alt Hacker NewsThis. We’ve just moved the bottleneck from during writing the code to reviewing. The failure mode here is that you can choose to just do a light review since the code “works” but if you had to write the code yourself you would have been forced to consider the design choices and broader implications of the code more.
Replies
The bottleneck they feel during coding is the lack of knowledge. Knowledge of the technical platform and knowledge of the deeper implications of the task’s requirements.
I’ve mentored juniors and the common trait is that they rush to code without understanding, then they spend days trying to fix things. Coding will seem like a real bottleneck for them.
AI agents can build a whole app and review PRs better than a lot of people, while having access to more context at once.
Honestly, where did the bottleneck end up now ? It sure isn't simply reviewing, or designing, so where is it ?
> We’ve just moved the bottleneck from during writing the code to reviewing
Ah, that's only true if whoever is in charge is willing to _have_ a bottleneck in reviewing. In some software development groups, there is a strong pressure for reviews to pass, i.e. expectation that reviewers approve; and the notion of veto is not common, so that if some reviewer objects, the implementing developer can find others who would be willing to accept, bypassing the objections. Also, few people are willing to conduct long reviews of a lot of code, and even less people are willing to read multiple comments in a long review of a lot of code.
Now, in some projects, there are only one or two "owners" or "marshals" for some areas of the code, so they can apply a strict standard if they want to, but - their time is limited, meaning that long LLM vibe-pull-requests are likely to either be ignored for a long time or rejected off-hand.
I have many years of the experience of of raising concerns, criticism and objections in reviews