OpenAI Agent Builder Shuts Down November 30: Your Migration Checklist

OpenAI Agent Builder Shuts Down November 30: Your Migration Checklist

OpenAI Agent Builder goes dark on November 30, 2026. Any automation you built on it stops working that day unless you act before then.

OpenAI announced the deprecation on June 3, 2026. The official deprecations page confirms the shutdown date. You have 25 days.

What breaks: The visual canvas, any workflow you built in it, and the hosted execution layer all stop. ChatKit remains available, but OpenAI has not confirmed whether existing workflow IDs will continue to execute through ChatKit after the shutdown. Assume they will not.

Two products close on the same date: Evals (read-only as of October 31) and the v1/prompts API with reusable prompt objects. If your stack touches any of these, the same 25-day clock applies.

What You Need to Do Before November 30 : 

The first step is an audit, not a migration. You cannot migrate what you have not documented.

Step 1: Export every workflow now.

Open each workflow in Agent Builder. Select Code at the top, choose Agents SDK, pick TypeScript or Python, and copy the full export into a file in your repository. Do this today. The export lives inside Agent Builder. After November 30, you cannot retrieve it.

OpenAI states directly that this export "does not convert your workflow graph or guarantee that every behavior transfers unchanged." Plan to test everything after migration.

Step 2: Document what the canvas holds.

For each workflow, record: The agent node's instructions and model configuration

• Every tool it uses: MCP servers, vector store IDs, OpenAI connectors, and any function tools with their target services.
• Guardrail settings: which checks run and what happens on failure.
• Approval nodes: what the reviewer sees and approves.
• Logic branches: If/else and While conditions are written in CEL expressions inside the canvas; copy them verbatim.
• State variables and Transform steps.

Step 3: Save test cases.

Run representative inputs through Preview and save the outputs. These become your regression baseline. After migration, you run the same inputs through the new path and compare.

Step 4: Note your ChatKit version and workflow IDs.

If your application calls these workflows via API, write down every workflow ID and the version your code pins. You will need these to map old calls to new ones.

Two Migration Paths: SDK or Workspace Agents

OpenAI documents two routes. Both start from the same export you created in Step 1.

Path 1: OpenAI Agents SDK (code-first)

Paste the exported code into your application, install the matching SDK, and configure your environment. From there you own the full execution stack: deployment, storage, authentication, approvals, and runtime validation. The SDK provides the primitives; your application provides the infrastructure.

Choose this path if:

• You need deterministic behavior: specific branching, exact tool call order, approval workflows that require audit logs.
• You are running this in a production system that your team deploys and monitors
• The workflow has conditional logic or state that needs to be predictable across runs

What to verify after migrating:

• Authentication and permissions in your runtime.
• Tool call behavior against your saved test cases.
• Approval flows, since the SDK leaves those to your application to implement

Path 2: ChatGPT Workspace Agents (platform-side)

Create a Workspace Agent in a ChatGPT Business, Enterprise, or Edu account. Paste the exported code using OpenAI's prompt, review the generated instructions, configure tools and permissions, and test in Preview before creating the agent.

Choose this path if:

• The workflow is simple and conversational in nature.
• Your team already operates inside ChatGPT Business or Enterprise.
• You do not need strict determinism from the workflow.

What to watch for: OpenAI's documentation notes that "workflows with strong determinism at their core may not migrate faithfully to a workspace agent." If your automation depends on exact branching or precise tool call sequences, test this path carefully before committing.

The Actual Risk :

Agent Builder was a hosted orchestration layer sitting above OpenAI's API. Convenient, but the vendor controlled the runtime. When the product gets repositioned, your automation goes with it.

Any automation that matters to how your business runs should sit as close as possible to the underlying API: code you deploy, version, and own. Hosted no-code tools add convenience in exchange for control, and that tradeoff tends to surface on exactly the timeline it did here.

If your team is working through this migration and is not sure which path fits your stack, we can help at zycosoft.com/consultation.

Source: OpenAI Deprecations page, confirmed November 2026. Additional coverage via Yahoo Tech and CamelAI.