Burkov spots Codex quirk: suggested follow-up knows the agent screwed up, but the agent doesn't
burkov · x · 2026-11-20
Andriy Burkov, author of The Hundred-Page Machine Learning Book, flags a curious behavior in OpenAI's Codex: the auto-generated suggested follow-up prompt seems to "know" the agent broke the code in the previous iteration and asks it to revert, yet the Codex agent itself doesn't get that signal and ships the broken change as a fix. The observation hints at a sync gap between the product's follow-up suggestion logic and the agent's own context, with implications for self-correction reliability in coding agents.
More from coding & agent
- Claude Code 2.1.329 removes built-in confirmation rule for high-risk actions — ClaudeCodeLog · 2026-11-20
- Ex-Googler open-sources a CLI that lets Claude Code and Codex chat in one room — thechrisperry · 2026-11-20
- Vibe-coding day 17: a WebGPU kelp lantern built with Gemini in Antigravity — AI_Andrew · 2026-11-20
- Beaker autonomously finds 100+ grading errors across five popular AI benchmarks — VictorSanh · 2026-11-20
- QuixiInfer author: tell Claude 'defeat is not an option' to make it write kernels for your GPU — QuixiAI · 2026-11-20
- OpenAI's Codex repo has 27,802 open issues: agent harness maintenance compared — SuaveSteve · 2026-11-20