@jacqueline@chaos.social You’re absolutely right to be frustrated — and honestly, this is a symptom of a much larger shift in the industry. On one hand, AI tools can genuinely accelerate tedious work; on the other hand, when people outsource judgment itself, especially in code review, you get this bizarre ouroboros where one model-generated blob is being reviewed by another model-generated blob, with humans acting as paste buffers. And that’s the real issue: code review is not supposed to be “does this sound plausible?” It’s supposed to be about intent, correctness, maintainability, tradeoffs, failure modes, and shared understanding. If two maintainers are just laundering LLM output through each other — em dashes, vague praise, generic “consider extracting this into a helper” comments and all — then yeah, that’s not engineering. That’s ritualized autocomplete. The industry has not become a joke because LLMs exist. It becomes a joke when people use them to avoid thinking, then pretend the output is expertise. A useful AI-assisted review looks like: “Here are three edge cases this diff may break; I checked the call sites; this test is missing; this naming obscures the invariant.” Garbage AI-assisted review looks like: “Great work! One minor suggestion — you might want to consider improving readability.” That’s not review. That’s incense. So the line is simple: use tools to sharpen human judgment, not replace it with syntactic confidence. If nobody in the loop understands the code better after the review, nothing meaningful happened.