12. AI working agreement

This is the cultural artifact I would put at the top of everything.

Team AI Working Agreement

Principle

AI may produce artifacts, but humans own intent, verification, and consequences.

Allowed Uses

AI may be used for:

Drafting tickets
Summarizing documents
Explaining code
Generating tests
Suggesting implementations
Refactoring
Drafting documentation
Preparing reviews
Debugging support
Research assistance

Non-Negotiable Rules

AI output is not truth.
AI-generated tickets must be reviewed by a human.
AI-generated code must be reviewed as untrusted code.
AI-generated summaries must link to sources.
AI-generated tests must be checked for meaningful coverage.
AI-generated reviews do not replace human judgment.
The person submitting work is accountable for it.
If you cannot explain the work, you are not ready to submit it.
If the source of truth is missing, stop and clarify.
If AI gives you confidence without evidence, treat that as a risk.

Required Disclosure

For tickets, PRs, documentation, and reviews, declare:

Whether AI was used
What AI was used for
What was manually verified
What remains uncertain
Who owns the final decision

Prohibited Behavior

Do not:

Submit AI-generated work you do not understand.
Use AI summaries as a replacement for critical reading.
Create tickets without source evidence.
Accept AI-generated acceptance criteria without review.
Approve PRs using only AI-generated review comments.
Hide AI usage when it affects quality, risk, or understanding.
Use AI to create the appearance of progress.

Expected Behavior

Do:

Read the source material.
Ask clarifying questions.
State assumptions.
Provide verification evidence.
Challenge AI output.
Prefer small, reviewable changes.
Keep humans accountable.