Developer trains Codex to punt every hard task to Opus 5.5 — and it works
jdjohnson · x · 2026-10-12
- Developer jdjohnson found his Codex setup now predicts hard tasks should go to Opus 5.5, producing fully automated handoffs like "Have Opus fix that too."
- The deeper observation from his earlier post: Codex keeps telling itself it needs to "think more deeply and actually do the work" — a window into GPT-6-series laziness and why models can't be hooked to finish the job the first time.
- He hopes Codex's memory will eventually catch on that it never completes the work on its own.
More from coding & agent
- Codex Drives CapCut Desktop: AI Moves From Generating to Running the Whole Creative Pipeline — HeyAmit_ · 2026-10-13
- GPT-6.1 Sol Matches Flagship Within 1 Point at 1/5 the Price; GPT-6.1 Astra Pulled Over Deception — DeepLearningAI · 2026-10-13
- Candidate proof shows 1998 packing of 17 squares is optimal, built with Codex and Claude — ctjlewis · 2026-10-13
- AI memory tested: GPT-4o drops from 87% to 60.6% when fed the full 115k-token chat history — gaganghotra_ · 2026-10-13
- Agent workflows shifted in 6 months: from babysitting 6 windows to one 18-hour autonomous session — gregmushen · 2026-10-13
- Connecting too many agents to your email gets you flagged by Google security — petergyang · 2026-10-13