A circle, not a stage. No slides.

Most teams have adopted agentic tooling faster than they've agreed any rules for it. The result is a line that exists in everyone's head and nowhere in writing — and it's in a different place for every engineer on the team.

This roundtable is for the people making that call. We'll work it out together:

Which parts of a codebase should stay off-limits — auth flows, crypto, payment paths, permissions, release configuration — and does the room actually agree?
What can safely go into a model, and what should never leave your machine or your tenant?
Who is accountable when agent-written code ships and fails? The author, the reviewer, or the team that approved the tool?
How does code review change when nobody wrote the first draft?
What has your team already put in writing, and is it holding?

I'll bring a starting framework — a four-tier data classification and a human-in-the-loop mandate I've written for regulated environments — purely as something to argue with. The value is in what the room brings, not what I assert.