Expanding the capabilities of ChatGPT, OpenAI has unveiled ‘workspace agents,’ a feature designed to automate routine tasks for businesses even when employees are not online. This development was announced on Wednesday, with OpenAI highlighting that these agents differ from previous custom GPTs as they leverage OpenAI’s Codex model. The new agents function as persistent assistants capable of integrating with external applications, maintaining information across various projects, and executing multi-step workflows autonomously.
“Workspace agents represent a progression in the evolution of GPTs. Utilizing Codex, these tools can handle numerous tasks that employees typically perform—such as drafting reports, coding, or managing messages,” stated OpenAI. “Operated in the cloud, they continue functioning offline and are crafted to be utilized collectively within organizations, enabling teams to develop an agent once for use across platforms like ChatGPT or Slack, with continuous enhancements over time.”
OpenAI explains that users can establish an AI agent via a new tab in ChatGPT by outlining a preferred workflow. Following this, ChatGPT assists in mapping the process, linking necessary tools, and testing the agent. Once operational, these agents can execute tasks according to set schedules or specific triggers.
“While AI has already expedited individual work processes, many critical organizational workflows rely on shared context, transitions, and collaborative decision-making,” OpenAI explained in their statement. “Workspace agents are tailored for such activities: they can extract relevant information from appropriate systems, adhere to team protocols, request approvals when necessary, and facilitate workflow continuity across various tools.”
This innovation emerges amid a heavily funded phase of agentic AI development, with tech behemoths like Google, Microsoft, and Amazon investing heavily in autonomous systems. Amid ongoing concerns about cybersecurity threats such as prompt injection, OpenAI noted that organizations can restrict data access for these agents, mandate human approval for sensitive actions, and monitor for potential prompt injection attacks.
Available now in research preview mode for ChatGPT Business, Enterprise, Edu, and Teachers plans, workspace agents will remain free until May 6, 2026, before transitioning to a credit-based pricing model. OpenAI has already integrated this technology within its teams but reassured that GPTs will continue to be accessible, ensuring easy conversion of GPTs into workspace agents.
As of now, OpenAI has not responded to Decrypt’s request for further comment.