AIFC-040: AI Retrospective

Status: Draft 0.1 Standard: AI-First Community Standard Abbreviation: AIFC Builds on:

• AIFC-004 Feedback and Change Proposals
• AIFC-020 Human-Managed AI
• AIFC-024 Human Capability Reserve
• AIFC-030 AI Capacity Planning
• AIFC-031 AI Autonomy and Intensity
• AIFC-033 AI Budget and Cost Control
• AIFC-034 AI Lock-in and Exit Strategy

Purpose of this document: Define AI Retrospective as a recurring learning mechanism that helps a community evaluate whether AI is creating durable value, waste, risk, dependency, skill improvement, or governance debt.

1. Purpose of this document

This document defines AI Retrospective.

AI use should not only be deployed, consumed, and forgotten.

The community must learn from:

• what AI improved,
• what AI made worse,
• what it cost,
• what it changed in human capability,
• what it moved into or out of the source of truth,
• what dependency it created,
• what workflows should be converted,
• what skills should evolve,
• what governance must change.

AI Retrospective turns AI use into community learning.

2. Core principle

The core principle of this document is:

AI use must feed learning back into the community.

If AI helps once, the community gained output.

If the community learns from why it helped, what it cost, and how the work should change, the community gained capability.

AIFC therefore says:

Do not only use AI.
Retrospect AI use.
Convert learning into source of truth, workflows and skills.

3. Definition

AI Retrospective is a structured review of AI use and its impact on the community.

It evaluates:

• value created,
• cost consumed,
• AI waste,
• AI dependency,
• Human Capability Reserve,
• source-of-truth impact,
• workflow conversion candidates,
• skill evolution candidates,
• AI Autonomy and AI Intensity,
• AI-NDA Boundary,
• lock-in risk,
• budget and capacity,
• quality of human review,
• governance changes needed.

Minimum requirement

Significant AI use must periodically be reviewed through AI Retrospective or an equivalent learning mechanism.

4. Why AI Retrospective matters

Without AI Retrospective, AI use tends to become invisible habit.

The community may not notice:

• low-value AI work,
• repeated prompting,
• review overload,
• growing AI dependency,
• lost human capability,
• knowledge trapped in chat,
• skills that should be updated,
• workflows that should be converted,
• risks that should become change proposals.

Minimum requirement

The community must have a way to learn from AI use, not only consume AI output.

5. Relationship with ordinary retrospective

AI Retrospective may be part of a normal team retrospective, but it has a specific focus.

Ordinary retrospective asks:

How did our work go?

AI Retrospective asks:

How did AI affect our work, knowledge, capability, risk and governance?

Minimum requirement

If AI is significant in the work, retrospective must include AI-specific questions.

6. Retrospective triggers

Periodic triggers

AI Retrospective may happen:

• monthly,
• quarterly,
• after a sprint,
• after a release,
• after a mission,
• during governance review.

Event-based triggers

It may also be triggered by:

• AI incident,
• budget incident,
• AI-NDA Boundary issue,
• model or vendor change,
• repeated AI waste,
• review overload,
• AI dependency signal,
• significant workflow conversion candidate.

Minimum requirement

Critical AI use must have periodic or event-based retrospective triggers.

7. Retrospective scope

AI Retrospective may cover:

• one workflow,
• one AI team member,
• one AI engagement,
• one team,
• one product,
• one operating mode,
• one budget period,
• one community.

Scope should be explicit.

Minimum requirement

AI Retrospective must define what AI use it reviews.

8. Retrospective ownership

AI Retrospective must have an owner.

The owner is accountable for:

• preparing inputs,
• inviting participants,
• preserving psychological safety,
• recording outputs,
• turning findings into change proposals,
• ensuring follow-up,
• updating the source of truth where needed.

Minimum requirement

AI Retrospective without follow-up ownership is incomplete.

9. Participants

Participants may include:

• workflow owner,
• AI engagement owner,
• human reviewers,
• affected team members,
• security or data owner,
• budget owner,
• knowledge owner,
• AI team member owner,
• community governance.

Minimum requirement

AI Retrospective must include people who can assess both value and risk.

10. Inputs to AI Retrospective

Useful inputs include:

• AI outputs,
• accepted and rejected proposals,
• cost data,
• review effort,
• incidents,
• AI waste signals,
• source-of-truth changes,
• workflow metrics,
• user or customer feedback,
• dependency indicators,
• human capability signals.

Minimum requirement

AI Retrospective should be based on evidence, not only impressions.

