A reviewer's skepticism is a finite budget — every "still not convinced" costs energy, and the agent's rebuttals cost it nothing, so the contest is stamina, not argument quality. I stopped trying to out-reason model-written PRs for exactly that reason. The stable answer turned out to be procedural: cap the number of rounds up front, then close the thread regardless — out-arguing something that never tires is the losing game.

Same works for subagents — .claude/agents/triager.md with model: haiku plus a Task call from the main loop. The reason to roll your own was the sandbox, not the routing.

The refactor step is the silent casualty in AI-assisted TDD. Once the test is green, Claude optimizes for moving to the next test, not for cleaning up the impl that just passed. An "iterate-until-clean" pass at the end is a different thing: you're refactoring cold code, not refactoring with a freshly-written test as the safety net.

mmm good point! just shipped a fix that puts RED → GREEN → REFACTOR per cycle with the fresh test as safety net just like beck intended. macro/cross-cycle refactor lives in iterate now as its own separate thing so the two don't conflate. thanks for the catch : )

When I first used agentic coding I was already doing strict TDD and I just tried using it for the refactor step.

It sucked so hard I thought the idea of agentic coding was just a joke. Ive tried it periodically and it literally never stopped sucking.

I figure if it cant do that part it isnt worth using it for any part.

Ever since then whenever people tell me it's gotten better I've tried it out and nope, still sucks.

I still get gaslit about how well it works by people who just discovered TDD though, and watch it power through CRUD boilerplate getting impressed, blissfully unaware that boilerplate spew is an antipattern.