AIFC-013: Human and AI Readable Content

Status: Draft 0.1 Standard: AI-First Community Standard Short name: AIFC Related to:

AIFC-000: Manifesto for AI-First Communities
AIFC-001: Core Concepts
AIFC-002: Community Model
AIFC-010: Knowledge Structure
AIFC-012: Metadata and Markdown

Purpose of this document: Define principles for creating content that is understandable to people, usable by AI agents, and verifiable by software. This document describes how knowledge should be written, structured, explained, and maintained so it becomes usable community capability, not only text.

1. Purpose of this document

This document defines rules for writing AIFC knowledge artefacts.

An AIFC knowledge base should not be only an information repository. It should be an environment where people, AI agents, and software work with the same Source of Truth.

To make that possible, content must be:

human-readable
agent-actionable
software-verifiable

This document focuses mainly on the first two layers:

how to write for people,
how to write for AI agents,
and how to align those two needs.

2. Core principle

The core principle of this document is:

Content must protect human attention and reduce AI ambiguity.

A person needs to understand meaning quickly. An AI agent needs clear context, rules, boundaries, and expected output.

Good AIFC content is not long because it is comprehensive. It is precise, structured, and usable.

3. Human-readable content

Human-readable content is content that a person can quickly understand, use, and approve.

It must be:

clear,
concise,
structured,
contextual,
readable without a special tool,
suitable for a new community member,
usable in decision-making,
not overwhelming.

Human-readable does not mean simplified to the point of losing meaning.

It means the content respects human attention.

Minimum requirement

Every critical knowledge artefact must have a human-understandable introduction explaining:

what the artefact is,
why it exists,
who uses it,
how to act based on it,
who owns it,
what status it has.

4. Agent-actionable content

Agent-actionable content is content an AI agent can use to safely perform work or propose a next step.

It must contain:

clear purpose,
rules,
boundaries,
inputs,
outputs,
allowed actions,
forbidden actions,
decision points,
approval rules,
link to the Source of Truth,
link to the human owner.

An AI agent must not guess what to do when it can be described clearly.

Minimum requirement

Every artefact intended for AI agents must explicitly say:

what the agent may do,
what the agent must not do,
when it should request approval,
where it should write output,
how it should mark uncertainty.

5. Software-verifiable content

Software-verifiable content is content for which at least some rules can be checked automatically.

For example:

the artefact has an owner,
it has status,
it has sensitivity,
it has lifecycle,
an AI workflow has fallback,
a change proposal has decision level,
restricted content does not have unlimited AI access,
a deprecated artefact is not used as active,
Operational DNA has protection.

Software does not have to understand the full meaning of the text. It is enough if it can verify critical rules.

Minimum requirement

Critical artefacts must have enough structure and metadata to validate at least minimal governance requirements.

6. The three-layer content model

AIFC recommends writing critical artefacts in three layers:

Meaning layer
Operational layer
Validation layer

6.1 Meaning layer

The meaning layer explains meaning to people.

It answers:

Why does this exist?
Which problem does it solve?
Which principle stands behind it?
How does it relate to purpose and values?

6.2 Operational layer

The operational layer says how to act.

It answers:

Who does what?
What are the inputs?
What are the outputs?
Which rules apply?
When is approval required?
What happens in an exception?

6.3 Validation layer

The validation layer says what can be verified.

It answers:

Which status must exist?
Which metadata is mandatory?
Which conflicts are unacceptable?
How do we know the artefact is incomplete?
How do we know the rule is violated?

Minimum requirement

Critical artefacts must have at least a meaning layer and operational layer. The validation layer is required for governance, security, AI workflow, Operational DNA, and compliance artefacts.

7. Clarity before completeness

AIFC prefers clear and usable content over exhaustive but hard-to-read content.

Long text without hierarchy creates attention debt.

If a document contains a lot of information, it must have:

clear purpose,
table of contents or navigation,
sections,
short summary,
decision points,
links to detail,
status,
owner.

A community that creates long documents without navigation transfers cost to every future reader.

Minimum requirement

Long artefacts must have structure that enables fast orientation for people and AI agents.

8. Context before instruction

An AI agent needs context.

A person also needs to know why an instruction matters.

AIFC therefore recommends that rules should not be written only as commands.

