OpenAI's coding agents now finish 46% of 2-4 hour tasks fully unattended
AIFlow_ML · x · 2026-11-21
Per OpenAI's own write-up (tasks run Aug 1–Oct 18): its coding agents now complete 46% of 2–4 hour research tasks with zero human help, up from 23% in March—doubling in six months.
By task type: 79% on monitoring training runs but only 25% on writing inference and serving code—agents are good at watching, still weak at building.
More from coding & agent
- New Multi-Agent Coordination Bench finds frontier agents coordinate poorly; only Opus gains — AlexGDimakis · 2026-11-21
- LukeW shares full notes from Crazy Agent Scale event on the agentic future — LukeW · 2026-11-21
- Solo dev runs agent swarm in Antigravity to automate SEO analysis — gaganghotra_ · 2026-11-21
- 3500-word guide: building agent evals from real LangGraph traces — strickvl · 2026-11-21
- Hands-on: discovering local AI agents on Windows via Defender for Endpoint — davemccollough · 2026-11-21
- Stanford CS329Z Lecture 9 Released: Optimizing AI Agents via Prompts vs. Weights — stanfordnlp · 2026-11-21