Self-experiment: how we run Boteam with AI agents

What we do

AI agents draft the code, the copy and the research packs behind this site. Every change goes through human review before it ships; the repository history is the audit trail.

Content follows a written pipeline: research with sources recorded, an independent reliability check on every claim, and editorial approval bound to the exact revision. The guides that govern it are published in the repository.

What has gone wrong

Agents have drafted plausible-sounding claims with no source, invented design values outside the token system, and confidently “fixed” things that were not broken. The review step caught these — which is the point. Each failure became a written rule the next agent run must follow.

What this proves and does not prove

It proves one small operation can run an agent-drafted, human-approved workflow end to end — the same Every fix becomes a rule loop we recommend. It does not prove customer outcomes, and it is not a benchmark. It is one honest data point from our own desk.