AIFC-024: Human Capability Reserve

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

AIFC-000 Manifesto of an AI-first community
AIFC-001 Core Concepts
AIFC-002 Community Model
AIFC-003 Values and Purpose
AIFC-010 Knowledge Structure
AIFC-011 Operational DNA
AIFC-020 Human-Managed AI
AIFC-021 AI as External Expert Capacity
AIFC-022 AI-NDA Boundary
AIFC-023 AI as Team Member

Purpose of this document: Define Human Capability Reserve as the deliberately maintained ability of people and the community to understand, perform, review, recover, and transfer critical work even without AI. Describe the relationship between AI acceleration, AI dependency, human competence, fallback, junior work, learning, and community resilience.

1. Purpose of this document

This document defines Human Capability Reserve.

AIFC assumes that AI can dramatically increase community performance.

AI may:

speed up work,
improve quality,
reduce routine effort,
help people formulate thoughts,
propose solutions,
clean the knowledge base,
support decision-making,
generate proposals,
support learning.

At the same time, an invisible risk may emerge:

The community begins to lose the ability to perform work without AI.

This may not show up immediately. At first, it looks like productivity. Only when AI, tokens, a vendor, a model, a tool, or permissions become unavailable does it become visible that part of the human capability has disappeared.

Human Capability Reserve is AIFC’s answer to this risk.

2. Core principle

The core principle of this document is:

AI should increase community capability, not replace it with dependency.

AIFC says:

A community must remain capable of understanding, validating and recovering critical work without AI.

If a token outage stops simple routine work, the organization has not gained intelligence. It has lost resilience.

3. Definition

Human Capability Reserve is the deliberately maintained ability of people and the community to perform or recover critical work without AI.

It includes the ability to:

understand the work,
assign the work,
perform the work,
review AI output,
fix an error,
explain a rule,
teach a new member,
continue during AI outage,
restore a workflow,
decide without AI recommendation,
and preserve community purpose without external intelligence.

Human Capability Reserve does not mean rejecting AI.

It means preserving human and community resilience.

4. AI acceleration vs AI dependency

AIFC distinguishes AI acceleration and AI dependency.

AI acceleration

AI acceleration is the state in which a human or community can perform the work, while AI makes it faster, more precise, or broader.

Examples:

a developer can write routine code and AI helps them do it faster,
an analyst understands the dashboard and AI helps propose a better description,
a team understands the process and AI helps find duplicates,
an owner understands the decision and AI prepares options.

AI acceleration is desirable.

AI dependency

AI dependency is the state in which a human or community cannot perform work without AI that, given their role, they should be able to handle.

Examples:

a developer stops on a simple routine implementation because AI tokens ran out,
a team cannot write a customer response without AI,
an owner cannot assess an AI proposal because they do not understand their own process,
nobody can edit the knowledge base without an agent,
decision-making stops because AI did not generate a recommendation.

AI dependency is a risk.

Minimum requirement

An AIFC community must regularly distinguish whether its AI use creates AI acceleration or AI dependency.

5. Why Human Capability Reserve matters

Human Capability Reserve matters for several reasons.

5.1 Resilience

The community must continue during AI outage.

An outage may be caused by:

token exhaustion,
model unavailability,
price change,
vendor incident,
legal restriction,
security incident,
access loss,
change in terms of service,
deliberate switch into AI-off mode.

5.2 Accountability

A human or community is accountable for decisions and outputs.

Accountability cannot be transferred to AI.

For a human to be accountable, they must be able to understand and assess the AI output.

5.3 Learning

If AI performs work in a way that makes people stop understanding the basic principles, the community loses the ability to teach new members.

Junior work, manual practice, and basic understanding are not inefficiency. They are mechanisms for reproducing capability.

5.4 Vendor independence

A community that cannot work without a specific AI is vulnerable.

A vendor, model, or tool may change.

Human Capability Reserve supports exit strategy.

5.5 Quality control

Someone who cannot perform or at least understand the work in principle cannot properly review AI output.

Without human competence, human review becomes an empty ritual.

Minimum requirement

Critical community capabilities must have a human or non-AI variant for understanding, validation, and recovery.

6. Critical capabilities

Not every capability must be maintained without AI at the same level.

AIFC focuses on critical capabilities.

A critical capability is a capability whose loss would significantly harm:

community purpose,
operations,
security,
quality,
trust,
customers,
governance,
legal accountability,
Operational DNA,
recovery capability,
or relationships with other communities.

Examples of critical capabilities:

decisions about values,
security incident response,
customer support in critical situations,
work with Operational DNA,
source of truth maintenance,
basic delivery capability of a team,
ability to review AI outputs,
ability to perform manual fallback,
onboarding new members,
approving change proposals.

