OpenAI is retiring Custom GPTs; migration tools open Sept 17
Every Custom GPT stops running Dec 11, and you can't make new ones after Sept 25. Also: ChatGPT ends auto-reasoning on Plus/Pro, and AWS ships an agent inbox.
Copy markdown
The timeline that bites
OpenAI's built-in migration tools go live Sept 17, new Custom GPT creation ends Sept 25, and every GPT stops running Dec 11. Enterprise admins were notified Sept 11; other plans follow the same schedule.
Migration isn't one button — actions rebuild by hand
The tool converts a GPT's instructions into plugin "skills," but custom actions don't transfer and must be rebuilt manually. Once migrated, the original GPT goes read-only until retirement.
Plugins are the replacement
OpenAI is folding Custom GPTs into Plugins, which bundle reusable instructions with connected apps. If you published a GPT, you re-share it as a plugin and users reinstall it with the right workspace permissions.
Also gone: auto-upgrade to Thinking on Plus and Pro
OpenAI discontinued automatic Instant-to-Thinking switching for Plus and Pro globally and pulled the "Higher intelligence" web setting. You now pick the reasoning model yourself; it only auto-switches for safety.
Build this weekend: an inbox for background agents
AWS open-sourced Pizza Bot (Apache 2.0), a local-first, email-style inbox for long-running agents across macOS, Windows, Linux, browser, and terminal — a ready harness for triaging async agent output instead of tailing logs.
Elsewhere: Vercel's AI SDK harness gets native subscription auth
Vercel's AI SDK harness now supports native subscription authentication, keeping credentials at the host boundary — handy if you're wiring subscription-gated model access into an agent instead of juggling raw keys.