11. What AI improved

The retrospective should ask where AI created value.

Examples:

• faster delivery,
• better analysis,
• clearer writing,
• reduced debt,
• detected risk,
• improved workflow,
• better source of truth,
• new skill,
• better decision support.

Minimum requirement

AI Retrospective must identify useful AI patterns worth preserving or converting.

12. What AI made worse

The retrospective should ask where AI created harm or friction.

Examples:

• low-quality output,
• review overload,
• confusion,
• inaccurate summaries,
• hidden assumptions,
• increased dependency,
• cost without value,
• knowledge outside the source of truth,
• weakened human learning.

Minimum requirement

AI Retrospective must identify negative AI effects, not only successes.

13. AI value

AI value should be assessed by durable outcome.

Useful questions:

• What was accepted?
• What changed?
• What risk was reduced?
• What capability improved?
• What knowledge entered the source of truth?
• What decision became better?

Minimum requirement

AI value must not be measured only by output volume.

14. AI cost

AI cost includes:

• money,
• tokens,
• compute,
• review time,
• attention,
• governance time,
• security work,
• dependency cost.

Minimum requirement

AI Retrospective must compare value with cost for significant AI use.

15. AI waste

AI Retrospective should detect AI waste.

Examples:

• repeated low-value prompts,
• outputs nobody uses,
• summaries nobody reads,
• review work without accepted value,
• AI-generated knowledge not maintained,
• agent work that creates cleanup burden.

Minimum requirement

AI waste findings must be recorded or triaged into the AI Waste Backlog.

16. AI dependency

AI Retrospective should detect whether the community is becoming AI-dependent.

Questions:

• What work stopped without AI?
• What can humans no longer do?
• What knowledge exists only in AI tools?
• What workflows have no fallback?

Minimum requirement

AI dependency signals must become observed signals or change proposals.

17. Human Capability Reserve

The retrospective should review whether AI strengthened or weakened human capability.

It should ask:

• Did humans learn?
• Did review quality improve?
• Did junior learning suffer?
• Is fallback still usable?
• Are human skills updated?

Minimum requirement

AI Retrospective must assess impact on Human Capability Reserve for critical work.

18. Source of truth impact

AI may create useful knowledge that remains outside the source of truth.

Retrospective should ask:

• What should be written back?
• What decision records are missing?
• What workflows or skills changed?
• What AI-generated knowledge should be rejected?

Minimum requirement

AI Retrospective must identify source-of-truth updates where AI created durable know-how.

19. Workflow conversion candidates

Repeated useful AI work may indicate a conversion candidate.

The community should ask:

• Is this repeated?
• Is the pattern stable?
• Would a template, validator, script, workflow, human skill, or AI skill reduce waste?

Minimum requirement

AI Retrospective must identify repeated AI work that should be evaluated for Workflow Conversion.

20. Skill evolution

AI Retrospective should identify skills that need to evolve.

Skill evolution candidates may come from:

• good outputs,
• repeated corrections,
• failures,
• human review findings,
• AI waste,
• workflow conversion.

Minimum requirement

AI Retrospective findings that affect capability must be considered for Skill Evolution.

21. Autonomy and intensity review

The retrospective should assess whether AI Intensity and AI Autonomy were appropriate.

It may recommend:

• increase,
• decrease,
• operating mode change,
• new approval boundary,
• de-escalation,
• stronger audit.

Minimum requirement

AI Retrospective must be able to recommend AI Intensity or AI Autonomy changes.

22. AI-NDA Boundary review

The retrospective should review whether data boundaries were respected.

Questions:

• Did AI see only approved data?
• Were prompts and outputs handled safely?
• Did derived knowledge need classification?
• Did memory create risk?

Minimum requirement

AI-NDA Boundary issues found in retrospective must be recorded and addressed.

23. Lock-in review

The retrospective should review lock-in risk.

Questions:

• Did new dependency appear?
• Is export possible?
• Is fallback available?
• Did human capability weaken?

Minimum requirement

AI Retrospective must flag significant AI lock-in risk.

24. Budget and capacity review

The retrospective should compare AI capacity use with value.

It should ask:

• Was budget used well?
• Was review capacity enough?
• Was attention overloaded?
• Should capacity be reallocated?

Minimum requirement

AI Retrospective must inform AI Capacity Planning and Budget Control.

25. Quality of human review

Human review quality matters.

The retrospective should ask:

• Did reviewers understand outputs?
• Were sources checked?
• Were values considered?
• Was review only formal?

