Using AI to write better code more slowly

4 hours ago

LLMs can be used to write high-quality code slowly, not just low-quality code quickly.
LLMs and AI agents are effective at finding bugs in codebases, though prioritization and validation remain challenges.
A technique involves using multiple models (like Claude, Codex, and Cursor Bugbot) for PR reviews to reduce hallucinations and false positives.
The workflow includes fixing critical/high bugs, skipping less valuable fixes, and abandoning PRs if the approach is misguided.
This method improves codebase health and understanding, focusing on quality over velocity, and can involve side-quests like writing unit tests.
Slowing down with AI-assisted review, using skills like 'grill-me' or multiple sweeps, leads to more careful and methodical coding.

Hasty Briefsbeta