functions still create clear economies of scale even as local GenAI capability expands?
- Which functions must remain independent from operating units to provide credible challenge?
- What parts of their workload can still be automated or reduced?
- How should central functions avoid expanding control simply because they have better AI-enabled visibility?
- What interfaces should allow local teams to work quickly without bypassing necessary central responsibilities?
- Could the same value be provided through shared standards or platforms rather than a large central team?
- What evidence would justify preserving the central function despite broader decentralization elsewhere?
Rebalancing Workforce Composition
Shifting Workforce Capacity Between Functions as GenAI Changes Demand Unevenly
- Which functions are experiencing meaningful reductions in routine workload because of GenAI?
- Which functions are experiencing increased demand for review, integration, judgment, or implementation?
- Are staffing levels still based on the pre-GenAI distribution of work?
- Which employees have transferable capabilities that could move toward emerging bottlenecks?
- What retraining would make redeployment more viable than external hiring?
- Which shifts in demand are durable enough to justify permanent movement of capacity?
- How should workforce planning account for functions where GenAI expands demand rather than reducing it?
Rebalancing the Mix of Junior, Mid-Level, and Senior Professionals
- Which levels currently perform work that GenAI changes most materially?
- Is the organization becoming overly dependent on senior employees for work that juniors could perform with structured support?
- Are junior roles disappearing faster than the organization can replace their developmental function?
- Which mid-level responsibilities become more or less important as routine coordination and production change?
- What mix of experience levels is needed to sustain both current delivery and future capability?
- How will the organization create progression if the traditional pyramid becomes narrower?
- What evidence should guide changes to workforce shape rather than assumptions about which levels AI will affect most?
Shifting from Routine Support Roles Toward Expert Review and Exception Handling
- Which support roles lose workload as GenAI automates routine preparation and processing?
- What review, exception, or quality responsibilities are growing simultaneously?
- Which employees can transition credibly into more judgment-intensive work?
- What training and supervision would be necessary for the transition?
- Should some review responsibilities remain with deeper specialists rather than redesigned support roles?
- How would pay, status, and career progression need to change if the work becomes more complex?
- What roles should be retired rather than repurposed if the residual responsibilities do not form a coherent job?
Rebalancing Generalist and Specialist Capacity
- Which work can AI-enabled generalists now perform that previously required specialist participation?
- Which specialist activities become more important as routine work moves outward?
- Is the organization maintaining too many specialists for routine execution or too few for difficult review?
- What specialist capacity should remain embedded, pooled, or centralized?
- How many broad generalist roles can one specialist group support effectively?
- Does increased generalist capability create new risks of shallow expertise?
- What workforce mix best balances broad end-to-end ownership with access to deep judgment?
Reassessing the Ratio of Managers to Individual Contributors
- How has GenAI changed the amount and type of managerial work per employee?
- Are managers spending less time on administration but more on judgment, development, or agent supervision?
- Can individual contributors operate more autonomously than before?
- Has team output expanded enough to increase rather than reduce managerial demand?
- What proportion of the workforce now performs work that requires close coaching or oversight?
- Would fewer managers improve the organization, or simply increase hidden coordination and development gaps?
- What measure of managerial load should guide workforce ratios instead of historical benchmarks?
Determining How Much Dedicated Technical AI Expertise the Organization Needs
- Which technical AI responsibilities require dedicated specialists rather than ordinary technology teams?
- How much of that expertise should reside centrally versus within business domains?
- Which capabilities can increasingly be provided through managed platforms and external services?
- What scarce expertise must remain internal to govern, integrate, or challenge external AI systems?
- How many teams can one technical AI specialist or shared group support effectively?
- Is current demand temporary because GenAI capability is immature, or likely to remain structurally important?
- How should workforce planning avoid both permanent overstaffing and dangerous dependence on a very small technical group?
Redeploying People Whose Existing Work Shrinks
- Which employees face durable reductions in the work that currently occupies them?
- What adjacent organizational needs could use their domain knowledge or relationships?
- Which new roles require capabilities that these employees could develop realistically?
- How much retraining is justified compared with hiring externally?
- What institutional knowledge would be lost if experienced employees simply left?
- Can redeployment support emerging review, customer, integration, or AI-evaluation work?
- How should the organization avoid creating nominal redeployment roles that do not have enough real work or value?
Rebalancing Human Capacity Across Production, Review, Integration, and Judgment
- How is human capacity currently distributed across production, review, integration, and judgment?
- Which categories shrink as GenAI takes on more routine cognitive production?
- Which categories become bottlenecks as output volume increases?
- Do staffing plans recognize that one hour of production saved may create new review or integration demand elsewhere?
- Which roles can shift between these categories without losing necessary expertise?
- How should organizational capacity be measured when fewer people produce directly but more people oversee complex outcomes?
- What future mix would best match the actual scarcity in the GenAI-enabled work system?
Correcting a Workforce That Becomes Too Senior-Heavy
- Has reduced junior hiring increased the proportion of senior employees beyond what the work requires?
- Are senior people performing tasks that should become developmental opportunities for less experienced employees?
- Does the organization still have a viable pipeline for replacing senior experts and managers?
- Are labor costs rising because GenAI-enabled work is being concentrated in expensive experienced staff?
- Which responsibilities can move safely to junior employees with AI support and supervision?
- How should senior roles shift toward mentoring, difficult judgment, and system improvement?
- What workforce shape would preserve expertise without creating an unsustainable top-heavy structure?
Planning Human Capacity Alongside Growing Agent Capacity
- What work volume is increasingly performed by agents rather than human employees?
- Which human roles are required to supervise, evaluate, integrate, and improve that agent capacity?
