Post #2097443

2026-05-06 16:55 UTC

In the ongoing ai-copalypse at work... I've been tasked with developing a "code review" process for the large LLM generated changes. I pushed back during our standup, basically stating that our "code review" is no longer a "code review" because 1) humans cannot review the code at the speed of LLM code generation and 2) the purpose of said review seems to have shifted. I asked for clarity. What is the purpose of the review going forward? It is decidedly NOT to keep a sound mental model/theory of the codebase, it is not to ensure code maintainability, and it is not to develop standard practices or to acculturate our team... I posited that what is being asked for is more akin to QA than "code review". 1/2

Replies (1)

@pythonbynight@hachyderm.io 2026-05-06 16:55

What was then described sounded more like what a business analyst might do. Look at business requirements. Prompt an LLM to bullet-point the changes. Examine tests to see if "they make sense". And somehow make assessments that the code is secure and performant? I complained that asking an LLM to review LLM generated code is redundant, and that we are making our work redundant based on the belief that the LLMs are fully capable of delivering product with little human intervention. Someone on the team mentioned that we would be spending less time coding and doing more "engineering"... ... at which point I just sort of gave up... lol... engineering? Naaah... Needless to say, "fun" start to the morning. 2/2

Open ##2430899