Poor:

Do not use AI for restricted data.

Better:

Restricted data may contain sensitive operational or personal information.
AI tools may process this data only within an approved AI-NDA Boundary.
Do not use external AI tools for restricted data unless the specific tool, purpose, and boundary are approved.

A good rule explains:

what,
why,
when,
who,
with which exception,
and what to do instead.

Minimum requirement

Critical rules must contain enough context for people and AI agents to apply them in new situations.

9. Use explicit ownership language

AIFC content must clearly distinguish:

who proposes,
who decides,
who approves,
who executes,
who verifies,
who owns the result.

AI may formulate a proposal. The community decides.

This boundary must be visible in the content.

Example

Poor:

AI updates the workflow.

Better:

AI may propose a workflow update.
The workflow owner reviews the proposal.
The approved change is written back to the Source of Truth.

Minimum requirement

Significant AI steps must clearly distinguish proposal, recommendation, decision, approved change, and implemented change.

10. Avoid hidden assumptions

Hidden assumptions are risky for people and AI.

A person may miss them. AI may fill them in incorrectly.

AIFC recommends writing important assumptions explicitly.

Example:

Assumption:
This workflow assumes that customer data is classified as restricted and cannot be processed by external AI tools without an approved AI-NDA Boundary.

Minimum requirement

Critical artefacts must state assumptions when misunderstanding them could lead to a wrong decision, security risk, or AI misuse.

11. Use examples and anti-patterns

Examples help people and AI understand a rule.

Anti-patterns help recognize what is outside the boundary.

AIFC recommends using these for important rules:

example,
counterexample,
anti-pattern,
edge case,
exception.

Example:

Rule:
AI-generated change proposals must be reviewed before becoming decisions.

Example:
AI suggests updating the maintenance workflow. The process owner reviews and approves it.

Anti-pattern:
AI detects a workflow issue and directly changes the active workflow without approval.

Minimum requirement

Critical rules with high risk of misinterpretation must contain an example or anti-pattern.

12. Separate facts, interpretation, and decisions

An AIFC knowledge base must distinguish:

fact
interpretation
proposal
decision

Fact

Verifiable information.

Interpretation

Explanation or synthesis of facts.

Proposal

Proposal for what should change.

Decision

Approved decision by responsible governance.

AI often creates interpretations and proposals. They must not automatically be treated as facts or decisions.

Minimum requirement

AI-generated interpretations and proposals must be marked as interpretations or proposals until approved.

13. Mark uncertainty

AI and people may work with uncertainty.

Uncertainty must not be hidden behind confident text.

AIFC recommends marking:

uncertain source,
missing data,
conflicting information,
unverified assumption,
outdated content,
AI-generated interpretation,
low confidence.

Example:

Confidence: medium
Reason: Based on two outdated process documents and one recent meeting note. Needs process owner review.

Minimum requirement

Artefacts created from incomplete, conflicting, or AI-interpreted sources must mark confidence and review need.

14. Write for future readers

An AIFC knowledge base is not written only for today’s team.

It is also written for:

a new member,
future owner,
auditor,
AI agent,
community after people change,
recovery situation,
future migration.

Content should explain not only what, but also why.

Without why, the community may later know which rule exists, but not why it exists.

Such knowledge is easy to delete, bypass, or automate badly.

Minimum requirement

Significant rules and decisions must contain a reason or link to a Decision Record.

15. Reduce attention debt

Attention debt emerges when content consumes too much human attention to understand basic meaning.

It emerges, for example, when:

documents are too long,
summaries are missing,
structure is missing,
status is missing,
owner is missing,
the same rule is described differently in several places,
it is unclear what is current,
a person must read a whole document for one action.

AIFC content should reduce attention debt.

The Human Cockpit Layer can help, but source content must also be structured.

Minimum requirement

Critical artefacts must have clear purpose, status, and navigable structure.

16. Reduce AI ambiguity

AI ambiguity emerges when an agent lacks enough information to act safely.

It emerges, for example, when:

owner is unclear,
status is unclear,
it is unclear whether content is draft or approved,
sensitivity is unclear,
forbidden actions are not described,
fallback is not defined,
it is unclear what the agent should do when uncertain.

The AI agent then fills gaps by probability.

That is inappropriate for governance.

