OpenAI Agent Builder shuts down Nov 30: export to the Agents SDK

If you built a support bot or an intake agent by dragging boxes around in OpenAI Agent Builder, the canvas goes away in four weeks. OpenAI's deprecations page schedules Agent Builder to shut down on November 30, 2026. Two other pieces go the same day: the Evals dashboard and API, and the v1/prompts API for reusable prompts. OpenAI gives you an exit, and it is code.

What actually happened

OpenAI announced the deprecation on June 3, 2026. Its deprecations page lists three shutdowns:

• Agent Builder: scheduled to shut down November 30, 2026.
• Evals: existing evals became read-only on October 31, 2026. The Evals dashboard and API shut down November 30. OpenAI points users to a cookbook guide for moving to Promptfoo.
• Reusable prompts: the v1/prompts API and reusable prompt objects shut down November 30.

OpenAI's migration path is the Agents SDK for workflows that should run as code, or ChatGPT Workspace Agents for staff-facing use. The deprecations page says ChatKit, the embeddable chat UI, "remains available." The Agent Builder guide shows a Code button that downloads your workflow as SDK code to run yourself.

What OpenAI's docs do not say: whether a published workflow keeps answering after November 30. We would plan as if it stops.

Why it matters for your business

A no-code tool is a lease. Five months' notice is fair. It is still five months of someone else's roadmap deciding when your customer-facing bot gets rebuilt. The workflow logic you spent weeks tuning was never in your repo.

Export now, not on the 29th. Click Code, commit the output to a repo, and run it against real traffic in staging. Exported code is a starting point. Check the tool calls, the guardrails and the handoff logic one by one.

Your evals are already frozen. If you used OpenAI Evals to check agent quality, those records went read-only October 31. Export the datasets and grading rules before the API goes dark, then rebuild them in a tool that runs in your own CI. Without evals, you cannot prove the migrated agent behaves like the old one.

Prompts belong in version control. Anything stored as a reusable prompt object on OpenAI's side needs to move into your codebase. That is where it should have lived anyway.

Sources: OpenAI: Deprecations, OpenAI: Agent Builder guide.