OpenAI workspace agents

OpenAI has added “workspace agents” to ChatGPT, giving business teams shared AI workers that can keep running after a user logs off. (openai.com) OpenAI said on April 22 that the agents are powered by Codex, run in the cloud, and are built for repeatable work such as writing code, preparing reports, and responding to messages. The company said teams can use them inside ChatGPT and in Slack. (openai.com) The rollout is gradual over the next few weeks to ChatGPT Business and Enterprise workspaces, according to OpenAI’s release notes. Those workspaces can create agents from templates or from scratch, preview them before publishing, share them across a workspace, and schedule runs. (help.openai.com) OpenAI is pitching the product as a successor to custom GPTs for team use, with agents that work inside company permissions and admin controls. Its business page says the agents can connect to tools including Slack, Google Drive, Google Calendar, Microsoft SharePoint, files, skills, and custom Model Context Protocol servers. (openai.com) An AI “agent” is software that can take a goal, use tools, and carry out a chain of steps without waiting for a prompt after every action. OpenAI’s academy page says workspace agents are aimed at jobs that usually require repeated instructions, shared systems, and handoffs between people and apps. (openai.com) The model choice matters because these agents are tied to OpenAI’s newer work stack. OpenAI said in March that GPT-5.4 is its main model for professional work, while GPT-5.4 Pro is the higher-performance option for harder tasks in ChatGPT and the application programming interface. (openai.com) That same GPT-5.4 launch said the model added native computer-use features and up to 1 million tokens of context, which is the amount of text and data a model can keep in view at once. Those features are part of why OpenAI is now describing ChatGPT less as a chat window and more as a place to assign longer jobs. (openai.com) OpenAI’s product newsroom shows the launch as part of a broader April push around Codex and agents, including “Codex for (almost) everything” on April 16 and an Agents software development kit update on April 15. The sequence shows the company bundling its coding model, tool use, and admin controls into workplace products. (openai.com) The immediate test is whether companies trust ChatGPT with routine internal work, not just one-off questions. OpenAI’s launch page says these agents are meant to be built once, shared across a team, and improved over time, which is closer to software deployment than ordinary chatbot use. (openai.com)