Minimum requirement

The community must identify critical capabilities where full AI dependency must not emerge without an approved risk.

7. Human understanding

Human Capability Reserve begins with understanding.

A human does not always need to perform all work manually.

But they must understand:

what is happening,
why it is happening,
what the input is,
what the output is,
what the limits are,
how to recognize an error,
what the fallback is,
who the owner is,
what values are affected,
when escalation is needed.

If a human merely accepts AI output without understanding, hidden decision risk emerges.

Minimum requirement

Critical AI workflows must have a human-readable explanation so the accountable human can understand what AI does and how to assess the output.

8. Human execution

Some critical work must be possible without AI.

It does not need to be equally fast.

It does not need to be equally cheap.

It must be acceptable for fallback mode.

Examples:

manually handling a critical support request,
manually creating a basic Jira ticket,
manually reviewing a security change,
manually updating the source of truth,
manually deciding a change proposal,
manually following a deployment checklist,
manually communicating with a customer,
manually creating a minimal report.

Minimum requirement

Critical workflows must have a defined non-AI execution variant or an approved risk of not having one.

9. Human validation

Human validation is the ability to review AI output.

It includes:

factual review,
source checking,
alignment with values,
security review,
data sensitivity review,
decision assumption review,
impact review for other communities,
review of whether the output belongs in the source of truth.

Without validation capability, AI output may become unverified truth.

Minimum requirement

Critical AI outputs must have a reviewer with sufficient competence to assess the output.

10. Human recovery

Human recovery is the ability to restore operations or knowledge state when AI fails.

It may include:

AI-off fallback,
manual checklist,
recovery runbook,
source of truth export,
offline documentation,
alternate workflow,
alternate vendor,
human review board,
disaster recovery test.

Human recovery is the practical part of Human Capability Reserve.

Minimum requirement

Critical AI-dependent workflows must have a recovery procedure or an approved risk of not having one.

11. AI-free work

AIFC recommends that the community deliberately preserve some work without AI.

This rule may be adjusted by work type.

Example:

At least 10 % of selected critical work types should be regularly performed without AI to preserve human capability.

This number is not a universal requirement.

It is a recommended pattern.

What matters is that the community deliberately maintains human practice where full AI dependency would be dangerous.

Minimum requirement

The community must have a mechanism for regularly practicing or verifying non-AI capability in critical areas.

12. Junior work and capability reproduction

Junior work is not only cheap work.

It is a mechanism by which the community reproduces its capabilities.

If AI removes all simple and repeatable tasks, the community may lose the natural path by which new members learn the basic principles.

This is especially important in:

development,
analytics,
support,
project management,
documentation,
UX,
security,
operations,
decision-making.

AI can support juniors very well.

But it should not take away their opportunity to learn basic work.

Minimum requirement

The community must consider the impact of AI automation on onboarding, learning, and reproduction of human capabilities.

13. Human skills

Human skills are a key part of Human Capability Reserve.

Every critical AI-assisted capability should have a corresponding human-readable skill.

A human skill describes:

principles,
procedure,
decision rules,
output quality,
examples,
anti-patterns,
review questions,
fallback,
when to use AI,
when not to use AI.

If only an AI skill exists and no human skill exists, there is a risk that know-how is available to the agent but not to the community.

Minimum requirement

Critical AI skills must be linked to human skills or human-readable knowledge in the source of truth.

14. AI skills and human dependency

An AI skill may accelerate agent work.

But if an AI skill contains know-how that humans cannot understand, AI skill dependency emerges.

The risk is especially high if the AI skill:

is not exportable,
is stored in a proprietary tool,
has no human-readable explanation,
has no owner,
is not reviewed,
contains decision logic,
works with Operational DNA.

Minimum requirement

Critical AI skills must be exportable, versioned, and explainable to humans.

15. AI dependency indicators

An AIFC community should monitor AI dependency indicators.

Examples:

work stops during AI outage,
people cannot explain AI output,
human review is only formal,
a team cannot create a basic output without AI,
junior learning worsens,
documentation is created only from AI chats,
agent memory contains context outside the source of truth,
AI-generated proposals are accepted without resistance,
people say “I cannot do this without AI” for routine work,
token limits stop simple implementation,
an AI workflow has no fallback.

Minimum requirement

AI retrospective must include a review of AI dependency indicators.

16. Human capability risk levels

AIFC may use human capability risk levels.

Level 0 - No dependency

AI helps, but people can perform and validate the work without AI.

Level 1 - Assisted dependency

People can do the work, but AI significantly accelerates it.

Risk is low if fallback exists.

Level 2 - Operational dependency

Without AI, work becomes significantly slower or worse.

Requires fallback and cost/risk control.

