Investigations 01–19

Question

What happens when increasingly capable AI enters organizations whose real work is distributed across people, software, rules, evidence, authority, and physical reality?

Surviving conclusion

Model capability does not determine organizational consequence; local capability enters a gated work system.

Map change

Established the initial gated adaptive work-system representation.

Relationship to v2.1

Foundation retained, but the earlier binding-constraint framing was later narrowed by v2.1.

Question

When does an AI-induced local task improvement become an end-to-end workflow improvement?

Surviving conclusion

Workflow propagation is a conversion problem: capacity released is not capacity converted.

Map change

Moved the unit of analysis toward reliable completed outcomes and propagation through workflow dependencies.

Relationship to v2.1

Retained. v2.1 adds that the workflow graph itself may change before propagation is evaluated.

Question

What determines which cases remain human and how those cases are routed, escalated, and resolved?

Surviving conclusion

Human residual work is produced by routing, risk, state, verification, authority, and workflow design, not simply by model incapability.

Map change

Made exceptions, escalation, and residual human roles explicit.

Relationship to v2.1

Retained and now interpreted as one consequence of organizational configuration.

Question

How are people able to judge, supervise, correct, and recover from AI systems when AI changes the work through which those abilities were historically learned?

Surviving conclusion

Assisted performance and durable human capability are different outcomes.

Map change

Introduced experience topology and capability reproduction as dynamic system variables.

Relationship to v2.1

Historically important. Investigation 19 replaces scalar expertise with a capability-stock vector.

Question

What must AI know about current organizational state when records, memory, evidence, and reality can diverge?

Surviving conclusion

Enterprise AI needs current adjudicated state, not merely retrieval or memory.

Map change

Separated record, evidence, memory, current action state, provenance, and reality closure.

Relationship to v2.1

Retained. v2.1 adds that state quality may itself be behaviorally endogenous.

Question

Who or what may convert a plausible AI action into a binding organizational commitment?

Surviving conclusion

Capability, technical permission, organizational authority, and accountability are distinct.

Map change

Added commitment boundaries, bounded delegation, reversibility, and authority envelopes.

Relationship to v2.1

Retained. v2.1 further decomposes formal authority, practical control, informational control, accountability burden, economic upside, and discretion value.

Question

How does AI change teams, management functions, coordination costs, and decision rights?

Surviving conclusion

AI redistributes management functions rather than simply eliminating managers.

Map change

Separated information routing from coordination, judgment, authority, coaching, and adaptation.

Relationship to v2.1

Retained as part of organizational configuration and workflow design.

Question

Who captures AI-created value, and which complementary assets determine durable economic ownership?

Surviving conclusion

Value created and value captured are different; model access alone does not determine economic ownership.

Map change

Added value capture, bargaining, complementary assets, and firm-boundary choices.

Relationship to v2.1

Retained. v2.1 additionally treats expected value and burden distribution as potentially upstream of actor response.

Question

Which competitive advantages persist as AI diffuses, and how do entry, demand, prices, and concentration respond?

Surviving conclusion

AI can lower some barriers while increasing returns to scarce complements; market effects differ by layer and diffusion stage.

Map change

Added entry, demand response, pass-through, and advantage migration.

Relationship to v2.1

Retained as a downstream and feedback layer rather than a universal direction of concentration.

Question

How does AI change task bundles, hiring flows, entry paths, seniority, wages, and occupational structure?

Surviving conclusion

Labor adjustment can appear in task and hiring flows before aggregate employment stocks move.

Map change

Replaced simple job-loss framing with task, vacancy, hiring, career-path, and workforce recomposition.

Relationship to v2.1

Retained with strict timing and causal-attribution limits.

Question

Why can the same AI capability produce different consequences across countries, firms, sectors, and institutional settings?

Surviving conclusion

AI capability travels more easily than organizational consequence.

Map change

Made wages, firm size, language, informality, institutions, and physical infrastructure explicit context.

Relationship to v2.1

Retained; Malaysia and Southeast Asia remain a local empirical frontier rather than assumed portability.

Question

Why do some AI experiments become durable operating infrastructure while others remain tools, pilots, or abandoned projects?

Surviving conclusion

Adoption is not access, and institutionalization is not adoption.

Map change

Added adoption depth, ownership, hidden implementation work, dependency, and institutionalization.

Relationship to v2.1

Retained and reinterpreted as actor response to expected value, burden, risk, and control.

Question

How can organizations know whether AI-enabled systems actually work and whether AI caused the observed outcome?

Surviving conclusion

Monitoring, observability, evaluation, causal attribution, and organizational learning are distinct functions.

Map change

Added evidence levels from infrastructure through reliable outcome, organization, economics, and distribution.

Relationship to v2.1

Retained and forms part of the v2.1 causal-validation contract.

Question

How do AI-enabled failures propagate, and what allows organizations to contain, recover, reconcile state, and learn?

Surviving conclusion

Resilience is not zero failure; failures must be bounded, observable, reversible where possible, and recoverable.

Map change

Added blast radius, cascade types, graceful degradation, and state recovery.

Relationship to v2.1

Retained as an organizational capability and a possible AI-induced dependency.

Question

What changes when AI must produce outcomes in a world constrained by matter, space, machines, people, safety, and time?

Surviving conclusion

Digital decision capacity can improve much faster than physical execution capacity.

Map change

Added reality closure, reality bandwidth, reality latency, and physical executability.

Relationship to v2.1

Retained as a distinct external and execution constraint class.

Question

What feedback loops, delays, thresholds, dependencies, and path dependence emerge as AI-induced changes accumulate over time?

Surviving conclusion

AI transformation is a dynamic adaptation process; the changed system changes what AI subsequently does.

Map change

Added stocks, feedback loops, transition debt, hysteresis, regimes, and dynamic adaptation.

Relationship to v2.1

Dynamic framing is retained, but the strong retrospective bottleneck-migration interpretation was narrowed in v2.1.