OpenAI Agent Builder Is Shutting Down: What to Do

OpenAI Agent Builder is scheduled to shut down on November 30, 2026, so export your workflows now. OpenAI gave notice on June 3, 2026, and its deprecations page lists two replacements, the Agents SDK and ChatGPT Workspace Agents.

As of 9 November 2026, that leaves about three weeks. This guide is part of our AI agents for business hub, and it was checked against OpenAI's documentation on that date.

What is shutting down, and when

Agent Builder is OpenAI's visual canvas for building multi-step agent workflows. The deprecation covers the builder itself and has a single date.

All of these dates come from OpenAI's deprecations page. Check it again before acting, because schedules can move.

Choosing your path

OpenAI's migration guide offers two destinations. They suit different kinds of teams.

The Workspace Agents caution matters. OpenAI says workflows with strong determinism at their core may not migrate faithfully. If your workflow must follow exact steps, such as approvals or data handling, the SDK is the safer starting point.

If you are also reconsidering your stack, our AI agent frameworks comparison and build versus buy guide cover the wider choice.

How to export a workflow

OpenAI documents the export in four steps:

1. Open the 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.

Save each export somewhere your team controls, along with notes on what the workflow does. Do this even if you have not chosen a destination, since exporting is quick and the deadline is fixed.

OpenAI is plain about the limits. It says this process does not convert your workflow graph or guarantee that every behavior transfers unchanged. The export is a starting point that still needs testing.

If you use ChatKit

ChatKit stays, but one way of using it depends on Agent Builder. OpenAI's ChatKit guide describes a hosted setup where your server creates a session tied to an Agent Builder workflow ID.

The guide says to use that path only if you already have an Agent Builder workflow behind your ChatKit implementation. For new work or migration planning, it points to advanced ChatKit integrations with your own server-side agent.

The page does not say what happens to hosted workflow IDs after shutdown. Treat any hosted ChatKit setup as something that needs a new backend before November 30.

If you use OpenAI Evals

Evals are on the same clock, and existing evals were scheduled to become read-only on October 31. OpenAI recommends moving your evaluation workflows to Promptfoo.

There is no export step for this one. OpenAI's guide says you recreate each eval by hand and check the new results against the original. It specifically says to validate recreated graders, including LLM-as-a-judge graders, before relying on them.

Copy your test cases and grading criteria into your own files now. If you have never set up evaluation for an agent, our guide on how to evaluate an AI agent explains the basics.

A three-week plan

This is our suggested order. Adjust it to the number of workflows you run.

This week

• List every workflow, its owner and whether it is in production.
• Export all of them to Agents SDK code.
• Copy your eval cases and grading criteria out of the Evals product.

Next week

• Choose a destination for each workflow.
• Rebuild the highest-risk workflow first, and test it with real examples.
• Configure tools, authentication and permissions again rather than assuming they carried over.

The week before November 30

• Compare old and new outputs side by side.
• Move production traffic and hosted ChatKit integrations.
• Keep a human approval step for anything that can spend money or change records.

OpenAI's guide tells you to apply the same safety practices as the original workflow, especially when the agent can reach private data. Our notes on human-in-the-loop agents cover where approvals belong.

The wider lesson

The Assistants API had a removal date of August 2026, and Agent Builder follows a few months later. Vendor-hosted tools can change on a timeline you do not control.

That does not mean you should avoid them. It does mean keeping your prompts, test cases and business rules somewhere you own, so a move like this takes days rather than months.

Next step

If you need help moving workflows before the deadline, see our AI agent development service. We can review what you have, recommend a destination for each workflow and help rebuild the ones that matter.

Frequently asked questions