OpenAI just added autonomous workspace agents to ChatGPT - bots that business users can build and deploy to handle tasks on their own, without someone manually prompting them each time.
The feature is live now for Business, Enterprise, Edu, and Teachers plan subscribers. OpenAI gave two concrete examples in its blog post: an agent that searches the web for product feedback and sends a summary report to Slack, and a sales agent that handles prospecting tasks independently. These aren't one-off chat responses - they're designed to run as recurring, scheduled workflows.
This changes how teams interact with ChatGPT. The standard model is reactive - you type, it responds. Workspace agents flip that: you configure what you want done, and the agent handles execution on its own schedule.
What Agents Replace
For a small marketing team, that could mean an agent pulling competitor mentions from the web every morning and dropping a summary into Slack - work that previously required either someone doing it manually or a custom automation built in a tool like Zapier. The difference is you describe the task in plain language rather than mapping out triggers and conditions manually. The agent handles the interpretation.
Access and Permissions
This is a business-tier feature only. Standard Plus subscribers don't get it. The full range of external app integrations isn't fully documented yet, but the Slack example confirms connectivity outside ChatGPT itself.
The permission question matters more than it might seem. An agent with standing access to your company's Slack or email has real reach, and it's not yet clear how granular the permission controls or revocation options are. That's a practical concern worth resolving before deploying any of these in a live workflow.
OpenAI's distribution advantage is real. Enterprise teams already use ChatGPT daily, which means these agents get real-world testing immediately - no separate onboarding, no new login required. Thousands of business teams will be running live workflows within weeks of launch, which will answer the reliability question faster than most product launches.