OpenAI launches Workplace Agents in ChatGPT Business

OpenAI launched workspace agents in ChatGPT, a research-preview feature letting teams on Business, Enterprise, Edu and Teachers plans build, share and manage persistent AI agents for repeatable workflows. OpenAI described it as an evolution of custom GPTs, which it said would eventually be phased out in favour of the new agents, though GPTs remained available alongside them for the time being.

Unlike a custom GPT, a workspace agent runs on OpenAI’s Codex coding model, executes in the cloud rather than only within an active chat session, and can keep working after a user has moved on. Agents could be connected to external tools, retain information across projects and multi-step tasks, and be triggered from Slack, on a schedule, or via API, with teams able to test an agent before publishing it and set its underlying model and reasoning effort. OpenAI pitched the underlying capability, built on Codex, as extending beyond coding into reporting and messaging tasks across a workplace.

The feature launched free of charge, with OpenAI saying it would move to credit-based pricing from 6 May 2026. It arrived the same week as GPT-5.5 and ChatGPT for Clinicians, part of a cluster of OpenAI product launches aimed at enterprise and vertical use cases rather than the consumer chat product, and a step toward retiring the GPT Store model that had anchored OpenAI’s platform strategy since 2023.