- Author
- Nolan Lawson
- Published
- 25 May 2026
- Kind
- Personal blog post, with a reader comment thread
- Original
- nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/
- Subject
- Coding with agents
What it is
Nolan Lawson is a programmer who works at Socket and writes a personal blog. The post argues against the idea that AI coding means fast, low-quality output. It gives a review workflow and urges careless AI coders to slow down. Lawson later shared the skill file in a comment.
What it says
- Lawson argues LLMs can also write high-quality code, more slowly.
- The post says current public models find plenty of bugs in unscrutinized code.
- Lawson's skill runs a Claude sub-agent, Codex and Cursor Bugbot on a PR. The main agent then rules out false positives.
- Lawson reports that this skill finds many bugs with a near-zero false positive rate.
- The workflow fixes criticals and highs. It skips costly edge cases and abandons PRs with too many criticals.
- Lawson says velocity has not necessarily risen, since review finds older bugs and side quests.
- In comments, Lawson agrees clearing context helps. Lawson leans toward valuing higher-level understanding over line-level checks.
What evidence it gives
The evidence is the author's own experience, with no counts or benchmarks. The skill text is paraphrased in the post and linked in a comment. Commenters give similar experiences and a few objections.
Interests and limits
The author works at Socket, per the blog's about box. The false positive claim is not measured. One commenter says the near-zero rate may hold only for experienced developers. Lawson calls this a good point. Lawson also says the tool sometimes flags unlikely or impossible issues.