- How should workforce planning account for the autonomy and complexity of agents rather than treating them as simple software?
- Which human bottlenecks will limit the value of adding more agents?
- Could additional agent capacity increase demand for specialist or managerial capacity rather than reduce it?
- What human redundancy remains necessary when critical work depends heavily on digital labor?
- How should the organization decide whether the next unit of capacity should be human, agentic, or a different workflow altogether?
Redesigning Entry-Level Work and Career Architecture
Redesigning Entry-Level Roles After Routine Learning Work Is Automated
- Which routine activities historically gave new employees the repetitions needed to build judgment?
- Which of those activities are disappearing because GenAI can now perform them more efficiently?
- What valuable responsibilities can replace automated production without expecting junior employees to operate at senior level immediately?
- How should the role combine productive contribution with deliberate capability development?
- Which AI-supported tasks can accelerate learning rather than bypass the reasoning the employee needs to develop?
- What supervision and feedback must accompany the redesigned entry-level work?
- How will the organization know whether the new role is actually producing competent future professionals?
Replacing Apprenticeship Paths That No Longer Arise Naturally from Production
- Which expertise was historically developed through progressive exposure to routine production work?
- What learning experiences disappear when that production is automated?
- Which real cases can still be assigned deliberately for developmental value even when AI could perform them faster?
- Where can simulation, shadowing, supervised practice, or structured case review replace lost repetition?
- How should employees progress from observing expert judgment to exercising it themselves?
- Which developmental experiences cannot be replaced adequately by formal training or AI tutoring?
- What new apprenticeship architecture would produce comparable expertise without depending on obsolete work?
Giving Junior Employees Supervised Responsibility in AI-Enabled Work
- What meaningful outcomes can junior employees own without exposing the organization to excessive risk?
- Which parts of the work should they perform themselves before using AI assistance?
- Where should supervisors review the reasoning process rather than only the final output?
- How should decision authority expand as employees demonstrate stronger judgment?
- Which mistakes can safely become learning opportunities rather than reasons to remove responsibility?
- How can AI support the employee without allowing weak underlying competence to remain hidden?
- What evidence should determine when supervision can become lighter?
Shifting Progression Criteria Toward Demonstrated Judgment
- Which current promotion criteria reward production volume, tenure, or task mastery that GenAI has made less scarce?
- What kinds of judgment should employees demonstrate before taking on broader responsibility?
- How can the organization assess decision quality across ambiguous or difficult cases?
- Which evidence would show that someone can recognize when AI output should be challenged or rejected?
- How should progression reflect the ability to integrate perspectives, manage tradeoffs, and handle exceptions?
- What safeguards are needed so subjective judgments about "good judgment" do not become opaque promotion criteria?
- How should demonstrated judgment interact with technical expertise, leadership potential, and role-specific capability?
- Which traditional career steps are losing enough work that they may no longer support distinct roles?
- What developmental experiences did those intermediate roles provide before they began shrinking?
- Can responsibilities from disappearing levels be redistributed into broader developmental roles?
- How can employees progress from entry-level work to senior responsibility without skipping necessary capability formation?
- Which lateral moves, rotations, or project assignments could replace missing hierarchical steps?
- How should compensation and status progress when the traditional title ladder becomes shorter?
- What new career architecture would preserve visible progression without recreating unnecessary organizational layers?
Using Rotations to Replace Narrow Repetition as a Learning Mechanism
- Which capabilities previously developed through years of repeated work in one narrow area?
- Could structured rotations expose employees to a broader range of cases more efficiently?
- Which functions or roles would provide complementary experiences needed for stronger judgment?
- How long must an employee remain in each rotation to learn more than surface-level terminology?
- What responsibilities should employees actually own during rotations rather than merely observe?
- How should learning from different rotations be integrated into a coherent professional capability?
- When would deep sustained practice be more valuable than broader rotational exposure?
- Which existing promotion signals depend heavily on visible production that GenAI can now increase cheaply?
- What outcome, quality, judgment, collaboration, or capability-building measures should carry greater weight?
- How should the organization recognize people who improve shared systems rather than maximize personal output?
- What evidence should distinguish genuine higher-level contribution from sophisticated use of AI to generate more artifacts?
- How should difficult review, mentoring, and integration work be credited when it produces fewer visible outputs?
- Which old promotion signals should be retired rather than simply supplemented?
- How can revised criteria remain understandable enough for employees to plan their development?
Avoiding Entry-Level Roles That Demand Senior Judgment Without Development
- Which junior roles are losing routine work while retaining difficult exceptions and ambiguous decisions?
- Are entry-level employees being expected to supervise AI outputs they lack the experience to evaluate?
- What foundational practice must occur before someone can own the remaining higher-judgment work?
- Which responsibilities should stay with experienced employees until a structured development path exists?
- Could AI-generated cases, simulations, or supervised work provide safe intermediate steps?
- How should workload expectations change if junior work becomes more cognitively demanding?
- What signs would show that the redesigned entry role has become an unrealistic compressed version of a senior job?
Preserving the Future Management Pipeline as Early-Career Work Changes
- Which current managers developed their judgment through roles that GenAI is now shrinking or removing?
- What operating experiences will future managers need if traditional feeder roles change substantially?
- How can early-career employees gain responsibility for people, resources, tradeoffs, and decisions before formal management?
- Which developmental assignments should replace managerial preparation that used to occur implicitly through progression?
- How should organizations identify potential managers when individual production becomes less revealing?
- Could a narrower junior workforce create future shortages of credible management candidates?
- What workforce and career investments are necessary now to preserve management capability several years from today?
