OpenAI is rolling out Workspace Agents in ChatGPT to enable collaborative, team-based autonomous workflows across organizations.

Organizations can create agents by describing workflows in ChatGPT, then develop, test, and deploy them, with agents shareable across teams and integrable into platforms like Slack.

Pricing remains the standard API pricing, with sandbox execution included in model API costs and hosted sandboxes billed at standard rates.

Enterprise controls are central: admins manage data access and tool usage, approvals, and monitoring, while a compliance API supports cross-organizational oversight.

Use cases highlighted include automating software request reviews, routing product feedback, generating weekly reports, qualifying leads, and handling third-party risk assessments.

Custom GPTs for business accounts are being deprecated, with a shutdown date of August 26, 2026.

New Codex-style filesystem tools (read_file, write_file, list_directory, run_command, edit_file, search_files, git_* tools) are provided as SDK primitives for autonomous code tasks and file operations.

Harness-compute separation decouples control from execution, enabling remote compute, scalable execution, auditability, and multi-agent coordination.

The update emphasizes safer, more capable agents and easier production deployment, with guidance to upgrade existing integrations and start with sandbox examples for new builds.

SDK now supports 100+ non-OpenAI models via the Chat Completions API, including Claude, Gemini, Llama, with Python support now and TypeScript coming later.

Custom MCP servers enable connections to proprietary systems for AI automation across enterprise environments.

Microsoft 365 integrations are in development; SharePoint is available at launch, with Teams, Outlook, Word, and Excel not yet supported.