Level 3 - Capability dependency

People are losing the ability to understand or perform the work independently.

Requires training, human skill update, and AI-free practice.

Level 4 - Critical dependency

A critical capability of the community is effectively transferred into AI, a vendor, or agent memory.

Requires immediate governance attention.

Minimum requirement

Critical workflows must have a human capability risk assessment if they are strongly AI-assisted or AI-dependent.

17. Relationship with AI-off fallback

Human Capability Reserve and AI-off fallback are closely related.

AI-off fallback is an operating procedure.

Human Capability Reserve is the ability of people to understand and perform that procedure.

A fallback document without capable people is not enough.

Capable people without a described fallback are also not enough.

AIFC requires both.

Minimum requirement

A critical fallback must be not only described, but also periodically verified or practiced.

18. Relationship with AI Retrospective

AI Retrospective must evaluate the impact of AI on human capabilities.

Questions:

Where did AI help?
Where did AI create dependency?
Where are people beginning to stop understanding the work?
Where is human review weakening?
Where is a human skill missing?
Where is AI-free practice needed?
Where does an AI skill contain know-how that has no human variant?
Where did an AI outage occur and what did it reveal?
Where is workflow conversion or fallback needed?

Minimum requirement

AI Retrospective must generate change proposals if it identifies a risk of human capability loss.

19. Relationship with maintenance

Human Capability Reserve requires maintenance.

Human capabilities, like the knowledge base, degrade if they are not used and maintained.

What the community stops practicing, it eventually loses.

This applies to:

routine implementation,
writing,
analysis,
decision-making,
troubleshooting,
support,
security response,
work with values,
reading the source of truth,
reviewing AI outputs.

Maintenance of human capabilities is part of community resilience.

Minimum requirement

Critical human skills must have a review, practice, or onboarding mechanism.

20. Relationship with source of truth

The source of truth supports Human Capability Reserve.

If know-how is stored only in AI chat, people cannot reliably learn from it.

If it is stored in a structured source of truth, it can serve:

onboarding,
fallback,
human skills,
review,
training,
knowledge transfer,
recovery,
vendor exit.

Minimum requirement

Know-how required for critical human capability must be stored in the source of truth, not only in an AI tool or agent memory.

21. Relationship with Human Cockpit Layer

The Human Cockpit Layer must make the state of human capability visible.

It may show:

critical skills,
skills without owner,
skills without review,
AI-dependent workflows,
missing fallback,
human capability risk,
AI-free practice schedule,
training needs,
junior learning paths,
AI dependency indicators,
capability incidents.

The Human Cockpit Layer helps prevent capability loss from becoming invisible.

Minimum requirement

The community must have a human-accessible way to see critical AI dependency and human capability risks.

22. Relationship with Operational DNA

Operational DNA contains critical workflows, skills, decision logic, and fallbacks.

Human Capability Reserve protects the community’s ability to actually use its Operational DNA.

Operational DNA without human capability may become a documented but dead system.

Human Capability Reserve ensures that the community does not lose the ability to:

read its own Operational DNA,
explain it,
perform it,
fix it,
transfer it,
recover it.

Minimum requirement

Critical Operational DNA must be accompanied by human-readable skills or fallback procedures.

23. Relationship with AI-NDA Boundary

Human Capability Reserve may reduce pressure toward risky AI use.

If people can perform work themselves, there is no need to give AI access to all data just to keep work moving.

Conversely, weak Human Capability Reserve may cause the community to violate the AI-NDA Boundary out of practical necessity.

For example:

“We have to put this into external AI, otherwise nobody can do it.”
“We do not have a person who can review it.”
“The agent has to remember it because nobody maintains the documentation.”

This is a warning signal.

Minimum requirement

Violating or bypassing the AI-NDA Boundary because of missing human capability must be handled as a governance risk.

24. Relationship with AI as Team Member

An AI team member may support or replace human capability.

It supports it when it:

explains steps,
teaches a human,
creates human skills,
proposes review questions,
helps a junior,
shows alternatives,
records know-how in the source of truth.

It replaces it in a risky way when it:

performs work without explanation,
stores know-how only in agent memory,
creates outputs nobody can assess,
removes all learning tasks,
becomes the only performer of routine work.

Minimum requirement

An AI team member must be evaluated by whether it strengthens or weakens human capability.

25. Human review quality

Human review is not automatically high quality.

Weak human review:

merely clicks approve,
does not check sources,
does not understand the output,
lacks time,
trusts AI too much,
is afraid to challenge AI,
lacks the needed skill.

Good human review:

understands the context,
checks critical assumptions,
recognizes uncertainty,
assesses values and risks,
knows when to escalate,
can reject the output,
records important learning back into the source of truth.

