That's mostly true, but I'm not convinced it's harmful to have agents also write tests for lower level details – code is cheap now, and I don't mind agents throwing away dozens of existing tests when they change how a feature works

→ View original post on X — @simonw