Workspace agents are in research preview and free until May 6, 2026, after which pricing will switch to a credit-based model.

OpenAI has launched Workspace Agents in ChatGPT, a Codex-powered layer for teams to manage repeatable tasks and long-running workflows across shared tools and approvals, not just respond to prompts.

These agents create a shared, cloud-based agent that can be reused across an organization and continuously improved, moving beyond single-user assistants.

Approval gates can require human review for sensitive actions like sending messages or updating records.

Key capabilities include scheduled execution, integration with external apps (CRM, IT tickets, Slack), cloud-based code writing/execution, cross-project memory, and configurable human approvals for sensitive actions.

Concrete internal-use cases include software request evaluation, IT ticket creation, product feedback routing with weekly summaries, monthly accounting closes, sales lead qualification with CRM updates, and supplier risk reporting.

OpenAI provides internal production examples such as a Software Reviewer for IT ticketing, a Product Feedback Router, a sales lead qualification agent, and an accounting agent for monthly closes.

Supported use cases include software request triage, weekly reporting, lead outreach, product feedback routing, and vendor risk reviews.

Enterprise admins get role-based controls to manage who can create/share agents and which tools they can access, with sensitive actions requiring manual approval.

Admins can set permissions, approval checkpoints, and monitoring to ensure automation is visible and controllable across workflows.

Admins can control tools/actions, require approvals for steps (e.g., editing spreadsheets, sending emails, creating calendar events), and monitor configurations/runs via a Compliance API; enterprise RBAC and workspace analytics are available.

Audit logs and monitoring track agent activity and workflow outcomes to understand behavior.