Minimum requirement

Artefacts used by AI agents must explicitly describe boundaries, uncertainties, and approval rules.

17. Use consistent terminology

AIFC content must use consistent terminology.

If one document uses Source of Truth, another main repo, a third knowledge store, and a fourth truth base, ambiguity emerges.

Core terms must come from AIFC-001 Core Concepts and AIFC-900 Glossary.

Minimum requirement

A new significant term must either be defined in Core Concepts or proposed as a change proposal.

AI agents must not freely introduce new terminology for existing concepts.

18. Avoid decorative language in standards

The manifesto may use stronger language. Standard documents should be precise.

AIFC distinguishes:

manifesto language,
standard language,
implementation guide language,
educational language,
book language.

Strong wording may be useful when it helps explain a principle. It must not replace a rule.

Example of a useful combination:

Everything the community does not care for tends to degrade or create debt.

Therefore, critical knowledge artefacts must have an owner and review mechanism.

Minimum requirement

Standard documents must distinguish principle, explanation, and requirement.

19. Content as capability

AIFC content is not only description.

Good content increases community capability.

For example:

a well-described workflow increases repeatability,
a well-described skill improves onboarding,
a well-described decision prevents repeated debate,
a well-described fallback increases resilience,
a well-described AI-NDA Boundary reduces leakage risk,
a well-described feedback loop increases learning capability.

Poor content creates debt.

Unclear content increases the need for interpretation. Unmaintained content degrades. Unverified content may lead AI and people in the wrong direction.

Minimum requirement

Critical knowledge artefacts must be written as operational capability, not only as notes.

20. Recommended artefact structure

AIFC recommends this structure for critical documents:

Title
Status
Purpose
Context
Definitions
Rules / Requirements
Workflow or Behavior
AI Role
Human Ownership
Fallback
Examples / Anti-patterns
Metadata
Review
Related Artefacts
Summary

Not every document must contain all of this.

But the higher the impact of the artefact, the more of this structure it needs.

Minimum requirement

Critical artefacts must have Purpose, Owner, Status, Rules or Requirements, and Review mechanism.

21. Writing for Human Cockpit Layer

The Human Cockpit Layer will extract:

status,
owner,
priority,
decisions,
risks,
pending approvals,
change proposals,
AI-generated proposals,
deprecated parts,
review needs.

Content must therefore be written so it can be split and displayed.

Long paragraphs without structure are hard to use.

Good structure allows display of:

decision card,
risk card,
change proposal card,
skill card,
workflow step,
approval request,
maintenance item.

Minimum requirement

Artefacts intended for display in the Human Cockpit Layer must have enough structure or metadata to be split into usable parts.

22. Writing for AI agents

An AI agent works better when the artefact contains clear instruction sections.

Recommended sections for an agent-used artefact:

Agent role
Allowed inputs
Allowed actions
Forbidden actions
Expected output
Approval rules
Uncertainty handling
Write-back rules
Fallback

Example:

Agent role:
Analyze submitted change proposals and prepare decision support.

Allowed actions:
Summarize, classify, identify risks, suggest decision level.

Forbidden actions:
Approve, reject, or implement significant changes.

Write-back rules:
Create draft analysis linked to the proposal. Do not update active workflow directly.

Minimum requirement

AI-facing artefacts must clearly describe allowed actions, forbidden actions, and approval rules.

23. Writing for validation

A validator needs predictable structure.

Validation rules must therefore be formulated unambiguously.

Example:

Every critical AI workflow MUST have fallback_defined: true.

Better than:

Critical AI workflows should have some kind of fallback if possible.

AIFC may use:

MUST,
SHOULD,
MAY,
MUST NOT.

Minimum requirement

Validation requirements must be formulated unambiguously and linked to metadata or structure.

24. Content lifecycle

Content evolves.

AIFC content may be:

draft
proposed
under_review
approved
active
deprecated
archived
rejected

People and AI must know whether content is an active rule.

Deprecated content may be useful historically, but must not be used as active guidance.

Minimum requirement

Critical artefacts must have lifecycle status, and AI agents must respect that not every text is current truth.

25. Content maintenance

Content without care degrades.

This applies to documentation, workflows, skills, decisions, AI rules, and value interpretation.

