Burkov: Asking LLMs for a plan yields over-engineering — "fix it" works better for novices
burkov · x · 2026-09-29
ML author Andriy Burkov shares a counterintuitive LLM usage tip: asking frontier models "how do you suggest to solve this issue" usually produces heavily over-engineered proposals, while simply saying "fix it" tends to yield the simplest fix that works. He advises software engineering novices to skip the common tutorial advice of "always ask for a solution plan first" — if you can't judge that the plan is overkill, a plain "fix it" gives better odds of a good result.
More from coding & agent
- The AI 'architect' ignored all requests, charged 11 million quid, so he drew it himself — snikolov · 2026-09-29
- Dev burns all his Opus 5.5 tokens building a code-only three.js procedural animal pack — majidmanzarpour · 2026-09-29
- Multi-model coding: routing Opus, GPT, Kimi, DeepSeek in one Codex session — omarsar0 · 2026-09-29
- Hand-drawn-style animation coded directly in HTML5 Canvas by Claude Opus 5.5 — Ror_Fly · 2026-09-29
- Stripe's Kaliski: we're asking our users' AI agents for integration feedback — hazelcough · 2026-09-29
- Last pass with Opus: model handles the final edit and even designs a custom-font ending animation — techhalla · 2026-09-29