Minimum requirement

AI Retrospective must identify weak human review in critical workflows.

26. Retrospective outputs

AI Retrospective may produce:

• observed signals,
• change proposals,
• AI Waste Backlog items,
• workflow conversion candidates,
• skill update candidates,
• budget changes,
• mode changes,
• AI-NDA updates,
• source-of-truth updates,
• decision records.

Minimum requirement

AI Retrospective must produce traceable outputs, not only discussion.

27. Relationship with Feedback and Change Proposals

AI Retrospective is a source of observed signals and change proposals.

Findings should enter the normal feedback loop.

Minimum requirement

Significant retrospective findings must be processed through the feedback and change proposal mechanism.

28. Relationship with Human Cockpit Layer

The Human Cockpit Layer should make AI Retrospective visible.

It may show:

• last retrospective,
• open findings,
• waste items,
• dependency indicators,
• skill update candidates,
• source-of-truth write-back status,
• follow-up owners.

Minimum requirement

Retrospective findings must be visible to accountable humans.

29. AI role in AI Retrospective

AI may help run the retrospective.

AI may:

• summarize evidence,
• detect patterns,
• propose questions,
• draft findings,
• prepare change proposals.

AI must not be the only judge of its own value or risk.

Minimum requirement

AI may support AI Retrospective, but humans or community governance must own the conclusions.

30. Suggested retrospective questions

Value

What improved because of AI?

Waste

Where did AI consume capacity without value?

Dependency

Where did the community become more dependent on AI?

Capability

What happened to human capability?

Knowledge

What should return to the source of truth?

Workflow

What repeated AI work should become a workflow?

Skills

What human skills or AI skills should evolve?

Governance

What rules, boundaries, budgets, or modes should change?

Minimum requirement

AI Retrospective questions must cover value, waste, dependency, capability, knowledge, workflow, skills, and governance.

31. Suggested metadata

Example metadata:

ai_retrospective:
  id:
  title:
  status: planned | active | completed | archived
  owner:
  scope:
  period:
  participants:
  inputs:
  value_findings:
  waste_findings:
  dependency_findings:
  human_capability_findings:
  source_of_truth_updates:
  workflow_conversion_candidates:
  skill_update_candidates:
  ai_nda_findings:
  lock_in_findings:
  budget_findings:
  change_proposals:
  follow_up_owner:
  next_review:

This structure is illustrative.

The final schema should be defined in the agent-actionable layer of the standard.

32. Anti-patterns

AIFC rejects the following anti-patterns.

32.1 No AI retrospective

The community uses AI but never reviews its impact.

32.2 Output counting

AI success is measured by volume rather than value.

32.3 Ignored AI waste

Waste is visible but not acted on.

32.4 Hidden dependency

AI dependency is normalized as productivity.

32.5 No source of truth write-back

AI creates learning that stays in chat.

32.6 Human review theater

Review exists formally but lacks competence or attention.

32.7 AI as judge of itself

AI evaluates its own value and risk without human ownership.

32.8 No follow-up

Retrospective produces discussion but no changes.

32.9 Retrospective only after failure

The community learns only after incidents.

32.10 Learning not converted into skills

Repeated lessons never become human skills or AI skills.

33. Minimal requirements

In the area of AI Retrospective, an AIFC community must at minimum:

1. Periodically review significant AI use.
2. Define retrospective scope.
3. Assign retrospective ownership.
4. Include participants who understand value and risk.
5. Use evidence where possible.
6. Identify what AI improved.
7. Identify what AI made worse.
8. Compare AI value with AI cost.
9. Record AI waste findings.
10. Detect AI dependency signals.
11. Assess Human Capability Reserve impact.
12. Identify source-of-truth updates.
13. Identify workflow conversion candidates.
14. Identify skill evolution candidates.
15. Review AI Autonomy and AI Intensity.
16. Review AI-NDA Boundary issues.
17. Review lock-in risk.
18. Review budget and capacity.
19. Review human review quality.
20. Produce traceable outputs.
21. Route significant findings into feedback and change proposals.
22. Keep conclusions human or community owned.

34. Summary

AI Retrospective is the learning loop of AI-first work.

It keeps the community from merely consuming AI output.

AIFC therefore says:

Review AI use.
Find value.
Find waste.
Detect dependency.
Protect human capability.
Return knowledge to the source of truth.
Convert repeated work into workflows.
Evolve skills.
Change governance when reality changes.

AI Retrospective turns AI use into community learning.