OpenAI announced this week that its enterprise agent platform, previously gated behind a limited preview, is now generally available to every ChatGPT Enterprise and Team customer. Administrators gain a connector catalog with granular scopes for systems such as Microsoft 365, Google Workspace, Salesforce, ServiceNow and internal HTTP APIs.

The release introduces persistent agent runs that can span hours, with checkpointing and human approval gates for sensitive actions such as sending email, posting to chat or executing payments. OpenAI is also exposing the same agent runtime through its API, so internal developer teams can build domain-specific agents on the same plumbing.

For IT leaders the practical change is governance. Agent activity now appears in the same audit log as user activity, and data residency selections made for ChatGPT Enterprise automatically apply to agent traffic. Pricing is metered per task-minute on top of the existing seat licence.