Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Agent Builder shuts down Nov 30: export now

Written November 11, 2026

OpenAI's Agent Builder, Evals and reusable prompts all shut down Nov 30. The migration path is export as Agents SDK code. Do it before the 30th.

What closes, and when

OpenAI notified developers on 3 June 2026 that Agent Builder was being deprecated, and its deprecations page lists Agent Builder as scheduled to shut down on 30 November 2026. The same page lists two more shutdowns on the same date: the Evals dashboard and API, and the v1/prompts API with reusable prompt objects. Evals became read-only on 31 October 2026, so an existing eval can no longer be changed, only moved. ChatKit remains available after the shutdown.

The hard part is the date. After 30 November you cannot open the Agent Builder canvas, so you cannot open a workflow to export it. Export while the canvas is still there.

How the export works

OpenAI's migration guide opens with the sentence that matters: use the guide to export an existing Agent Builder workflow as Agents SDK code. The steps are short:

  1. Open your workflow in Agent Builder.
  2. Select Code in the top navigation.
  3. Select Agents SDK in the code dialog.
  4. Select TypeScript or Python, then copy the complete export.

From there the guide names two destinations. The Agents SDK is for building agents through code: copy the export into your application, install and configure the matching Agents SDK, and test the workflow in your runtime. You then own hosting, authentication, the interface, and uptime. ChatGPT Workspace Agents is for building agents through natural language and sharing them with teams, and it needs a ChatGPT Business, Enterprise, or Edu workspace with access to workspace agents and permission to create them. In ChatGPT, create a workspace agent, paste the exported code with a prompt asking for the workflow to be converted into an agent, review what the builder flags as needing changes, and create the agent only after previewing representative inputs against the original workflow's expected behaviour.

The export is not lossless

The migration guide says so up front: the process does not convert your workflow graph and does not guarantee that every behaviour transfers unchanged. Treat the export as a starting point, not a finished move. Review control flow, triggers, tools, and permissions as you test the migrated agent, and validate the application's configuration and behaviour before deploying it.

For the Workspace Agents destination, the guide adds three limits of its own. Workflows with strong determinism at their core may not migrate faithfully to a workspace agent. Connected apps, authentication, publishing, and permission configuration need separate review in ChatGPT. And the same safety practices you used for the workflow still apply, especially when the agent can access private data or take actions through connected tools. A workflow that has not been tested in its new home has not been migrated. It has only been copied.

What else goes on the same date

The two companion shutdowns need their own work, and neither is covered by the Agent Builder export:

There is no hosted replacement on the way. OpenAI's support account said in September that it has no timeline for an equivalent hosted replacement, so whatever the dashboard did for you, editing without a deploy and one-click rollback, needs an owner on your side before the 30th.

Order of work

  1. Inventory every workflow now. List each Agent Builder workflow, including ones owned by former team members or embedded in scheduled jobs.
  2. Export each one as Agents SDK code. Open the workflow, select Code, select Agents SDK, pick TypeScript or Python, and copy the complete export into version control.
  3. Pick a destination per workflow. Code-first work goes to the Agents SDK in your own application. Team-shared work that fits natural-language building goes to Workspace Agents, if your workspace qualifies.
  4. Test with representative inputs. Compare the migrated agent's behaviour with the original workflow's expected behaviour, and review control flow, triggers, tools, and permissions before you create anything.
  5. Cut over before 30 November. After that date the canvas is gone, and the export path goes with it.

Verdict

Export before 30 November. After that the canvas is gone and the export path with it. The export is a starting point, not a finished migration, so budget time to test each workflow in its new home.

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: OpenAI shuts Agent Builder, Evals, and prompts November 30 · November shutdowns: Runloop, Conductor, Moondream · GPT Image 2 shuts down Feb 10, 2027