Designing Parallel Generalist, Specialist, and Leadership Career Paths
- Which employees create the most value through breadth, deep expertise, or leadership rather than through the same progression path?
- Where does the current career architecture force specialists into management simply to advance?
- How should AI-enabled generalist roles progress without becoming shallow collections of unrelated responsibilities?
- What levels of specialist depth should carry status and compensation comparable to management?
- How should movement between generalist, specialist, and leadership tracks remain possible?
- What common capabilities should all three paths develop as GenAI becomes embedded in work?
- How can the organization prevent one career path from becoming the default prestige route while others become dead ends?
Preserving Expertise, Learning, and Succession
- Which routine activities historically helped experts maintain pattern recognition and technical fluency?
- What happens to expertise when those activities are increasingly delegated to GenAI?
- Which kinds of direct practice should experts continue even when manual execution is no longer economically necessary?
- How can experts remain exposed to enough ordinary cases to recognize emerging changes in the field?
- What difficult work should become the primary focus of expert practice?
- How should the organization distinguish unnecessary manual work from practice that preserves critical competence?
- What evidence would reveal that expert capability is degrading despite strong AI-supported performance?
Capturing Tacit Knowledge Before Experienced Employees Leave
- Which critical knowledge exists mainly in the experience of a small number of people?
- What decisions, exceptions, warning signs, and contextual distinctions do these employees know that formal documentation does not capture?
- How can GenAI help elicit and organize this knowledge without pretending that all tacit judgment can be fully codified?
- Which cases should experienced employees explain in detail before transition or retirement?
- Who should validate whether captured knowledge is accurate and usable by others?
- How should tacit knowledge be connected to real workflows rather than stored as passive documentation?
- What knowledge will still require direct apprenticeship or shared experience even after extensive capture?
Creating Deliberate Practice When Real Work Provides Fewer Repetitions
- Which skills require repeated practice even though AI now handles many of the relevant real cases?
- What kinds of exercises can reproduce the judgment required in authentic work?
- How difficult should practice become as capability develops?
- What feedback must learners receive to understand why their reasoning succeeded or failed?
- How often must deliberate practice occur to maintain competence rather than only build it initially?
- Which skills cannot be maintained credibly through simulation alone?
- How should deliberate practice be treated as legitimate work rather than optional training performed around productive demands?
Building Successors for Highly Leveraged AI-Enabled Experts
- Which experts now support disproportionately large amounts of work because GenAI amplifies their reach?
- What would happen if one of those experts became unavailable?
- Which elements of their judgment, relationships, and system knowledge must successors develop?
- How can potential successors receive enough real responsibility when the incumbent expert can perform so much with AI?
- Should more than one successor be developed for particularly critical expertise?
- What work should experts deliberately delegate for developmental purposes even if doing it themselves would be faster?
- How will the organization know when a successor is genuinely capable rather than merely proficient at using the same AI tools?
Maintaining Selected Skills Through Periodic Unaided Practice
- Which skills must humans retain independently even if AI usually performs or assists the work?
- How often do those skills need direct practice to remain reliable?
- What kinds of unaided exercises best reveal whether competence still exists?
- Should unaided practice focus on routine fundamentals, difficult exceptions, or both?
- How should performance be assessed without turning the exercise into ceremonial compliance?
- What productivity cost is justified to preserve strategically important independent competence?
- When can a skill safely stop being maintained internally because the organization no longer needs credible human fallback?
Using Simulation to Develop Judgment for Rare or Difficult Cases
- Which high-value situations occur too rarely for normal work to provide enough learning opportunities?
- What realistic scenarios would expose learners to the relevant ambiguity and tradeoffs?
- How can simulations include incomplete information, conflicting goals, and plausible AI errors?
- Who should debrief the reasoning rather than merely score the final decision?
- How should scenarios evolve as organizational systems and GenAI capabilities change?
- Which decisions are too context-dependent to simulate adequately?
- How can simulation performance inform readiness for greater real-world responsibility?
Preserving Professional Communities After Expertise Becomes More Distributed
- Which specialist communities currently learn through proximity to colleagues performing similar work?
- What happens to professional development when specialists become embedded across many teams or business units?
- Which shared forums, standards, case reviews, or communities should connect distributed experts?
- Who should own the health of the professional discipline when no single function employs all specialists?
- How can distributed experts contribute lessons back to the wider community?
- What career and mentoring mechanisms are necessary so isolated specialists continue to develop?
- When does distributed expertise become too fragmented to sustain consistent professional capability?
Transferring Expertise When Critical Work Moves Between Roles or Functions
- What expertise must move with the work rather than remain in the function that historically performed it?
- Which knowledge can be documented and which requires direct mentoring or joint work?
- How long should old and new owners overlap before responsibility transfers fully?
- What cases should the receiving team handle under supervision before independent ownership?
- Which expert responsibilities should remain with the original function even after routine work moves?
- How will the organization detect capability gaps after the transfer?
- What should happen if the receiving role proves unable to absorb the expertise at the level originally assumed?
Preventing AI Assistance from Replacing Human Mentorship
- Which developmental needs are employees increasingly directing to AI instead of experienced colleagues?
- What can AI coaching provide effectively, and what does it lack compared with human mentorship?
- Which forms of feedback require knowledge of the employee, organization, and professional context?
- Are senior employees spending less time teaching because AI can answer routine questions?
- How should managers and experts remain accountable for developing others even when AI support is abundant?
- Could AI reduce low-value mentoring tasks and create more time for deeper human guidance?
- What signs would indicate that employees are receiving information but not the professional socialization and judgment development mentorship provides?
Maintaining Independent Human Competence as AI Takes Over Routine Execution
- Which underlying capabilities must humans retain to supervise, challenge, or recover AI-enabled work?
