ChatGPT Workspace Agents

OpenAI on April 22 introduced workspace agents in ChatGPT, a new shared automation system for teams that can also run inside Slack. (openai.com) OpenAI said the agents are available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans, with rollout to Business and Enterprise workspaces happening over the next few weeks. The company said teams can create them from the Agents tab in ChatGPT by describing a repeatable workflow in plain English. (openai.com) The product is built on Codex, OpenAI’s coding agent, and OpenAI said workspace agents can run in the cloud, be shared across an organization, operate on schedules, and connect to tools including Google Drive, Google Calendar, Slack, and Microsoft SharePoint. (help.openai.com) OpenAI is positioning the new system as the next step after custom GPTs. Its help center says GPT feature availability now varies by plan and permissions, and that ChatGPT Business, Enterprise, and Edu customers keep GPT-4o inside custom GPTs only until April 3, 2026, when that model is retired across plans. (help.openai.com) That shift comes as OpenAI is separating what customers can do inside ChatGPT from what they can do through the application programming interface, or API. VentureBeat reported that OpenAI told customers model availability will increasingly differ between ChatGPT surfaces and the API, steering enterprise workflow building toward ChatGPT-based agents instead of older GPT wrappers. (venturebeat.com) In plain terms, a workspace agent is a persistent software helper for office tasks: it can watch for a trigger, pull information from connected systems, and send back a draft or action without a person retyping the same instructions each time. OpenAI’s academy materials describe them as tools for repeatable workflows that depend on shared systems, standard handoffs, and timing rules. (openai.com) OpenAI said its own sales team uses one agent to combine call notes and account research, qualify leads, and draft follow-up emails, while support teams can route requests and operations teams can assemble recurring reports. The business landing page says the same framework is aimed at sales, operations, information technology, and customer success teams. (openai.com 1) (openai.com 2) For software teams, OpenAI tied the launch to Codex features that let agents work in isolated worktrees, produce reviewable diffs, and interoperate with Codex in the command line interface and integrated development environment. OpenAI’s release notes say eligible workspaces can also add custom Model Context Protocol servers, which are connectors for outside tools and data sources. (help.openai.com 1) (help.openai.com 2) Slack is the first outside surface OpenAI is emphasizing. OpenAI’s help docs say agents can post results into Slack channels on a schedule, but Slack Enterprise Grid deployments require both a workspace-level connection and organization-level approval before the app can run. (help.openai.com) OpenAI is pitching the system as shared infrastructure rather than a personal chatbot: build an agent once, let a team reuse it in ChatGPT or Slack, then revise it over time as the workflow changes. The company’s release notes say agents can be previewed before publishing, shared within a workspace, and managed with admin controls. (openai.com) (help.openai.com)