← Back to the blog
Product

Your agent tool could disappear. What 30 November teaches about building with AI

OpenAI shuts down Agent Builder on 30 November. For anyone building with AI, the deadline is a good reason to review what sits inside the platform and what stays with the company.

Interlocking blocks in dark lines on a light background, beside the glass form with a blue core

Building an agent by dragging blocks is quick and feels good. The risk appears on the day the tool changes its plans.

In OpenAI's case, that day has a date: 30 November.

What will happen

OpenAI announced in June that it is closing three parts of its platform:

  • Agent Builder, the visual agent builder, shuts down on 30 November.
  • Evals, the testing platform, became read-only on 31 October and goes offline on the same 30 November.
  • Reusable prompts, through the v1/prompts API, also end on that date. The guidance is to move prompts into the application code.

The recommended path is to export the flow to the Agents SDK, in TypeScript or Python, or to recreate the agent in ChatGPT's Workspace Agents. The migration guide warns that the export does not convert the visual flow and does not guarantee that all behaviour will carry over unchanged.

If your company uses these tools, this week's task is simple: export, test and confirm what still works. Three weeks go by quickly.

The problem is not OpenAI

Platforms change, sometimes for good reasons. That applies to any supplier of a visual builder, a subscription chatbot or an app generated in a few clicks. The useful question is a different one: what would be lost if this tool vanished tomorrow?

Comparison between what depends on the platform, such as hosted flows and tests, and what stays with the company, such as code, prompts and test history

What should stay with the company

In an AI product, the accumulated value sits in four places:

  • Prompts and rules, versioned in the code, with a history of who changed what.
  • Tests, with real examples from your business and the expected answer. They show whether a model change has made the agent worse.
  • Integrations, with the ERP, CRM, shop and WhatsApp, written so that the model can be swapped.
  • Data and accounts, such as history, domain and access, in the company's name.

With this in place, changing model or supplier becomes a project of days, not a fresh start.

Is a visual builder still worth it?

For testing an idea, yes. What matters is not to build your operation on top of the prototype without knowing how to get out of it.

If your app or agent started in one of these tools, our review of an AI-built app covers security, payments and performance and delivers the plan to take it to production. In the projects we build, code, domain and accounts are in your company's name.