OpenAI launches workspace agents in ChatGPT with Slack, Linear, and scheduled actions

• OpenAI launched shared “workspace agents” inside ChatGPT, aimed at long-running team tasks instead of one-off chats, according to OpenAI launch and tool workflow post.
• The new agents can pull context from docs, email, chats, code, and internal systems, then take approved actions such as updating Linear issues, creating docs, or sending messages, as OpenAI says in tool workflow post.
• OpenAI says teams can “build an agent once” and share it across the workspace, with ChatGPT turning a job description into an agent that follows team practices shared agent post.
• Availability starts as a research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans, per preview availability.

OpenAI is positioning this as a workflow surface, not just a chat surface. In OpenAI’s description, the agent can “pull context from docs, email, chats, code, and systems,” then take “approved actions” across connected tools tool workflow post.

• OpenAI’s examples include updating Linear issues, creating documents, and sending messages tool workflow post.
• In Slack, the agent can join a thread, “understand what’s needed,” gather context, help resolve the issue, and update external systems tool workflow post.
• It can keep running “in the background or on a schedule,” extending the interaction beyond a single active session tool workflow post.

Linear’s own post confirms the first-party workflow OpenAI highlighted: workspace agents can read Linear context, create issues, and “move work forward” without a second prompt Linear example. A supporting demo roundup also describes “Skills, Connectors, and Scheduling” as visible configuration surfaces in the product feature roundup.

The build model is prompt-first. OpenAI says teams can “describe the job,” then ChatGPT turns that into a working agent that uses the team’s “best practices,” with suggested use cases including lead qualification, feedback routing, request review, reporting, and vendor research shared agent post.

• The sharing model is workspace-wide rather than personal: “Build an agent once, then share it across teams” shared agent post.
• Access starts in ChatGPT rather than a separate developer console, with a product page linked from OpenAI’s launch post and business-facing details in the preview page.
• OpenAI’s Greg Brockman said the feature sits “on top of a cloud-hosted Codex harness,” can be connected to tools, assigned recurring tasks, and reached from surfaces “like Slack” Codex harness note.

The launch scope is limited for now: OpenAI calls it a “research preview” and restricts access to Business, Enterprise, Edu, and Teachers plans preview availability. That makes this a managed workspace feature first, with approvals, sharing, and scheduled execution bundled into ChatGPT’s existing team tiers rather than a general consumer rollout.