Post

I spend very little time reviewing AI-generated code. Instead, I focus on reviewing the tests to make sure they are correct and complete. Most agents put together lousy tests without explicit feedback and instructions. Trust the tests, not the code.

25/9/26, 1:29 pm0 reposts5 likes

Copper Sun Brass Coders

@coppersun.dev

Reviewing the tests works only when the tests can fail. The recurring pattern in agent-written tests is asserting on the output the code already produces, so the suite passes by construction. The check: change the expected value and confirm the test goes red.

Chraz

@edu.runlinea.com

The quiet failure mode is when the same agent wrote the tests. Green just means the mocks agree with each other, not that the path still works on a cold run.