Strategy

Custom GPTs and Agent Builder are closing. What should you do with the tools you built?

Custom GPTs are scheduled to retire on 11 December 2026 and Agent Builder to shut down on 30 November 2026, so any business process running on them needs a decision now: migrate it to OpenAI's replacement, or rebuild it as something you own.

Migrating is quick for a GPT that is mostly instructions and files. If it connects to your systems through custom actions, or a real process depends on it, rebuild it so that you, not the platform, decide when it changes.

What's changing, and when.

OpenAI is retiring custom GPTs and moving them to plugins, which bring instructions, reference files and connected apps together. Separately, it is shutting down Agent Builder. The dates below are OpenAI's scheduled or planned dates, and OpenAI says they can change.

26 October 2026 (planned). Creation of new custom GPTs ends in affected Enterprise workspaces. Existing GPTs can still be edited until they are migrated.

30 November 2026. Agent Builder is scheduled to shut down. OpenAI points users to its Agents SDK or to ChatGPT Workspace Agents. ChatKit remains available.

11 December 2026. Custom GPTs are scheduled to retire across all ChatGPT plans, and their GPT pages become inaccessible. Enterprise workspaces with an approved deferral retire on 11 February 2027.

Until then, existing GPTs keep working. The notice in the account or workspace where a GPT was created is the one that applies to it.

Sources: OpenAI's Custom GPT retirement and migration FAQ and API deprecations page, checked 26 October 2026.

What carries over, and what doesn't.

OpenAI's migration turns a GPT into a plugin, using the GPT's latest published version.

What carries over: the GPT's instructions, which become a skill in the plugin; its knowledge files, copied in as reference files; and its connected apps.

What doesn't: custom actions, the model you selected, the GPT's sharing settings, and drafts or unpublished edits. The plugin starts private, so the people who used the GPT won't have it until you share it.

Custom actions are the catch. They are how a GPT connects to other services, such as your CRM or booking system. Features that depend on them won't work in the plugin until you set up a replacement, and rebuilding one may need a custom MCP server.

Your old conversations stay accessible after retirement. Once you migrate a GPT, the original becomes read-only, so finish any edits first. OpenAI also says a migrated plugin may respond differently, so test it on familiar tasks before anyone relies on it.

Migrate, rebuild or replace: a simple test.

Ask two questions of every GPT and Agent Builder workflow: does it connect to your systems, and does a real process depend on it? Two noes, migrate it. Any yes, rebuild it.

Migrate it

Move it to a plugin, then test it

  • Mostly instructions and files A set prompt, a house style, a library of reference documents.
  • No custom actions Nothing has to be rebuilt before it works again.
  • Helps a person with a task Drafting, summarising or answering questions, with a person doing the work around it.
  • Losing it would be an inconvenience Not a stoppage. If it behaved differently for a week, the business would carry on.

Rebuild it as something you own

Move it into your own systems, on your timetable

  • Uses custom actions It reads from or writes to your systems, and that connection has to be rebuilt anyway.
  • Sits inside a process Quoting, onboarding, reporting or anything else the business runs on every week.
  • Runs on Agent Builder It has to move by 30 November regardless, so move it to something you control.
  • Several people or customers depend on it Access, permissions and records need to sit with you.
  • Needs to behave the same next year Changes happen when you decide, not when a platform does.

If a tool sits partly in both columns, rebuild it: one custom action or one process that depends on it is enough. And if nobody would notice it had gone, don't migrate it at all. Let it retire, or replace it with a ready-made tool that already does the job.

Why tools your business runs on should be yours.

A custom GPT is a quick way to try an idea. The trouble starts when the idea works and a real process grows around it. At that point you're renting part of your operation, and the platform's timetable decides when that process changes.

That's what this retirement shows. Nothing is wrong with the GPTs themselves; the platform has moved on. Platforms do: models are replaced, features move and products close, and each change arrives on someone else's schedule.

Owning a tool doesn't mean avoiding AI platforms. It means the parts that are specific to your business, the instructions, the connections to your systems, the data and the record of what happened, sit where you control them. Then a supplier's change is a piece of work you plan, not a deadline you react to.

How we handle this in a bespoke build.

When a tool has outgrown a custom GPT, we rebuild it as part of your own systems. You own the result: no licences and no vendor lock-in.

  • Your logic in your codebase. The instructions and rules that made the GPT useful move into code you own, not a settings page on someone else's platform.
  • Connections engineered into your systems. Where a custom action used to reach your CRM or database, the integration is built into your own infrastructure, with access limited to what the tool needs.
  • Sign-off and records you control. The points where a person still signs off follow your approval rules, and what the tool does is recorded in your own systems.
  • Tested before anyone switches. The rebuilt tool is checked against the work the old one does today, and the old one stays in use until the new one has proved itself.

See how we approach bespoke AI development.

What to do before 30 November.

  1. List what you rely on. Every custom GPT and Agent Builder workflow in use: who built it, who uses it, and what it's for. Note where each GPT was created, because that account's timeline applies.
  2. Start with custom actions and Agent Builder. They need the most time, and Agent Builder is scheduled to close first.
  3. Run the two-question test. Two noes, migrate. Any yes, rebuild.
  4. Finish edits, then migrate. Publish the changes you need first: drafts don't transfer and the original becomes read-only. Test the plugin on familiar prompts and at least one harder case before you share it.
  5. Plan each rebuild with the people who use the tool. Keep the old tool running until the replacement has proved itself on your own work.

Aim for 30 November rather than 11 December. Agent Builder goes first, and a rebuild needs time to be tested before the tool it replaces closes.

Rebuild what you rely on, on your timetable.

The AI Opportunity Audit scores which tools are worth rebuilding and what you'd own at the end. Fixed price, and credited against your build. It's one way in, not a required first step: if you already know what needs rebuilding, book a call about the project.

FAQs

Frequently asked questions.

When do custom GPTs stop working?

Custom GPTs are scheduled to retire on 11 December 2026 on all ChatGPT plans, when the GPTs and their pages become inaccessible. Enterprise workspaces with an approved deferral have until 11 February 2027. Until then, existing GPTs keep working. OpenAI says the dates can change, so check the notice in your own account or workspace.

Will our custom actions still work after migration?

Not automatically. Custom actions don't transfer through OpenAI's migration workflow, and features that depend on them won't work in the new plugin until you set up a replacement. Rebuilding the connection may need a custom MCP server and some technical work, so allow extra time for any GPT that uses them.

What happens to tools built in Agent Builder?

Agent Builder is scheduled to shut down on 30 November 2026. OpenAI points users to its Agents SDK or to ChatGPT Workspace Agents, and says ChatKit remains available. Either way, a workflow you rely on has to move before the shutdown date, so start with anything a business process depends on.

Should we migrate or rebuild?

Migrate a GPT that is mostly instructions and files, has no custom actions and doesn't sit inside a process the business depends on. Rebuild anything that connects to your systems or that a real process runs on, so you control when it changes. If nobody would miss it, let it retire.