GPT-6 Astra Pro user hits agent stuck in edit loop, suspects another OpenAI 'lobotomy' update
the-human-user · reddit · 2026-10-02
A user who has had ChatGPT's agent edit their repo for weeks found it stuck in an apologetic loop, repeatedly failing to strip draft language from a markdown file despite confirmed GitHub access.
- Around the same time, popups offered "retry with Thinking enabled" — even though fast answers were off and they were on GPT-6 Astra Pro (200 uses/week).
- The agent's eventual edit was only partial; the user paused their near-complete project, suspecting one of OpenAI's "helpful but broken" updates. Anecdotal, with no official explanation yet.
Related event: Users report GPT acting dumber, burning quota in agent loops(2 posts)→
More from Models
- Kev 1.0 ships open-weight decision models: Kev-27B with 64k context and fine-tuning skills — victormustar · 2026-10-02
- OpenAI and Anthropic are hiding stronger models internally, Google likely isn't — haider1 · 2026-10-02
- Base models predict "her" via two separate pathways: token gender and post-verb pronoun rules — Sauers_ · 2026-10-02
- Vibe coding vs Overleaf: back to the classic tool in under 2 days — jasondeanlee · 2026-10-02
- Adaptive effort is likely the killer app for dynamic looped transformers — willcb · 2026-10-02
- Dev says Claude and Claude Code are getting worse day by day — Pavan_Belagatti · 2026-10-02