OpenAI launches cloud workspace agents for ChatGPT business users

OpenAI began rolling out cloud-based workspace agents that can do more than answer prompts, pushing ChatGPT further into the role of a digital coworker for sales, operations and support teams. The feature entered research preview for Business, Enterprise, Edu and Teachers plans, and OpenAI framed it as an evolution of GPTs rather than a replacement.

The company described the new agents as Codex-powered and designed to run in the cloud, with examples that range from routing product feedback to weekly reporting, qualifying leads, drafting follow-up emails and handling third-party risk management. OpenAI said teams can build an agent once and share it across an organization, then use it in ChatGPT or Slack and improve it over time. That is a sharper workplace pitch than a consumer chatbot: the goal is to automate repeatable work that usually crosses inboxes, calendars, shared drives and messaging apps.

For managers and IT departments, the bigger question is not whether the agent can write a decent summary, but whether it can be trusted to operate inside company systems without creating a new layer of risk. OpenAI said the agents can run on schedules and use connected apps, tools, skills, files and custom MCP servers, while governance features include role-based access control, admin approvals, audit logs, monitoring and centralized management. The help center said admins can enable or disable workspace agents and control access and usage, and that agents can be shared privately, by link or in the workspace directory.

The rollout was not universal. Workspace agents were off by default at launch for ChatGPT Enterprise workspaces, and they were not available to Enterprise workspaces with EKM at launch. That detail will matter to security teams that handle sensitive data and to legal and compliance officers weighing how far autonomous software should be allowed to reach into internal systems.

OpenAI’s own materials showed the intended use case in concrete terms. A sales-meeting-prep agent, built in the developer cookbook, pulled tomorrow’s calendar data, gathered account context from Microsoft SharePoint, searched the web for company news, saved a brief back to SharePoint and sent a summary. In practice, that kind of workflow could save time for knowledge workers while also widening the blast radius if an agent misreads data, sends the wrong note or pulls from a protected file.

Slack is central to that strategy. OpenAI said the company has more than 170 Slack Connect channels and has sent over five million messages since becoming a Slack customer in 2018, underscoring why a workplace agent that lives in chat could spread quickly inside an organization. As rival vendors race to sell enterprise AI automation, OpenAI is betting that the next frontier is not a better chatbot, but a system that can act across the tools where work already happens.