OpenAI has launched Workspace Agents to succeed GPTs, with agents able to autonomously run continuously in the cloud.

According to Dynamic Beating monitoring, OpenAI has launched Workspace Agents in ChatGPT, positioned as an upgrade to GPTs. Workspace Agents are powered by Codex, run in the cloud, and can continue executing tasks even when users are offline, supporting scheduled triggers and Slack deployment, enabling collaborative development and sharing by teams.

Users can click on "Agents" in the ChatGPT sidebar, describe a workflow in natural language, and the system will guide them through building the agent step by step, including defining steps, connecting tools, adding skills, and testing. OpenAI has listed 5 example agents: software procurement approval, product feedback routing, automated weekly report generation, follow-up on potential customers, and third-party vendor risk assessment. The official templates also provide preset skills and tool configurations for finance, sales, marketing, and other scenarios.

The agent has its own workspace for files, code, tools, and memory, allowing users to write code, invoke connected applications, and maintain context across steps. For sensitive operations (such as editing spreadsheets, sending emails, adding calendar events), manual approval can be set up. Workspace Agents also support the Compliance API, allowing administrators to view each agent's configuration, updates, and run records, and suspend agents if necessary.

This feature is available in a research preview format for ChatGPT Business, Enterprise, Edu, and Teachers plans, free until May 6th, after which it will switch to a quota-based billing model. Existing GPTs will be temporarily retained, and in the future, one-click migration to Workspace Agents will be supported.