Unmaintained content creates:

knowledge debt,
decision debt,
AI ambiguity,
human attention debt,
security risk,
AI dependency risk,
onboarding debt.

AIFC content must have review and an owner.

Minimum requirement

Critical knowledge artefacts must have a review mechanism and responsible owner.

26. AI-generated content

AI-generated content must be marked if it has not yet been approved.

AI may create:

document drafts,
summaries,
interpretations,
change proposals,
workflows,
skills,
decision support,
risk analyses.

Until approved, the output is a proposal or interpretation.

It is not the Source of Truth.

Minimum requirement

AI-generated knowledge artefacts must have a status and review owner before they can be considered approved or active.

27. Human-approved content

Human-approved content is knowledge that a responsible person or community has accepted as valid.

Approval must be traceable.

It may be recorded as:

metadata status,
Decision Record,
approval log,
Git review,
governance decision,
cockpit approval.

Minimum requirement

Critical approved artefacts must have a traceable approval mechanism.

28. Content migration

When migrating from existing documentation, the community must distinguish:

original content,
AI extraction,
AI interpretation,
human-approved result,
uncertain parts,
conflicting parts,
deprecated parts.

Migration is not only moving text.

It is transformation of content into a structure usable by the community, AI, and validation.

Minimum requirement

Migrated content must not automatically be treated as approved Source of Truth unless it has been reviewed or marked with appropriate confidence and status.

29. Conflict handling

The knowledge base may contain conflicting information.

A conflict may exist between:

old and new document,
two owners,
declared value and workflow,
AI interpretation and human decision,
local rule and higher governance,
public text and internal Source of Truth.

A conflict must not be hidden.

It must be marked and converted into:

review task,
change proposal,
Decision Record,
or governance issue.

Minimum requirement

Detected conflicts in critical content must be marked and assigned to an owner.

30. Recommended language patterns

AIFC recommends clear language patterns.

Purpose

This artefact exists to...

Rule

The community MUST...

AI limitation

AI may propose...
AI must not decide...

Human ownership

The responsible owner reviews and approves...

Fallback

If AI is unavailable, the workflow continues by...

Review

This artefact must be reviewed...

Uncertainty

Confidence is low because...

Change proposal

This observation should be processed as a change proposal because...

Minimum requirement

AIFC implementation guides should provide reusable language patterns for common artefacts.

31. Anti-patterns

AIFC rejects the following anti-patterns.

31.1 Long text without structure

Neither people nor AI have clear navigation.

31.2 Rules without context

Rules are short, but not applicable in new situations.

31.3 Context without decision

A document explains the situation at length, but it is unclear what applies.

31.4 AI suggestion as approved truth

AI output is stored without review as valid knowledge.

31.5 Mixed fact and interpretation

It is unclear what is fact, what is interpretation, and what is proposal.

31.6 Missing owner

Nobody is responsible for the content.

31.7 Missing status

It is unclear whether content is draft, active, deprecated, or rejected.

31.8 No uncertainty marking

Unverified content appears certain.

31.9 Human-hostile structure

Content is validatable, but people cannot read it.

31.10 AI-hostile prose

Content is literary, but an AI agent cannot safely act on it.

32. Minimal requirements

For human and AI readable content, an AIFC community must at least:

Give critical artefacts a human-understandable purpose.
Give critical artefacts an owner.
Give critical artefacts a status.
Give critical rules context.
Define allowed actions in AI-facing artefacts.
Define forbidden actions in AI-facing artefacts.
Define approval rules in AI-facing artefacts.
Mark AI-generated content as a proposal until approved.
Give significant decisions a reason or link to a Decision Record.
Mark incomplete or uncertain information with confidence or review need.
Distinguish fact, interpretation, proposal, and decision.
Give critical artefacts a review mechanism.
Give long artefacts navigable structure.
Allow the Human Cockpit Layer to extract key status information from content.
Allow validators to verify minimal governance requirements.

33. Summary

AIFC content must serve people, AI agents, and software validation.

A person needs meaning, context, and overview.

An AI agent needs rules, boundaries, and expected behavior.

Software needs structure and verifiable requirements.

AIFC therefore says:

Protect human attention.
Reduce AI ambiguity.
Make critical knowledge verifiable.

Well-written content is not only information.

It is community capability.

Human and AI readable content turns knowledge into shared understanding and safe action.