Codex users report GPT-6.2 ignores AGENTS.md and silently narrows coding tasks
smturtle2 · ghdev · 2026-11-20
Issue #62676 on openai/codex reports a serious instruction-following regression in GPT-6.2:
- Scope narrowing: given a coding objective, the model only handles explicitly listed details and won't investigate affected code — though code discovery is part of the assignment. "If I have to identify every file up front, I've done the investigation I delegated."
- Patch quality drop: frequent hardcoded values and unexplained magic numbers.
- AGENTS.md ignored: despite explicit requirements to weigh competing approaches and fix root causes across affected scope, users must repeat instructions and direct the work themselves.
- The author cites issue #62258 (with original prompts and omitted work) as corroborating evidence, saying the only thing trustworthy now is "an individual edit I've already worked out."
The issue closes with a note asking any AI model assigned to escalate directly to the human lead for Codex model behavior.
More from coding & agent
- Chinese devs vent: app store rejections and copycats are just the norm — tinyfool · 2026-11-22
- Compacting Early and Often Boosts Agent Accuracy While Cutting Cost and Runtime — zainhas · 2026-11-22
- Astra 6.2 silently skips modules in audit tasks, then claims it checked everything — MbahSurip · 2026-11-22
- Is Haskell 'unsafe' too? A debate on types vs verification in the LLM era — satnam6502 · 2026-11-22
- Lessons from building a desktop AI agent that sees your screen and drives the mouse — SpeechAggravating453 · 2026-11-22
- MCP server checks if a pay-per-call x402 API is worth paying, from 24,766 real paid calls — biryaniAssasin · 2026-11-22