Minimum requirement

For critical AI workflows, human review must be assigned to a human or team with sufficient competence, time, and authority.

26. Training and practice

Human Capability Reserve requires training.

Training may include:

AI-free task days,
manual fallback drills,
junior practice tasks,
review calibration,
pair work without AI,
incident simulation,
source of truth navigation,
decision record writing,
manual support handling,
code without AI,
AI output critique sessions.

The goal is not to reduce productivity.

The goal is to protect community capability.

Minimum requirement

Critical capabilities must have a training or practice mechanism proportionate to risk.

27. Capability transfer

The community must be able to transfer capabilities.

Capability transfer includes:

onboarding,
mentoring,
documentation,
human skills,
examples,
anti-patterns,
practice tasks,
review patterns,
decision records,
fallback drills.

AI may support capability transfer, but it must not be the only carrier of it.

Minimum requirement

A critical capability must not depend on one human, one agent, or one agent memory.

28. Human capability incidents

A community may record a human capability incident.

Examples:

work stopped because AI was unavailable,
a reviewer approved an incorrect AI output because they did not understand it,
a team could not restore a workflow without an agent,
a junior is not learning basic skills,
agent memory contained know-how that people did not know,
nobody could maintain AI-generated documentation.

Such incidents are not only individual failures.

They are signals of systemic risk.

Minimum requirement

Significant human capability incidents must be processed as observed signals or change proposals.

29. Suggested metadata

Example metadata for human capability assessment:

human_capability_assessment:
  id:
  title:
  status: draft | active | under_review | deprecated | archived
  capability:
  owner:
  related_workflow:
  related_ai_workflow:
  related_human_skill:
  related_ai_skill:
  criticality: low | medium | high | critical
  ai_dependency_level: 0 | 1 | 2 | 3 | 4
  human_execution_possible: true | false | partial
  human_validation_possible: true | false | partial
  ai_off_fallback_defined: true | false
  ai_off_fallback_tested: true | false
  junior_learning_path_available: true | false
  training_required: true | false
  last_reviewed:
  review_cycle:
  risks:
  mitigation:

This structure is illustrative.

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

30. Anti-patterns

AIFC rejects the following anti-patterns.

30.1 AI productivity masking skill loss

A team appears productive, but loses the ability to perform work without AI.

30.2 Token outage stops routine work

Simple routine work stops because tokens or AI access ran out.

30.3 Human review without competence

A human formally approves an output they do not understand.

30.4 AI skills without human skills

An agent has instructions, but humans do not have a corresponding human-readable skill.

30.5 No AI-off fallback

A critical workflow has no non-AI path.

30.6 Junior work removed

AI automation removes simple tasks through which new members learned.

30.7 Agent memory as hidden teacher

An AI agent remembers know-how that is not in the source of truth and is not known by people.

30.8 Capability owned by vendor

A critical capability of the community is effectively owned by an AI vendor or proprietary tool.

30.9 No practice

Fallback exists in a document, but nobody can perform it.

30.10 AI dependency treated as innovation

AI dependency is presented as progress even though it reduces resilience.

31. Minimal requirements

In the area of Human Capability Reserve, an AIFC community must at minimum:

Distinguish AI acceleration and AI dependency.
Identify critical capabilities.
Give critical capabilities a human-readable description.
Give critical AI workflows a human validation mechanism.
Give critical AI workflows an AI-off fallback or approved risk of not having one.
Periodically verify or practice critical fallback.
Link critical AI skills to human skills or human-readable knowledge.
Use AI retrospective to monitor AI dependency indicators.
Monitor human capability risk in strongly AI-assisted workflows.
Maintain a mechanism for AI-free practice or verification of non-AI capability.
Ensure onboarding and junior learning are not fully replaced by AI.
Process significant human capability incidents as observed signals or change proposals.
Assign human review of critical AI outputs to a competent reviewer.
Ensure critical know-how is not stored only in an AI tool or agent memory.
Make AI dependency and human capability risks visible in the Human Cockpit Layer or governance interface.
Treat AI-NDA Boundary violations caused by missing human capability as governance risk.
Evaluate AI team members by their impact on human capability.
Give critical human skills an owner, review, and maintenance mechanism.

32. Summary

Human Capability Reserve protects the community from an AI-first approach turning into AI dependency.

AI may speed up work.

But the community must remain able to:

understand its own work,
review AI outputs,
decide,
transfer capabilities,
teach new members,
recover critical workflows,
and continue during AI outage.

AIFC therefore says:

Use AI to strengthen people.
Do not use AI to quietly remove capability from the community.

An AI-first community should be faster because of AI.

It must not be helpless without it.

Human Capability Reserve turns AI acceleration into resilient community capability.