ChatGPT Workspace adds custom agents

OpenAI is trying to turn ChatGPT into something closer to workplace software. The new thing is called workspace agents, and the pitch is simple — instead of prompting a chatbot from scratch every time, teams can build a reusable agent once and let it handle a repeatable workflow. These agents are now in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans, and OpenAI is positioning them as the next step beyond custom GPTs. ### What is a workspace agent? A workspace agent is basically a team version of a custom GPT, but with more power and more structure. It lives inside a ChatGPT workspace, can be shared with other people, and is meant for jobs that happen over and over — lead qualification, meeting prep, weekly reporting, inbox triage, customer follow-up, that kind of thing. OpenAI’s own examples lean hard into sales and operations workflows rather than one-off chat use. ### What changed from custom GPTs? The big change is that custom GPTs were mostly conversational tools. Workspace agents are meant to do work across systems. OpenAI says they can connect to tools like Slack, Google Drive, Google Calendar, SharePoint, and custom MCP servers, then run multi-step tasks using those tools and shared context. That makes them less like a smart prompt wrapper and more like lightweight workflow automation sitting inside ChatGPT. ### Why does the cloud part matter? Because these agents are not tied to a person sitting in front of the chat window. OpenAI describes them as Codex-powered agents that run in the cloud, which means they can keep going on scheduled or triggered work even when the user logs off. That is a pretty meaningful shift. A normal chatbot helps in the moment. A cloud agent starts to look like background office software. ### What can teams actually do with them? OpenAI’s docs and examples show a pretty practical set of use cases. Teams can build agents from templates or from scratch, connect data sources and apps, add files and skills, then share the finished agent privately, by link, or through a workspace directory. Agents can also run on a schedule and show up in connected Slack channels, which matters because a lot of business work already lives there. ### Why is Slack getting so much attention? Because Slack is where a lot of approvals, handoffs, and status updates already happen. OpenAI’s help documentation explicitly says workspace agents can be created, used, shared, and managed in ChatGPT and Slack for Business and Enterprise workspaces. That means the company is not just adding another builder inside ChatGPT — it is trying to put agents where teams already coordinate work. ### What about controls and governance? This is where the enterprise angle becomes real. OpenAI is framing workspace agents around governance and admin controls, and its business materials point to role-based controls and centralized management. Business release notes also show preview-before-publishing, workspace sharing controls, and admin-managed all who builds, shares, and runs the agents. ### What is the pricing catch? The preview is not staying free. Multiple OpenAI-adjacent materials say workspace agents are free until May 6, 2026, after which usage moves to a credit-based model. OpenAI’s public product pages emphasize that Business and Enterprise customers can add credits as needed, which fits the broader idea here: ChatGPT is becoming a metered work platform, not just a seat-based app. ### Bottom line This is really OpenAI making a product category bet. Chat used to be the interface. Now the company wants the interface to be shared agents that sit across Slack, files, calendars, and internal processes. If that sticks, workspace agents will matter less as a flashy AI feature and more as the beginning of ChatGPT turning into everyday enterprise plumbing.