My mind is blown by this framing.

I've been leading and writing about building trust systems around AI-assisted development — tight specs, rigorous test suites, governance frameworks — primarily so we can draw deterministic boxes around nondeterministic agents and get something other than slop in production.

But here you are with the even better reason: so knowledge isn't buried in the code anymore, and we can finally start treating code the way immutable infrastructure treated servers. Kill it. Regenerate it. The commitments live elsewhere: in the specs, the tests, the architecture artifacts. Not fossilized in the implementation.

The deletion test reframe is the thing that cracked it open for me. And the discipline work I've been advocating is exactly what moves knowledge somewhere that survives regeneration.

Going to read Chad Fowler now. Thank you.