OpenAI launched workspace agents in ChatGPT today, a research preview that the company is positioning as the replacement for custom GPTs. The new agents are powered by Codex and designed to handle complex, long-running workflows on behalf of teams rather than individual users.
The move effectively ends OpenAI's multi-year push to turn custom GPTs into a mainstream platform. Custom GPTs, introduced in 2023 as the headline feature of OpenAI's first developer event, never gained the kind of traction the company once projected. Workspace agents now inherit the slot, but with a firmly enterprise-first design.
What workspace agents actually do
According to OpenAI, workspace agents are shared agents that a team can configure once and then reuse across tools. They run in the cloud and continue working when the user who triggered them is offline — a sharp departure from the largely conversational, in-session nature of custom GPTs.
The company describes the agents as capable of carrying out tasks "people do at work — from preparing reports, to writing code, to responding to messages." They can be invoked from inside ChatGPT or from Slack, and OpenAI says they improve over time as teams collaborate on them.
The underlying engine is Codex, OpenAI's agentic coding and automation stack, which was itself repositioned earlier this month from a developer tool into a broader "computer use" product with an in-app browser and plugin ecosystem. Workspace agents are the first major consumer-facing surface to run on that revamped Codex foundation.
Availability and controls
The research preview is rolling out to ChatGPT Business, Enterprise, Edu, and Teachers customers. OpenAI has not published standalone pricing; the capability is bundled into those existing team tiers. Admins can scope what workspace agents are allowed to access, reflecting the company's broader push to make its tools defensible inside regulated and security-conscious enterprises.
OpenAI has also committed to a migration path: existing custom GPTs will eventually be convertible directly into workspace agents, easing the transition for the long tail of customers that built internal GPTs during the prior push.
Why this matters
The announcement is less about a single feature than about OpenAI's structural bet. Instead of a marketplace of single-turn GPTs aimed at consumers, the company is pushing a shared, persistent, team-scoped agent model aimed at knowledge work — competing head-on with Microsoft Copilot workspace features and a growing field of enterprise agent platforms.
It is also a quiet admission that the original GPT Store experiment has run its course. OpenAI is effectively absorbing the concept into a product surface that sits closer to Codex, closer to the cloud, and much closer to the enterprise workflows where the company now makes most of its revenue.
