| > If you’re not reviewing, your vibe coding imo. Before AI, copy-pasting from the Internet was seen as a noob thing to do. Immediately after ChatGPT, copy-pasting from ChatGPT was seen with the exact same sentiment. Maybe the code is more tailored towards you, but it's average, context-less, thoughtless. With agents reading the codebase and writing code in what appears to be a context that approximates a developer's, we've recently crossed the threshold where a serious software developer can say they don't write their own code any more and not get chastised. When you claim reviewing is the golden threshold that must not be crossed: Sure, but we very recently started accepting that code can be written entirely by agents. I don't yet automate code review, but I've been reading about some code review agents called Hickey and Löwy: https://kolu.dev/blog/hickey-lowy/#what-the-two-lenses-are -- > A Hickey reviewer reads code the way a lockpicker reads a tumbler2 — looking for concepts that shouldn’t be in the same position. The output is always “split these apart.” > A Löwy reviewer reads code the way an actuary reads a portfolio3 — looking for things coupled to unrelated schedules. The output is always “draw a boundary that encapsulates this volatility.” So the art of reviewing can, perhaps, be synthesized. Perhaps not for complete human replacement. I believe that a lot of wetware is still highly mechanical and can be extracted as agentic workflows. The goal is to move the human towards "software architect" and "product owner". The real danger is cutting off the bridge that qualifies current seniors to be hands-off. I don't see the skill atrophy of being an AI manager paying back what took me 20 years of pain to qualify. |