OpenAI workspace agents

- OpenAI turned custom GPTs into persistent, always-on workspace agents that automate team workflows rather than just answer questions. - They run 24/7, can orchestrate cross-tool tasks, and are reportedly powered by Codex as a background execution layer. - The product shift frames agents as durable processes instead of single queries, signalling persistence and orchestration are becoming core product features (the-decoder.com) (testingcatalog.com) (help.openai.com)

OpenAI has started rolling out workspace agents in ChatGPT, turning shared custom assistants into tools that can keep working after the chat window closes. (openai.com) OpenAI said on April 22 that the new agents are powered by Codex, run in the cloud, and can handle “complex tasks and long-running workflows” for teams inside the permissions set by an organization. The company said they are an evolution of GPTs, which previously lived mostly as one-off custom chatbots. (openai.com) In OpenAI’s help documentation, the company says workspace agents can be created from templates or from scratch, shared across a workspace, previewed before publishing, and scheduled to run. OpenAI also says they can work inside ChatGPT or Slack and connect to apps including Google Drive, Google Calendar, Slack, and SharePoint. (help.openai.com) The shift is from answering prompts to handling repeatable office tasks. OpenAI’s Academy materials describe the target jobs as work people now do by re-explaining steps, copying information between tools, and repeating the same handoffs across systems. (openai.com) That puts the product closer to workflow software than a chatbot. OpenAI says teams can build one agent and reuse it together over time, while the agent keeps operating within shared files, tools, and organizational controls instead of a single user’s session. (openai.com) The rollout is not universal. OpenAI’s enterprise release notes say workspace agents are rolling out gradually over the next few weeks to ChatGPT Business and Enterprise workspaces, while a separate help article says the feature is off by default at launch for Enterprise admins and is not available to Enterprise customers using enterprise key management, or EKM. (help.openai.com (help.openai.com)) OpenAI is also widening what those agents can plug into. The company’s help pages say builders can add skills, files, and custom Model Context Protocol, or MCP, servers, a standard that lets models pull tools and data from outside services in a more structured way. (help.openai.com) Outside reporting has filled in some of the mechanics. The Decoder reported that the agents keep persistent context, can request approvals, and are being introduced as a research preview for Business, Enterprise, Edu, and Teachers plans, while TestingCatalog and 9to5Mac both described them as always-on and tied to Codex in the background. (the-decoder.com (9to5mac.com)) OpenAI has been moving toward action-taking systems for months, including ChatGPT agent features for consumers and a steady expansion of Codex. Workspace agents package that direction for companies that want scheduled, shared, admin-controlled automation inside the tools employees already use. (help.openai.com (openai.com)) The immediate test is whether teams treat these agents like smarter chatbots or like software coworkers with a job description, a schedule, and a permissions map. OpenAI’s launch materials are written for the second case. (openai.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.