Auto-research with codex: How I achieved a 232x Faster Kernel
- ID
- 14513
- Status
- summarized
- Published
- 15 Aug 2026, 7:00 PM
- Fetched
- 17 Aug 2026, 2:35 PM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://sankalp.bearblog.dev/autoresearch/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.0
- Created
- 17 Aug 2026, 2:37 PM
- Tags
- Audience
- developersai_ml_learnersai_agent_users
What happened
A participant in GPU Mode's auto-research contest used Codex in an automated loop to optimize a batched Householder QR factorization CUDA kernel, achieving 232x speedup over baseline and placing 12th of 183. The post details how introducing 'idea diversity' to escape local maxima and feeding the agent progressively better mathematical context (blocked Householder algorithm, serial work reduction) drove breakthroughs.
Why it matters
This is a concrete first-hand account of 'loop engineering' — using an AI coding agent iteratively to solve a hard optimization problem where the human doesn't need deep domain expertise upfront. If you build or use AI agents for technical tasks, the key actionable insight is that breaking out of local optima required deliberately injecting diverse strategy prompts rather than letting the agent refine one approach.
Discussion angle
How far can 'auto-research' go when the human's main contribution is curating which questions and strategies to feed the agent, rather than solving the problem directly — and what does that mean for how developers should spend their time?