@apropos @7666 the logic is that it's easier to write tests than code so the LLM gets it right or closer to right far more often, then you can just automate the code generation process so even if it takes longer, it took less of your productive time.