OpenAI DevDay preview: Codex 1,000-hour runs, ChatGPT Sites and WebMCP leads
johnseach · x · 2026-09-29
OpenAI has published no product list for DevDay; the official agenda promises only a keynote unveiling "what's next" plus 22 technical sessions.
Strongest hints from session titles:
- Long-running coding agents: a single Codex run of 1,000+ hours keeping builds and tests passing, with goals, subagents, and persistent plans
- Agentic architecture beyond simple loops: durable sessions, hosted sandboxes, subagents (partly live in the Agents API beta)
- ChatGPT Sites + WebMCP: official copy cites "Astra, ChatGPT Sites, and WebMCP" enabling new kinds of interfaces
- Agents that provision their own compute
- Closing Q&A with Sam Altman, Tejal Patwardhan, and Thibault Sottiaux
Fortune, citing people familiar, says OpenAI plans a dozen or more enterprise and consumer products — treat with caution.
More from AGI Musings
- Mocking the ASI superpersuasion hype: catchy jingles and flashy production value — burny_tech · 2026-09-29
- OpenAI internal models show ~15-min 80% time-horizon on real research tasks, far below METR — ben_j_todd · 2026-09-29
- SV builds agents for a 47-tasks-at-once world, while real workers doomscroll out of boredom — IndraVahan · 2026-09-29
- LLM Can't Enforce Hard Boundaries: Keep Final Decisions Outside the Probabilistic Engine — forevergeeks · 2026-09-29
- Cute Personal AI Agents Are the New Dubai Chocolate, Podcast Quips — marlene_zw · 2026-09-29
- Law school applicant boom continues despite "AI will replace lawyers" takes — jkubicki · 2026-09-29