Running a verification session with a team

A verification habit that lives in one person's head is a habit that leaves when they do. Turning the ladder into a team practice means changing what a review meeting is actually for: the explicit, stated goal of the session is to break the result under discussion, not to approve it, and everyone in the room needs to hear that goal stated out loud before the session starts.

The closest working analogy is a postmortem culture, except every reviewed piece of work gets one, whether or not anything went wrong downstream. That is a real cost, a review session takes time that could go toward the next task, and it is worth paying deliberately rather than only after something has already broken in production.

What makes this work in practice is picking one person per session whose only job is adversarial: find the assumption nobody stated, ask what the two agreeing checks actually share, ask what happens at the next size up. Rotate who holds that role so it does not become one person's permanent, resented job, and treat a session that finds nothing wrong as a session that still did its job, not a session that wasted time.

Start today: before your team's next review of an AI-assisted result, assign one person explicitly to try to break it, out loud, in the meeting, before anyone is allowed to say it looks fine. See what changes.