- Are employees still learning how the work actually functions beneath the AI-supported interface?
- What tasks should humans periodically perform directly to preserve independent understanding?
- Which roles require enough technical or domain depth to diagnose failures the AI cannot explain reliably?
- How should independent competence be tested rather than assumed from successful AI-assisted performance?
- What happens if the organization can operate normally only while its AI systems remain available and correct?
- Which minimum human capability should be treated as a resilience requirement rather than an optional development goal?
Allocating Productivity Gains and Released Capacity
Distinguishing Time Savings from Usable Organizational Capacity
- How much time does GenAI appear to save at the individual task level?
- What new review, integration, coordination, or exception work offsets those savings elsewhere?
- Does the end-to-end process actually complete more outcomes with the same resources?
- Are saved hours concentrated in fragments too small or unpredictable to redeploy meaningfully?
- What happens to the time employees report saving in practice?
- Are productivity gains stable across normal users, workloads, and operating conditions?
- What evidence would justify treating measured time savings as durable organizational capacity?
Using Released Capacity to Increase Output
- Is there valuable unmet demand that existing capacity previously prevented the organization from serving?
- Which outputs can increase without overwhelming downstream review, decision, or implementation capacity?
- Does producing more of this work create proportional customer or organizational value?
- Which bottleneck will constrain output once the original production constraint is relaxed?
- What additional resources outside the AI-enabled team will higher output require?
- How should targets change so growth reflects real demand rather than production for its own sake?
- What evidence would show that using released capacity for volume creates more value than alternative uses?
Using Released Capacity to Improve Quality
- Which parts of the work would benefit most from additional human attention once routine production becomes cheaper?
- Could released capacity support better verification, deeper analysis, stronger customer understanding, or more thorough testing?
- How should the organization define the quality improvement it expects?
- Which quality dimensions matter enough to justify retaining capacity rather than removing it?
- Are employees actually using freed time for higher-quality work, or simply producing more?
- How can quality gains be measured when they involve avoided errors or better judgment rather than visible output?
- At what point would further quality investment produce less value than another use of the capacity?
Using Released Capacity to Shorten Cycle Times
- Which delays in the end-to-end process are caused by genuine capacity constraints?
- Does GenAI remove work from the critical path or only accelerate tasks that were not limiting completion time?
- Which downstream approvals or dependencies would become the next source of delay?
- What customer or business value would faster completion create?
- How much spare capacity must be retained to achieve reliably shorter response times?
- Could shorter cycle expectations create unhealthy work intensification despite AI assistance?
- What end-to-end measure would confirm that released capacity is genuinely increasing speed?
Using Released Capacity to Reduce Backlogs and Unmet Demand
- What valuable work is currently delayed or abandoned because capacity is insufficient?
- Which backlog items remain worth doing now that GenAI lowers their processing cost?
- How much released capacity should be directed toward existing backlog before normal staffing assumptions change?
- Will clearing the backlog reveal recurring demand that requires permanent capacity?
- Which old backlog items should be retired rather than processed simply because capacity is available?
- What secondary bottlenecks will emerge as delayed work begins moving again?
- How will the organization distinguish a one-time backlog reduction from a sustained productivity improvement?
Preserving Deliberate Slack for Resilience and Unexpected Work
- Which teams currently operate so close to full utilization that unexpected work causes immediate disruption?
- How much released capacity should remain uncommitted rather than be converted into higher targets?
- What kinds of disruptions, incidents, demand spikes, or opportunities justify maintaining slack?
- Where does slack support experimentation, learning, or process improvement in addition to resilience?
- How can useful slack be distinguished from structurally unnecessary excess capacity?
- What performance system would prevent managers from automatically filling all freed capacity?
- What evidence would justify preserving a capacity buffer even when short-term cost reduction is possible?
Converting Durable Excess Capacity into Staffing Reduction
- What evidence shows that the workload reduction is persistent rather than temporary?
- Have review, exception, coordination, and demand effects already stabilized?
- Does the function still need current staffing for resilience, peaks, learning, or future capability?
- Could excess capacity create more value through redeployment than through removal?
- Which roles genuinely become unnecessary rather than merely lighter?
- What knowledge or responsibilities must be transferred before positions are removed?
- How should the organization monitor whether staffing reduction creates hidden workload or service deterioration elsewhere?
- What work has historically been omitted because the required human effort exceeded its expected value?
- Which of that work becomes worthwhile once GenAI lowers production cost?
- Does the new work improve decisions, customers, quality, resilience, innovation, or another meaningful outcome?
- What human review or integration does the newly feasible work still require?
- Could large amounts of newly cheap work overwhelm attention or decision capacity?
- Who should decide which previously uneconomic activities now deserve organizational resources?
- How should the organization prevent low-cost production from turning into unlimited low-value activity?
Responding When Lower Costs Create Additional Demand
- Has GenAI made a service or capability inexpensive enough that internal or external demand is growing?
- How much of the apparent productivity gain will be consumed by the additional demand?
- Does serving more demand create sufficient value to justify preserving or expanding capacity?
- Which parts of the system will become constrained as volume grows?
- Should the organization ration, prioritize, price, or automate additional demand differently?
- Could staffing cuts based on old demand levels create shortages once the service becomes easier to use?
- What demand assumptions should be built into future capacity planning?
Reallocating Capacity After the Bottleneck Moves Elsewhere
- Which activity became less constrained because of GenAI?
- Where has waiting, workload, or scarcity increased as a result?
- Which skills are needed at the new bottleneck?
- Can employees from the relieved area be redeployed toward the new constraint?
- Would adding more capacity at the old activity create little additional throughput?
