OpenAI’s AgentKit Launches as It Quietly Sunsets Older Tools

OpenAI just dropped AgentKit, a comprehensive suite for building agents, and it comes with a quiet admission: the company is already killing off two of its earlier agent products. The new stack includes a visual Agent Builder, a Connector Registry for enterprise data, ChatKit for UI, and upgraded evals. But the same day, OpenAI updated a support page to confirm that its original Agent Builder and Evals tools will be completely shut down by November 30, 2026.

It’s a remarkably fast product lifecycle, even for AI. The move signals that the experimental phase of agent development is over. The original tools were essentially placeholders while the company figured out what developers actually needed. AgentKit’s visual canvas is the centerpiece here—a drag-and-drop interface for multi-agent workflows that Ramp used to build a buyer agent in hours instead of months. The team claimed it slashed iteration cycles by 70%. LY Corporation, a Japanese tech firm, spun up a working multi-agent workflow in under two hours.

The enterprise play is obvious. The Connector Registry gives admins a single pane of glass to manage data sources like Google Drive, SharePoint, and Microsoft Teams across ChatGPT and the API. That’s the kind of governance IT departments demand before letting agents anywhere near company data. There’s also an open-source safety layer called Guardrails that can mask PII and detect jailbreak attempts, deployable via Python or JavaScript.

I’m genuinely curious how long this iteration survives. AgentKit feels more coherent than the piecemeal tools it replaces, but the quick sunset of the previous products raises a pragmatic question for enterprise teams: if you build on this, will you be migrating again in 18 months? The new evals capabilities—datasets, trace grading, automated prompt optimization—might be the stickiest part of the package, since reliable testing is what separates demo agents from production ones. HubSpot is already using ChatKit for customer support, and Canva claims it saved two weeks of development time embedding a support agent in under an hour. The speed gains are real, but so is the churn underneath them.