OpenAI launches workspace agents for ChatGPT

OpenAI introduced workspace agents in ChatGPT, persistent cloud assistants powered by the Codex model that can run multi-step team workflows, integrate with external apps and operate in Slack.

Users build an agent from a new tab in ChatGPT by describing the desired workflow. The interface helps map steps, connect tools and test the agent. Once active, agents can run on schedules or respond to specific triggers, retain context across projects and complete multi-step tasks without repeated prompts.

The feature is available now in a research preview for ChatGPT Business, Enterprise, Edu and Teachers plans. OpenAI said the preview will remain free through May 6, 2026; after that date the company plans to move to a credit-based pricing model.

Administrators can limit which data and tools an agent may access, require human approval for sensitive actions and monitor for prompt injection and other security risks. OpenAI also described tools to convert existing custom GPTs into workspace agents and to share and improve an agent within a team.

OpenAI wrote, “Workspace agents are an evolution of GPTs. Powered by Codex, they can take on many of the tasks people already do at work-from preparing reports to writing code to responding to messages. They run in the cloud, so they can keep working even when you’re not.”

The company said agents can gather context from designated systems, follow team processes, request approvals when needed and hand off work across tools. Teams can use agents inside ChatGPT or through Slack integrations.

OpenAI reported that its own teams are already using the agents to automate recurring workflows that require shared context and cross-team handoffs. The launch follows broader industry investment in autonomous and agent-style AI systems. Security researchers and enterprise IT teams have raised concerns about prompt injection and unintended data exposure, issues OpenAI cited when describing administrative controls.