Is OpenAI Agent Builder being discontinued?

Yes. OpenAI added a notice dated 3 June 2026 to its AgentKit announcement: it is winding down the Agent Builder and Evals products, and from 30 November 2026 they will no longer be available on the OpenAI platform. OpenAI recommends the Agents SDK for workflows that should continue as code, and Workspace Agents in ChatGPT for natural-language use cases.

What happens to my Agent Builder workflows after 30 November 2026?

They stop being available on the platform. Export or document everything before then: each workflow's steps, prompts, tools, guardrails and the variables passed between nodes. The underlying model and Responses API keep working, so the logic can be rebuilt in code; the visual workflow itself cannot be carried over as is.

What is still available from AgentKit?

Reporting on the wind-down says ChatKit, the embeddable chat interface, remains available, and the Responses API is unaffected. Reusable prompts (the v1/prompts API) are reported to share the 30 November shutdown date. Evals becomes read-only for existing users on 31 October 2026, so export datasets and configurations before then.

What should I migrate Agent Builder workflows to?

For a workflow your business depends on, code: OpenAI's Agents SDK, or another framework or provider, versioned and tested like the rest of your software. For simple internal assistants, Workspace Agents in ChatGPT. For multi-step automations across apps, a workflow tool such as n8n can fit. Choose by how much the workflow matters, not by what is closest to the old canvas.

How much does it cost to migrate an Agent Builder workflow?

Mostly the time to rebuild and test it. At Reflekt Lab, scoping one workflow costs $2,300 to $4,700, and a first workflow rebuilt and running in production starts at $17,500, with running costs from about $470 a month. Model usage is billed separately and usually small.

OpenAI Agent Builder Shutdown (Nov 30, 2026): Migration Guide

A visual workflow canvas with fading nodes next to a calendar, while a glowing path carries the same nodes into a solid code machine

OpenAI is shutting down Agent Builder on 30 November 2026. The visual workflow canvas it launched with AgentKit, and the Evals product, "will no longer be available on the OpenAI platform" from that date, according to a notice dated 3 June 2026 on OpenAI's AgentKit announcement. Evals goes read-only on 31 October. OpenAI points builders to the Agents SDK for workflows that should continue as code, and to Workspace Agents in ChatGPT for natural-language use cases. The models and the Responses API keep working, so nothing you built is lost in principle, but the canvas itself can't be carried over. If your team runs anything on Agent Builder, you have about six weeks to document it, choose where it goes, and rebuild it. Below: what stops, what stays, your options, and a migration plan.

What stops, and when

From OpenAI's notice and the coverage of the wind-down by The Rundown, TheRouter.ai and Developers Digest:

What	When	Status
Evals	31 October 2026	Read-only for existing users; export datasets and configs
Agent Builder	30 November 2026	No longer available on the platform
Evals	30 November 2026	No longer available
Reusable prompts (v1/prompts API)	30 November 2026	Reported to share the shutdown date
ChatKit	—	Reported to remain available
Responses API, models	—	Unaffected

Check OpenAI's deprecations page for your own account before planning around a date.

Your three options

1. Rebuild in code, with the Agents SDK or another framework. The right answer for any workflow your business depends on: it becomes versioned, testable and reviewable like the rest of your software, and it no longer lives in a canvas a vendor can retire. It also frees you to choose the model per step, from OpenAI or another provider.

2. Move simple assistants to Workspace Agents in ChatGPT. Fine for internal helpers driven by natural language, where your team already works in ChatGPT and nothing critical depends on exact behaviour.

3. Move multi-app automations to a workflow tool. If the Agent Builder workflow was really "when X happens in one app, do Y in another", a tool like n8n may fit better than an agent. We compared the options in n8n vs Zapier in 2026.

One more option is worth naming: Anthropic's Claude Managed Agents, in public beta since April, bills tokens plus a small runtime fee per session-hour. Whatever you pick, the lesson of this shutdown is the same: the workflow logic should be yours, in code or in a portable form, not only in a vendor's canvas.

A migration plan for the next six weeks

Inventory now. List every Agent Builder workflow, who uses it, how often, and what breaks if it stops.
Export before 31 October. Evals datasets and configurations first, because Evals goes read-only then.
Document each workflow as text: every node, prompt, tool, guardrail, condition and variable. This document is the specification for the rebuild.
Sort by importance. Business-critical workflows go to code. Simple assistants go to Workspace Agents. Cross-app automations go to a workflow tool.
Rebuild the critical ones with tests on real inputs from the last month, and run old and new side by side before the cut-off.
Add what the canvas didn't have: error alerts, retries that stop on errors that can't clear, a review step where mistakes are expensive, and logging.

What we do

We build AI workflows and agents in code and run them in production: a model chosen per task, errors that stop and alert instead of looping, a person in the loop where it matters, and changes users can see and undo. Moving an Agent Builder workflow is that job with a head start, because the canvas already documents what the workflow should do. Our prices are published on the AI integration page: scoping one workflow $2,300 to $4,700, a first workflow in production from $17,500, running it from about $470 a month.

OpenAI Agent Builder Is Shutting Down on 30 November 2026: Where to Move Your Agents

Agent Builder, the visual canvas OpenAI launched with AgentKit, leaves the platform on 30 November. Evals goes read-only on 31 October. If your team built workflows on it, you have weeks, not months. Here is what changes, your options, and how to move.