- What organizational boundary or incentive prevents capacity from moving toward the new bottleneck?
- How should workforce and budget planning change once the true system constraint is identified?
Reconsidering Organizational Scale, Sourcing, and Firm Boundaries
Bringing a Capability In-House After GenAI Reduces Its Minimum Viable Scale
- Which capability was historically outsourced because maintaining sufficient internal capacity was too expensive?
- How much does GenAI reduce the staffing or scale required to perform the work internally?
- What proprietary data, context, or strategic advantage would improve if the capability moved in-house?
- What specialist expertise would still need to be retained or purchased externally?
- What full costs would internalization create beyond direct labor?
- Would in-house ownership improve learning, responsiveness, resilience, or control enough to justify the change?
- What evidence would show that the capability is now economically and operationally viable internally?
Reconsidering Outsourcing Built Primarily on Labor Cost Advantages
- Which outsourced services derive most of their economics from lower-cost human labor?
- How much of that labor can now be automated or augmented through GenAI?
- Does the external provider retain meaningful scale, expertise, infrastructure, or risk advantages beyond labor arbitrage?
- Could the organization now perform the work internally with a smaller AI-enabled team?
- How are provider pricing and service models changing as their own labor requirements fall?
- Which knowledge or strategic capability is currently being lost through outsourcing?
- Should the relationship be redesigned around outcomes or specialist capability rather than volumes of external labor?
Buying Agentic Services Instead of Building Internal Delivery Capacity
- Which end-to-end activities can now be purchased as agentic services rather than staffed internally?
- How much organizational context must an external agentic service access to perform effectively?
- Who remains accountable internally for the results?
- What visibility will the organization have into decisions and actions taken by the service?
- How easily can the service be replaced if quality, cost, security, or strategic needs change?
- What internal expertise must remain so the organization can govern and challenge the external capability?
- When would buying the service create better economics than building and operating the capability internally?
Insourcing Strategically Critical Expertise Despite External Availability
- Which expertise is sufficiently important that the organization must understand and control it internally?
- Does the capability contribute directly to differentiation, critical decisions, resilience, or regulatory responsibility?
- What risks arise if the organization relies entirely on vendors or consultants for this expertise?
- How much internal capacity is necessary to remain an intelligent buyer and credible challenger?
- Can external specialists still supplement internal capability for rare or highly advanced needs?
- What investment is required to rebuild expertise that has previously been outsourced?
- How should the organization decide when strategic ownership matters more than short-term sourcing efficiency?
Reassessing Contractor and Contingent Workforce Models as Routine Cognitive Work Shrinks
- Which contractor roles exist primarily to provide flexible volumes of routine cognitive production?
- How much of that demand is likely to decline because of GenAI?
- Which contingent roles remain valuable because they provide rare expertise, surge capacity, or independent perspective?
- Could smaller internal teams absorb work previously assigned to large contractor populations?
- How should variable demand be covered if routine contractor capacity is reduced?
- What knowledge should be transferred before external contingent roles disappear?
- How should the organization redesign its mix of employees, contractors, agents, and specialist vendors as the work changes?
Reassessing the Role of Consultants in AI-Enabled Knowledge Work
- Which consulting work can internal teams now perform more effectively with GenAI support?
- Where do consultants still provide scarce expertise, independence, external perspective, or implementation capability?
- Are consultants being used mainly to supply analytical production that has become cheaper internally?
- How should consulting engagements shift if the organization can produce research, analysis, and drafts itself?
- What internal capabilities are necessary to evaluate AI-enabled consultant outputs critically?
- Could smaller specialist engagements replace larger teams built around junior analytical labor?
- Which consulting relationships should remain because external challenge creates value independent of production capacity?
Choosing Whether to Build, Buy, or Partner for a GenAI-Enabled Capability
- How strategically differentiating is the capability for the organization?
- How much proprietary context, data, or workflow integration does it require?
- What internal expertise exists to build and operate it credibly?
- How mature, substitutable, and economically attractive are external offerings?
- Which risks increase under vendor dependence versus internal ownership?
- Could a partnership combine external technical capability with internal domain ownership more effectively than either extreme?
- What total lifecycle economics should determine the choice rather than initial development cost alone?
Using External Specialists for Expertise Needed Intermittently Rather Than Continuously
- Which specialist capabilities are important but required too infrequently to justify permanent internal staffing?
- How quickly must external expertise be available when an exceptional case arises?
- What internal capability is needed to recognize when external specialist involvement is necessary?
- How much organizational context must specialists understand before their advice becomes useful?
- Could a standing relationship improve continuity compared with ad hoc sourcing?
- Which expertise remains too strategically important to rely on externally even if demand is intermittent?
- How should the organization compare the economics of selective external access with maintaining underutilized internal specialists?
Reassessing the Minimum Viable Size of a Function
- Which fixed responsibilities determine the minimum human size of the function regardless of workload?
- How much routine production can GenAI absorb?
- Which review, resilience, separation-of-duties, development, or coverage requirements prevent further reduction?
- Could one smaller function serve several units that previously needed separate teams?
- Does reducing the function create unacceptable key-person dependency?
- How much demand variability must the function still absorb?
- What evidence would establish a smaller sustainable size rather than a temporarily understaffed one?
Redrawing the Boundary Between the Organization and Its External Ecosystem
- Which activities that were previously internal can now be sourced more effectively through AI-enabled external providers?
- Which previously outsourced capabilities can now be owned internally at viable scale?
- Where does strategic value come from controlling the capability rather than merely receiving the output?
- Which ecosystem relationships create access to innovation or expertise the organization should not reproduce internally?
- What dependencies would become dangerous if too much critical capability moved outside the firm?
