OpenAI Agent Builder shutdown: timeline and how to migrate

The OpenAI Agent Builder shutdown is set for 30 November 2026: the visual workflow editor launched with AgentKit in October 2025 goes away that day. OpenAI announced the deprecation on 3 June 2026, together with its Evals platform. If you have workflows in Agent Builder, export them now: OpenAI’s migration path is either code with the Agents SDK, or a ChatGPT workspace agent. ChatKit, the embeddable chat interface, stays.

The timeline

Date	What happens
6 October 2025	Agent Builder, ChatKit and new Evals features launch at OpenAI DevDay
3 June 2026	OpenAI announces the deprecation of Agent Builder and the Evals platform
31 October 2026	Existing evals become read-only (already in effect)
30 November 2026	Agent Builder and the Evals dashboard and API are scheduled to shut down

What was Agent Builder?

A visual canvas for multi-step agent workflows. You dragged nodes onto a canvas (agents, file search, MCP tools, guardrails, loops, approval steps), connected them with typed inputs and outputs, and previewed runs with live data. A published workflow got an ID and versions, and you either embedded it with ChatKit or downloaded it as Agents SDK code.

What stops working?

Agent Builder itself, on 30 November. Existing users can keep using it until then.

If you embedded a workflow with ChatKit, you can keep using that hosted workflow during the transition. OpenAI’s docs tell new ChatKit apps, and anyone migrating, to connect ChatKit to their own server-side agent instead. They don’t spell out that hosted workflow IDs stop answering on 30 November, but with Agent Builder gone, plan for it.

How do I export a workflow?

Agent Builder can export any workflow as Agents SDK code:

Open the workflow in Agent Builder.
Select Code in the top navigation.
Select Agents SDK.
Choose TypeScript or Python, and copy the complete export.

Do this for every workflow before the shutdown, even if you don’t migrate yet. The export is the only portable copy.

OpenAI is explicit that the export doesn’t convert your workflow graph and doesn’t guarantee that every behavior carries over. Test it before you switch traffic.

Option 1: the Agents SDK

Code you run yourself. Install it:

pip install openai-agents

or for TypeScript:

npm install @openai/agents zod

Paste your export into your project, set OPENAI_API_KEY, and run it. The smallest possible agent, for orientation:

import asyncio

from agents import Agent, Runner

agent = Agent(
    name="History tutor",
    instructions="You answer history questions clearly and concisely.",
)


async def main() -> None:
    result = await Runner.run(agent, "When did the Roman Empire fall?")
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())

Tools become decorated functions, and the routing between specialist agents you drew on the canvas becomes handoffs. You now own hosting, authentication and permissions, which Agent Builder used to handle.

Option 2: ChatGPT workspace agents

For teams that would rather describe an agent in plain language than maintain code. You need a ChatGPT Business, Enterprise or Edu workspace with permission to create agents. Open the agent studio, paste your export with “Please help me convert this workflow into an agent:”, review what the builder says needs changing, set up apps and permissions, and test with real inputs before you create it.

The determinism trap: OpenAI warns that workflows with strong determinism at their core may not migrate faithfully to a workspace agent. If your workflow is a fixed sequence that must run the same way every time, take the Agents SDK route.

What about Evals?

The Evals platform is read-only since 31 October and shuts down on 30 November. OpenAI recommends Promptfoo, an open-source tool for evaluating and red-teaming AI apps. There is no export: you recreate prompts, providers, test cases and assertions in a promptfooconfig.yaml, then run:

promptfoo eval -c promptfooconfig.yaml --no-cache
promptfoo view

Your past results stay behind in OpenAI’s read-only Evals until the shutdown. Save what you need.

Why are visual agent builders going away?

OpenAI isn’t alone. In 2026 AWS closed its Bedrock Agents to new customers and Microsoft is retiring the visual workflows in Foundry, while code-based agents keep growing. See GenAI app builders for the full picture, and Foundry’s declarative workflows for our own experience building a workflow engine nobody wanted.

GenAI app builders: what each cloud offers after 2026’s changes
Bedrock Agents Classic: the AWS side of the same shift
What is an AI coding agent?: the agent loop the Agents SDK runs