OpenAI Agent Builder and Evals shutdown: what to migrate before November 30

OpenAI has scheduled three platform surfaces for retirement: Agent Builder, the Evals platform, and reusable prompt objects. The official deprecations page says Evals become read-only on October 31, 2026, and all three shut down on November 30, 2026. Teams using visual agent workflows, hosted eval runs, prompt optimizer assets, or /v1/prompts should start by exporting the artifacts that would be hard to reconstruct later.

Key takeaways

• Agent Builder is deprecated, but ChatKit remains available and OpenAI points users toward the Agents SDK or ChatGPT Workspace Agents.
• The Evals platform becomes read-only for existing users on October 31, 2026, then the Evals dashboard and API shut down on November 30, 2026.
• Reusable prompt objects and the /v1/prompts API are also scheduled to shut down on November 30, 2026.
• OpenAI's Agent Builder migration guide says exported workflows do not guarantee identical behavior, so parity testing is required before deployment.
• The practical work is inventory, export, ownership, replacement design, and regression testing before late-Q4 change freezes.

Practical LinkLoot angle

This is a migration deadline for agent operations, not just a dashboard cleanup. A visual agent can hide tool permissions, routing decisions, approval logic, prompts, and trace-review habits. An eval project can hide datasets, graders, acceptance thresholds, historical baselines, and prompt optimizer evidence.

Use a simple triage pass before writing code:

For LinkLoot readers, the useful workflow is direct: create an inventory of every Agent Builder workflow, Evals project, reusable prompt object, and production call path. Assign an owner to each item. Then mark it as keep-as-code, recreate-as-workspace-agent, replace-with-eval-runner, move-to-prompt-registry, or retire.

What to verify before you act

Check live usage before migrating. Search repositories and API logs for /v1/prompts, Agent Builder workflow IDs, Evals API calls, prompt optimizer assets, and release checks that depend on OpenAI-hosted eval runs. Export Agent Builder workflows and test the exported TypeScript or Python in the target runtime; OpenAI explicitly warns that the export does not convert the workflow graph or guarantee unchanged behavior.

For Evals, preserve definitions, graders, datasets, run history, and baseline decisions before the October 31 read-only date. For reusable prompts, confirm where prompt text will live, how it will be reviewed, and whether deployments can pin a prompt version. If your team has a Q4 freeze, treat October 15 as the internal completion date rather than waiting for November 30.

Source check

OpenAI's deprecations page confirms the June 3, 2026 announcements, the October 31 read-only date for Evals, the November 30 shutdown date, and the recommended replacement paths. The Agent Builder docs confirm that ChatKit remains available and that Agent Builder is scheduled to shut down. The migration guide confirms the two main paths, Agents SDK and ChatGPT Workspace Agents, and warns that exported workflows still need validation. TheRouter independently summarizes the same timeline and frames the operational impact around evals, reusable prompts, routing, and gateway-side prompt management.

Migration checklist

• List every Agent Builder workflow and export the current version.
• Copy tool definitions, approval rules, connected app requirements, and authentication notes into the new owner repo.
• Export or document Evals datasets, graders, historical baselines, prompt optimizer outputs, and release thresholds.
• Replace /v1/prompts calls with a versioned prompt source that your deployment pipeline can review and roll back.
• Run parity tests with representative inputs before moving any migrated agent into production.

For adjacent implementation patterns, compare this with LinkLoot's guide to AI workflow automation and the broader AI agent tools guide.