May 25, 2026
Ctrl+Alt+Delusions
Using AI to write better code more slowly
AI coding’s new twist: fewer mistakes, more arguing, way less speed
TLDR: The article argues AI can help people make better software by slowing down and checking for mistakes, not by rushing out sloppy work. Commenters were torn between “this careful approach really helps” and “great, but where’s the proof?” — exposing a bigger fight over whether AI is useful or just exhausting.
The big plot twist in this debate? The author says artificial intelligence for coding doesn’t have to be a speed demon spraying out messy work. Instead, it can be used like an obsessive proofreader: slower, fussier, and constantly hunting for mistakes before anything goes live. In plain English, the pitch is: stop using AI as a chaos machine and start using it as the annoying friend who keeps saying, “Are you sure that works?” The author even says this slower method can uncover old hidden problems, sending people on side quests to fix forgotten bugs and write tests instead of bragging about “10x productivity.”
And the comments? Absolute split-screen drama. Some readers were nodding along hard. One person said they use AI like a patient tutor that points out everything they got wrong until the code finally works — but admitted the whole thing took two hours, which is not exactly the glossy future the hype promised. Another commenter basically described a full-blown AI relay race, with multiple bots reviewing each other’s work in a long, dramatic back-and-forth. It’s giving “too many cooks,” except everyone is a robot. Meanwhile, one skeptic came in with the classic buzzkill energy: where are the actual examples, and why are so many articles about AI coding just vibes and opinions? That complaint landed because, honestly, a lot of people seem tired of grand theories without receipts. The result is a deliciously nerdy culture war: is AI making people better builders, or just slower managers of very needy machines?
Key Points
- •The article argues that AI coding tools can be used for slower, higher-quality development rather than rapid generation of minimally reviewed code.
- •It says LLM agents are effective at finding bugs, citing Mythos and public models from Anthropic and OpenAI as examples.
- •The author describes a multi-model pull request review workflow using Claude, Codex, and Cursor Bugbot to identify and validate bugs.
- •The workflow prioritizes fixing critical and high-severity issues, skipping low-value fixes, and sometimes abandoning a pull request if the approach appears fundamentally flawed.
- •The article says this method may not increase coding speed, but can improve codebase health, reveal older defects, and deepen understanding of system failure modes.