BonanKou

6 ms·

1. 1.▲
   by BonanKou 1mo ago

   I agree with the direction, but I think "stop doing code reviews" is slightly too far. The review boundary should move up. Engineers should review the plan made by agents and trust they will carry out the implementation correctly.
2. 2.▲
   by BonanKou 1mo ago

   This matches my experience. For a lot of agent-written code, reviewing the final diff feels too low-level and too late. I'd rather review the plan&assumptions. If those are right, AI will probably get the implementation right as we
3. 3.▲
   by BonanKou 1mo ago

   +1 to the timeline request, with a warning from having built one: it looks like a logging feature but it's actually a data model decision.
4. 4.▲
   by BonanKou 1mo ago

   TBH I think those who went back to hand-written code overreacted. In my experience, once you have nailed down the data structures and test cases (and for frontend, the visual design), letting AI fill the gap in between is a valid approach.
5. 5.▲
   by BonanKou 1mo ago

   TBH I just use another ClaudeCode to review or ask the agent to explain its reasoning process with me. I personally believe the reviewable unit shouldn't be the final diff. Instead, review the decision trace: plan, assumptions, etc. An
6. 6.▲
   Looking to connect by helping founders find their first users

   2 points by BonanKou 3mo ago|0 comments
7. 7.▲
   EOL – Find an audience from public posts, talk to it as one(earthonlines.com)

   2 points by BonanKou 3mo ago|0 comments