Move fast.
Preserve what is true.

Establish an authority order

When documentation, code, old chats and new requests conflict, the project needs a known order of authority. Without one, an AI system may confidently choose the wrong source.

A simple pattern is: current owner instruction → approved product/brand truth → current technical specification → latest approved code → historical material → external research → recommendation.

Separate decisions from discussion

Exploration can be messy. Approved state cannot. Record final decisions in a compact ledger: what was decided, what it supersedes, why it matters and which downstream artifacts need to change.

This is especially important for prices, entitlements, data models, navigation, access rules and deployment behavior.

Write preservation rules into every build

Every material implementation instruction should say what must not change. Preservation rules convert institutional memory into an executable constraint.

PRESERVE:
- approved offer and pricing,
- existing user flow unless explicitly changed,
- server-side entitlement checks,
- brand tokens and approved assets,
- verified mobile behavior,
- current production data model.

Do not replace working architecture merely because another pattern is possible.

Use gates before irreversible actions

Not every change needs a ceremony. Production promotion, payment changes, data migrations and public indexing do. Define the evidence that must exist before those actions happen.

Keep human approval where consequence is high

AI can prepare and verify a great deal of work, but the appropriate human should retain approval for credentials, security posture, legal/compliance interpretation, production release, destructive data operations and other consequential actions.

The purpose of governance is not to slow the builder. It is to let the builder move quickly without losing control of what the system is allowed to decide.