01What happened
The story, straight
A Hacker News post by developer @vnbrs argues that AI coding agents have shifted the bottleneck from implementation to code review, and that he regularly rejects all AI-generated changes and starts over. He says the cognitive overload of reviewing code he hasn't personally thought through outweighs the speed gains — even when following best practices like plan mode, phased tasks, and small commits.
dev @vnbrs on Hacker News says he rejects all AI-generated code and rewrites it himself, even when it works. the problem: AI makes writing code fast but reviewing it brutal. he hasn't internalized the reasoning behind the changes, so explaining them to coworkers and maintaining confidence is harder than if he'd just done it the slow way.
02Spread timeline
Where it actually started
03Source receipts
Every claim, linked
04What's solid, what isn't
What's solid and what isn't
- @vnbrs published a blog post arguing AI coding agents shift the bottleneck from implementation to review.
- He reports regularly rejecting all AI-generated changes and starting over from scratch.
- The broader prevalence of this practice among professional developers (the post is anecdotal, not a survey).
- Whether 'more often than not' is precise or rhetorical.
05Why it matters
The editorial take
This captures a growing tension in software engineering: AI coding tools accelerate output but create a review bottleneck that experienced developers find cognitively expensive. The post resonates on Hacker News because it articulates a problem many engineers feel but few have formalized — that speed without understanding trades one bottleneck for another.
the AI-code-speed debate isn't about whether the code works. it's about whether you can own it after. this post nails the thing every dev using copilot/cursor agents has felt but hasn't said out loud: reviewing AI output you didn't think through is harder than just writing it yourself.