Skip to content

AI Usage Standard

Version: 1.1.0

Status: Active

Owner: Rafael da Silva Guerra

Purpose

This document defines how AI assistance is used in building and documenting W7, and where its output requires human judgment before it becomes official.

Context

AI tools are used throughout W7's engineering and documentation work — including, transparently, in the drafting of parts of this repository. That fact is stated here rather than hidden, because a standard about AI usage that pretends AI wasn't used would not be credible. What matters is not whether AI was involved, but whether a human took responsibility for the result.

The Core Rule

AI output becomes official only when a human has reviewed it and is willing to be accountable for it. An AI-generated document, commit, or decision has no more authority than a first draft until someone applies the same judgment they would to their own writing: does this hold up, is it accurate, does it avoid buzzwords and hype, does it match how W7 actually thinks (see the Manifesto)?

This mirrors the Engineering Standard's position on tooling generally: a tool that produces output faster does not remove the obligation to understand and stand behind that output.

Where AI Assistance Is Encouraged

  • Drafting documentation for human review and revision, especially for restating existing canonical content in a new context (with a link back to the canonical source — see Knowledge Architecture).
  • Exploring alternatives during the Decision Framework process — AI is a useful source of counterarguments precisely because it has no stake in the outcome.
  • Boilerplate engineering work: scaffolding, tests for well-understood behavior, refactors with a clear specification.
  • Reviewing existing documentation and code for buzzwords, inconsistency, or drift from these standards.

Where Human Judgment Is Required Before Anything Ships

  • Any change to Foundation, Identity, Governance, Framework, Architecture or Standards — these encode shared reasoning and constraints, and reasoning that nobody accepts responsibility for cannot govern the ecosystem.
  • Any Type 2 or Type 3 decision under the Decision Framework.
  • Anything presented as a fact about W7's history, philosophy, or founder intent — the Founder documents in particular are first-person and are not to be AI-authored beyond assisting with phrasing of the founder's own account.
  • Security-sensitive code or configuration, per the Engineering Standard.

What AI Output Must Avoid

Per the Documentation Standard: buzzwords, marketing language, startup clichés, and AI-hype phrasing. AI-generated drafts have a documented tendency toward exactly this kind of language, which is precisely why review against this standard is not optional.

Transparency

There is no requirement to annotate every sentence with its origin — that would be noise, not signal. There is a requirement that anyone contributing to W7 with AI assistance follows this standard the same way they follow the Engineering and Documentation standards: as an expectation, not a suggestion.