- How should partnerships, vendors, contractors, and AI services fit together around retained internal ownership?
- What should the organization's future boundary reflect about where its distinctive knowledge, authority, and accountability need to remain?
Replacing Output Volume as the Primary Measure of Knowledge Work
- Which current performance measures primarily reward the amount of work produced?
- How easily can GenAI increase those measures without increasing organizational value?
- What outcome measures better capture whether the work solved the intended problem?
- How should quality, judgment, integration, and downstream effects be represented?
- Which roles still have meaningful volume measures that should remain?
- How can the organization avoid replacing one simplistic output metric with another equally narrow outcome proxy?
- What performance evidence should matter more as cognitive production becomes increasingly abundant?
- What does the organization actually want employees to achieve through GenAI rather than simply use it?
- Could high AI usage reflect unnecessary automation, weak independent capability, or low-value activity?
- Which roles legitimately need very different levels of AI use?
- How should appropriate non-use be recognized when human work is better for a particular case?
- What outcome or quality measures should replace adoption counts in performance evaluation?
- How can managers distinguish sophisticated AI use from frequent AI use?
- What incentives would encourage employees to choose the right method rather than maximize visible AI activity?
Rewarding People for Sharing Reusable AI-Enabled Practices
- Which employee-created workflows, agents, methods, or lessons could create value beyond the originating role?
- What currently discourages people from sharing effective practices?
- How should contributions to shared capability be recognized in performance evaluation?
- Should employees receive credit when others reuse or improve what they created?
- How can the organization distinguish genuinely reusable practices from local tricks that do not generalize?
- What mechanism should turn useful local learning into accessible organizational capability?
- How can sharing be rewarded without creating incentives to publish large amounts of low-quality material?
Preventing Productivity Gains from Automatically Becoming Higher Workload Targets
- How is the organization currently responding when employees report significant time savings from GenAI?
- Are saved hours automatically converted into higher output expectations?
- What alternative uses of released capacity could create greater value?
- Could ever-rising targets discourage employees from revealing or sharing productivity improvements?
- How should expectations change only after net capacity gains have been demonstrated?
- What level of slack, development, or improvement time should be preserved deliberately?
- How can managers prevent GenAI from becoming a mechanism for permanent work intensification?
- How much employee effort is now spent checking, correcting, testing, or validating AI-enabled work?
- Do current metrics recognize that work as valuable contribution?
- Which quality failures are being prevented through review but therefore remain invisible in output measures?
- How should difficult verification be distinguished from unnecessary rework caused by poor AI use?
- Should reviewers receive credit for stopping bad work even when doing so reduces apparent throughput?
- What measures would reveal whether review effort is improving system quality?
- How can the organization avoid underfunding verification simply because production remains more visible?
Balancing Individual Contribution with Team and System Outcomes
- Which outcomes depend so heavily on shared AI, data, platforms, and colleagues that individual attribution is inherently incomplete?
- What individual contributions still deserve direct recognition?
- How should team performance influence evaluation without allowing weak individual contribution to disappear?
- How should people who improve shared systems receive credit even when their personal output declines?
- Which behaviors support collective performance but are currently invisible in individual metrics?
- How can performance systems discourage people from optimizing local results at the expense of end-to-end outcomes?
- What balance between individual and collective measures best fits each type of work?
Rewarding Disciplined Experimentation Without Rewarding Careless Failure
- Which GenAI-related experiments are valuable even if they do not produce an immediate successful outcome?
- What should distinguish a well-designed failed experiment from poor execution?
- Were the hypothesis, success criteria, risk limits, and learning objectives clear beforehand?
- Did the experiment produce evidence that changed subsequent decisions or practices?
- How should employees receive recognition for stopping an unpromising experiment early?
- What would prevent "experimentation" from becoming an excuse for weak accountability?
- How can performance systems encourage intelligent exploration while still rewarding reliable operational delivery?
Removing Incentives to Preserve Manual Work for Budget or Headcount Protection
- Do managers lose budget, status, positions, or influence when they automate work successfully?
- Are employees disadvantaged when they share methods that reduce the need for their current activities?
- Which organizational incentives make maintaining inefficient manual work rational?
- How should leaders be rewarded for eliminating unnecessary work rather than defending organizational territory?
- Could released capacity be redeployed in ways that reduce fear of sharing productivity gains?
- How should budgets recognize value creation rather than simply the size of the workforce controlled?
- What structural incentive changes are needed before the organization can expect people to redesign their own work willingly?
Preventing Metrics from Rewarding Excessive or Unsafe Automation
- Which targets could encourage teams to automate more work regardless of suitability?
- Are employees or managers being rewarded for automation rates, headcount reduction, or AI adoption without quality adjustment?
- What risk, error, customer, or review measures should counterbalance efficiency metrics?
- How should the organization recognize a decision not to automate when human involvement remains important?
- Could teams improve their local metrics by transferring risk or workload downstream?
- What measures would reveal whether automation is increasing hidden failure or fragility?
- How should incentive structures change if teams begin optimizing for the appearance of AI transformation rather than useful outcomes?
- What end-to-end outcomes matter more than the activity of any individual function?
- Which upstream and downstream effects should be included when assessing performance?
- How can cycle time, quality, cost, customer outcomes, resilience, and learning be considered together?
- Which local metrics should remain because they reveal important control or capability conditions?
- What system-health measures would expose hidden review queues, overload, dependency, or capability erosion?
- How should teams share accountability for outcomes they jointly influence?
- What set of measures would reveal whether GenAI-enabled redesign is improving the organization rather than merely increasing visible activity?
