According to developer Vinicius Brasil, featured on Hacker News, the real bottleneck in AI-assisted coding isn't writing code—it's reviewing and judging it. When agents can implement features in minutes, human judgment becomes the limiting factor. Brasil rejects AI-generated code regularly, even when it technically works, because he hasn't fully understood the problem. Before AI, complex tasks required days of exploring the codebase and experimenting—building deep understanding and confidence. Now, he often discards the entire AI output and starts over, letting his own thinking drive the solution. His criteria are specific: he rejects code he cannot explain in his own words, changes larger than the problem, premature abstractions, solutions that work locally but obscure the system, and situations where he trusts the AI more than his own understanding. The core principle: code that passes tests isn't the same as good engineering. Sustainable, scalable solutions require skilled engineers guiding AI agents, not the other way around.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton