ChatGPT Workspace Agents
OpenAI
Codex-powered shared agents for repeatable team workflows in ChatGPT and Slack.
Overview
Freshness note: AI products change rapidly. This profile is a point-in-time snapshot last verified on April 23, 2026.
ChatGPT Workspace Agents are OpenAI’s new shared-agent layer for teams, not just a renamed version of individual ChatGPT agent mode. OpenAI positions them as Codex-powered agents that live in the cloud, work across tools, and can be shared across an organization so teams can reuse the same workflow instead of rebuilding it in every chat thread.
The practical distinction matters. Standard ChatGPT agent mode is the personal “do this task for me” lane. Workspace Agents are the team-operating lane: shared context, reusable instructions, Slack deployment, approval rules, analytics, and admin controls for repeatable work.
Key Features
The strongest feature is that the agent is built around an actual recurring workflow instead of a one-off conversation. OpenAI’s launch examples are specific: software-request triage, product-feedback routing, weekly metrics reporting, lead outreach, and third-party risk screening. The product can connect tools, run in the cloud, ask for approval on sensitive actions, and keep working when the user is away.
OpenAI also made the governance story much more explicit than with older GPT-style customization. Agents can be shared in ChatGPT or Slack, analytics show how they are used, and admins get controls over who can build, share, or use them plus visibility through the Compliance API. That is a much more operational posture than “make a custom GPT and hope people use it correctly.”
Strengths
Workspace Agents are strong when the team already knows the workflow and wants to turn it into a reusable asset. Shared context and approvals are the real value here. Instead of every operator improvising a slightly different brief or handoff, the agent becomes a repeatable operating pattern with guardrails.
They also look especially useful for Slack-heavy teams. OpenAI is clearly leaning into “meet people where the work already happens” rather than forcing every request back into the ChatGPT UI.
Limitations
This is still a research-preview product. Availability is limited to Business, Enterprise, Edu, and Teachers plans, Enterprise admins must enable it, and OpenAI’s help docs note launch constraints such as no support for Enterprise workspaces with EKM.
It is also easy to overestimate what this replaces. Workspace Agents reduce coordination overhead, but they do not remove the need for a system of record, human approval on sensitive writes, or explicit scoping of what the agent can see and do.
Practical Tips
Start with one workflow that already has a clear owner, a clear output, and a clear approval step. Good first candidates are weekly reporting, lead qualification drafts, product-feedback routing, and internal request triage.
Keep permissions narrow and separate draft generation from official writeback. If the agent is allowed to send email, edit sheets, or update records, require approval first. Also decide early whether the primary surface is ChatGPT or Slack so the usage pattern is consistent instead of split across both.
Verdict
ChatGPT Workspace Agents are one of the more important 2026 shifts in OpenAI’s product stack because they turn ChatGPT from an individual assistant into a shared workflow surface for teams. They are best used for repeatable, approval-aware work where the team wants one governed agent instead of many personal prompt habits.