Designing Organizational Resilience Around Critical AI Dependencies
Managing Dependence on a Single Model or AI Provider
- Which critical workflows currently depend on the same model, provider, or underlying service?
- What would stop or degrade if that provider became unavailable, changed terms, raised prices, or materially changed capability?
- How difficult would it be to move each critical workload to an alternative provider?
- Which proprietary integrations, data structures, evaluations, or workflows increase switching difficulty?
- Where is concentration acceptable because alternatives would add more cost and complexity than resilience?
- What contractual, technical, and organizational capabilities should be preserved to keep substitution realistic?
- At what level of dependency should diversification become an explicit organizational requirement?
- Which business processes would be affected simultaneously if the shared platform became unavailable?
- How long can each dependent process operate without the platform before consequences become unacceptable?
- Which critical activities need alternative routes rather than simply waiting for recovery?
- Who owns continuity planning for business processes that depend on centrally operated AI infrastructure?
- How should local teams know when to switch from normal operation to a fallback mode?
- What recovery priorities should determine which capabilities return first after a major platform failure?
- How often should the organization test whether critical work can continue when the shared platform is unavailable?
Reducing Dependence on a Small Number of Critical Specialists
- Which AI-enabled systems or workflows depend on knowledge held by only one or two people?
- What would the organization be unable to operate, diagnose, or recover if those specialists became unavailable?
- Which knowledge can be documented, automated, or transferred without pretending that all tacit expertise is codifiable?
- Who should be developed as credible backup for each critical specialist capability?
- Which responsibilities can be distributed so one specialist is no longer the mandatory point of intervention?
- How should workload be redesigned so specialists have time to transfer knowledge rather than remaining permanently overloaded?
- What evidence would show that the organization has moved from nominal backup to genuine capability redundancy?
Restoring Human or Non-AI Fallbacks for Critical Work
- Which critical workflows have become so AI-dependent that no credible alternative operating method remains?
- What minimum service must continue if GenAI becomes unavailable or inappropriate for use?
- Which parts of the old human or conventional process should be preserved as fallback capability?
- How quickly could employees return to the fallback method after a prolonged period of AI-enabled work?
- What skills, documentation, access, and staffing must remain available for the fallback to be real?
- How should fallback operation be tested without imposing unnecessary duplicate work during normal conditions?
- When is maintaining a fallback no longer worth its cost because another resilient alternative is stronger?
Preserving Capability When Employees Can No Longer Work Without AI
- Which roles have become dependent on AI for activities employees can no longer perform or understand independently?
- Does that dependency matter because the underlying capability is critical to supervision, recovery, judgment, or professional responsibility?
- Which skills should employees retain even if AI performs them during normal operation?
- How can independent capability be tested without requiring constant manual duplication of automated work?
- Are new employees learning the underlying work or only learning how to operate AI-supported interfaces?
- Which roles can reasonably become fully AI-dependent because independent human execution no longer creates material value?
- What minimum human competence should the organization deliberately preserve for each critical AI-enabled activity?
Preventing Central Review Functions from Becoming Single Points of Failure
- Which workflows depend on the same central review or assurance function before they can proceed?
- What happens across the organization if that function becomes overloaded or unavailable?
- Which review decisions can be distributed without weakening independence or standards?
- What low-risk cases could move to automated checks, sampling, or predefined local authority?
- Which specialist judgments genuinely require the central function to remain involved?
- How should backup capacity or alternate reviewers be organized for high-consequence work?
- What indicators would reveal that central assurance has shifted from a control mechanism into a resilience risk?
Maintaining Viable Alternative Providers or Technical Paths
- Which critical capabilities need at least one credible alternative provider or technical route?
- What would make an alternative genuinely usable rather than merely listed in a contingency plan?
- How much compatibility should be preserved in data, interfaces, workflows, and evaluation methods?
- Which capabilities are too expensive to duplicate continuously but still require tested migration plans?
- How often should alternatives be exercised sufficiently to confirm they remain viable?
- Who owns the organizational knowledge required to switch paths under pressure?
- What level of additional cost is justified to preserve meaningful optionality for critical capabilities?
Designing Reduced-Function Operating Modes for AI Disruption
- What minimum outcomes must the organization still deliver during a significant AI disruption?
- Which nonessential features, services, or process steps can be suspended temporarily?
- Which decisions can revert to simpler rules or conventional systems during degraded operation?
- What staffing pattern is needed to support reduced-function operation?
- Who has authority to declare, modify, and end the reduced-function mode?
- How should customers, employees, and dependent teams understand the temporary change in service?
- What exercises would reveal whether reduced-function operation is actually sustainable for the required period?
Preparing for Common-Mode Failures Across Many Agents or Workflows
- Which apparently independent agents or workflows rely on the same models, platforms, data sources, permissions, or technical components?
- What single failure could affect many of them simultaneously?
- Are monitoring systems capable of recognizing a correlated failure rather than treating each incident separately?
- Which shared dependency should be diversified, isolated, or constrained because its failure radius is too large?
- How can one faulty agent, prompt, policy, or shared component be prevented from propagating problems across other workflows?
- Who has authority to disable an entire class of agents when a common vulnerability is discovered?
- How should recovery be prioritized when many AI-enabled processes fail for the same underlying reason?
Balancing Efficiency Gains Against Concentration and Resilience Risks
- Which consolidation or automation decisions create the largest efficiency gains?
- What technical, organizational, supplier, or expertise concentration does each gain introduce?
- How much resilience is lost when duplicate systems, teams, providers, or capabilities are removed?
- Which redundancies are genuinely wasteful and which provide economically valuable protection?
- How severe would a failure need to be before the avoided resilience investment no longer looks worthwhile?
- Can the organization achieve most of the efficiency benefit while preserving selective redundancy at critical points?
- What explicit resilience threshold should proposed efficiency improvements be required to satisfy?
Testing, Transitioning, and Reversing Structural Change
Piloting a Structural Change Before Enterprise-Wide Adoption
- What specific organizational hypothesis is the pilot intended to test?
- Which team, function, or workflow provides a sufficiently representative but bounded test environment?
- What should remain unchanged so the structural effect can be distinguished from unrelated changes?
- Which outcome, workload, quality, coordination, and capability measures should be captured before and during the pilot?
- How long must the pilot run to expose ordinary operations, difficult cases, and adaptation effects?
- What conditions would justify expanding, modifying, or stopping the experiment?
- Which aspects of the pilot would not generalize automatically to the rest of the organization?
Running Old and New Organizational Arrangements in Parallel
- Which activities are important enough to justify temporary parallel operation?
- What should the organization learn from comparing the old and new arrangements?
- How long can duplicate structures operate before their cost and confusion outweigh the learning benefit?
- Which arrangement owns the final outcome while both are active?
- How should conflicting results or recommendations from the two arrangements be resolved?
- What evidence should determine when the old structure can be retired safely?
- How can parallel operation avoid becoming permanent duplication because nobody wants to make the final transition decision?
Using Temporary Roles During an Organizational Transition
- Which transition responsibilities are necessary only while old and new structures coexist?
- What temporary roles are needed to migrate knowledge, coordinate dependencies, or stabilize new workflows?
- Which authority should temporary role holders possess, and where should their authority end?
- How can temporary roles avoid duplicating responsibilities that permanent owners should already hold?
- What knowledge or relationships must transfer before each temporary role disappears?
- What explicit sunset date or completion condition should be attached to the role?
- How will the organization prevent successful temporary coordinators from becoming permanent intermediaries without a lasting structural purpose?
Defining Reversal Criteria Before a Structural Experiment Begins
- What specific evidence would indicate that the redesign's underlying hypothesis is wrong?
- Which deterioration in quality, risk, workload, customer outcomes, or capability should trigger reversal?
- What thresholds should lead to modification rather than complete reversal?
- Which consequences would be too serious to wait for the experiment's scheduled end?
- Who has authority to reverse the structural change when the criteria are met?
- What baseline must be preserved so reversal remains practically possible?
- How can precommitted reversal criteria reduce the tendency to defend a visible redesign after the evidence turns against it?
Sequencing Permanent Changes After Capabilities and Workflows Stabilize
- Which structural decisions depend on GenAI capabilities that are still changing rapidly?
- Which workflows need to demonstrate stable performance before permanent roles or reporting lines change?
- What temporary arrangements can preserve flexibility while evidence accumulates?
- Which changes are easy to reverse and can therefore occur earlier?
- Which irreversible or capability-destroying changes should wait for stronger evidence?
- What dependencies require one redesign step to happen before another can succeed?
- What evidence would show that the organization has learned enough to move from experimentation into permanent structure?
Preserving Institutional Knowledge Before Removing a Role or Unit
- What decisions, relationships, historical context, and exception knowledge reside in the role or unit being removed?
- Which of that knowledge exists only in people's experience rather than formal systems?
- What recurring responsibilities might become invisible once the organizational unit disappears?
- Who should inherit each important body of knowledge or relationship?
- Which difficult cases should be reviewed with successor owners before the transition is complete?
- What knowledge should be captured in systems, and what still requires direct handover or mentoring?
- How will the organization detect several months later whether important institutional knowledge was lost?
Tracking Hidden Work After Responsibilities Are Removed or Moved
- Which responsibilities were expected to disappear under the redesign?
- Where are employees actually spending time compensating for those removed responsibilities?
- Has coordination, checking, relationship management, or exception handling become informal and therefore harder to see?
- Which teams have absorbed new workload without corresponding staffing, authority, or recognition?
- Are automated workflows generating support work that was not included in the redesign assumptions?
- What measures or observations can reveal hidden work before it becomes chronic overload?
- When hidden work appears, should it be eliminated, formally assigned, automated, or evidence used to reverse part of the redesign?
Revising or Reversing a Redesign That Does Not Produce the Expected Results
- Which expected outcomes failed to materialize?
- Is the problem caused by the structural hypothesis itself or by poor implementation of an otherwise sound design?
- What unexpected dependencies, bottlenecks, or capability losses have emerged?
- Which parts of the redesign are still working and should be preserved?
- Can the problem be corrected through a narrower adjustment rather than restoring the entire previous structure?
- What costs and risks would reversal create now that people and work have adapted to the new arrangement?
- What should the organization learn from the failed redesign before attempting another structural change?
Sunsetting Temporary Structures After Their Transition Purpose Ends
- What original transition problem justified the temporary structure?
- Has that problem now been resolved or transferred to a permanent owner?
- Which responsibilities should disappear entirely rather than move elsewhere?
- Which valuable capabilities developed inside the temporary structure need a permanent home?
- Are stakeholders still routing work to the temporary structure out of habit rather than necessity?
- What would happen operationally if the structure closed on the planned date?
- How should closure be completed so the organization does not retain unnecessary layers after the transition is finished?
Maintaining Clear Accountability While Old and New Structures Coexist
- Who owns each important outcome during the transition period?
- Which responsibilities remain with the old structure and which have already moved to the new one?
- Where could dual ownership create conflicting instructions or allow accountability to fall between structures?
- Who has final decision authority when old and new roles disagree?
- How should incidents or failures be attributed while responsibilities are still moving?
- What communication does each participant need to understand the current rather than eventual accountability model?
- What transition milestone should trigger the final transfer of accountability to the new structure?