What Are ChatGPT Workspace Agents?
ChatGPT Workspace Agents are shared AI agents that your entire team can build once, deploy to Slack or any browser, and let run around the clock, even when everyone is offline. Launched by OpenAI on April 22, 2026, they replace Custom GPTs with something far more powerful: scheduled workflows, multi-tool integrations, and the ability to act autonomously across Gmail, Slack, Notion, and Salesforce without anyone at the keyboard.
If you have used Custom GPTs before, think of Workspace Agents as their grown-up sibling. A Custom GPT answers questions. A Workspace Agent completes a job, from start to finish, on a schedule, for your whole team.
Workspace Agents are currently in research preview for ChatGPT Business, Enterprise, Education, and Teachers plans. They are free until May 6, 2026, after which OpenAI moves to credit-based pricing.
What Can a Workspace Agent Actually Do?
A Workspace Agent is not a chatbot. It is a worker that runs jobs. The practical difference matters: a chatbot waits for you to ask. A Workspace Agent runs on a schedule, picks up requests from Slack automatically, and finishes without you being present.
Here is what OpenAI's own documentation confirms Workspace Agents can do today:
--- Monitor a shared Slack channel for client questions, draft a response using your brand voice, and send it to a designated approver before posting publicly.
--- Pull new form submissions from Google Sheets every morning, cross-reference them against your CRM in Salesforce, and send a daily digest to your sales lead at 9 AM.
--- Read new emails in Gmail tagged with a specific label, summarise the key action points, and create tasks in Notion or Atlassian Rovo automatically.
--- Prepare a weekly performance report by pulling data from Google Analytics, formatting it in Docs, and posting a summary in a Slack channel every Friday at 5 PM.
According to OpenAI's April 22 announcement, Workspace Agents are powered by Codex and are designed to operate within an organisation's permissions and controls, which means IT teams can restrict what agents can access before giving teams the green light to build.
How to Build Your First Workspace Agent in Under 20 Minutes
You do not need to know how to code. You need to know what job you want the agent to do. Here is the exact process, based on OpenAI's Workspace Agents Academy documentation:
Step 1: Open the Agent Builder
In ChatGPT, click Agents in the left sidebar. If you do not see this option, confirm your account is on the Business, Enterprise, or Education plan. Click New Agent.
Step 2: Describe the Job in Plain Language
The builder chat opens. Write a description of the task you want automated. Be specific: name the tools it should touch, how often it should run, and what a good result looks like. A vague description produces a vague agent.
Step 3: Review the Generated Workflow
ChatGPT will turn your description into a structured workflow with defined steps. Read through each step. If anything looks wrong or you want to add a constraint, type your change directly in the builder chat. This iterative refinement usually takes 3 to 5 minutes.
Step 4: Connect Your Tools
The builder will prompt you to authorise the integrations the agent needs. Workspace Agents natively support Slack, Gmail, Google Drive, Google Docs, Google Sheets, Google Calendar, SharePoint, Outlook, Salesforce, Notion, and Atlassian Rovo. You grant permissions once; the agent uses them each time it runs.
Step 5: Set a Schedule or Slack Trigger
Choose between two deployment modes: a scheduled run (e.g. every weekday at 8 AM) or a Slack trigger (the agent wakes up when it sees a specific keyword or mention in a channel). Both can be active simultaneously.
Step 6: Test, Then Share
Run the agent manually once to verify the output. If it looks right, share it with your team from the Agents panel. Everyone with access can use it in ChatGPT or call it from Slack with a simple @mention.
Which Tools Do Workspace Agents Integrate With?
As of the April 22, 2026 launch, Workspace Agents connect natively to the following tools without requiring Zapier or any middleware:
--- Communication: Slack (read channels, send messages, respond to threads)
--- Google Workspace: Gmail, Google Drive, Google Docs, Google Sheets, Google Calendar
--- Microsoft: SharePoint, Outlook
--- CRM: Salesforce
--- Project management: Notion, Atlassian Rovo (Jira, Confluence)
The practical implication: if your daily workflow touches any combination of these tools, you can automate that workflow today without writing a single line of code or learning Zapier's interface. This is a meaningful shift for marketers, content teams, operations managers, and project coordinators who live in Google Workspace and Slack.
Additional integrations are expected to expand through 2026, according to OpenAI's product blog. Teams wanting to connect tools not yet in the native list can use the agent's API access layer, though that requires some technical setup.
What Workspace Agents Cannot Do Yet (And Where They Still Fall Short)
This audience already knows that AI hype often outruns reality. Here are the honest limitations you should know before building:
They are not free long-term. Workspace Agents are credit-free only until May 6, 2026. OpenAI has not published the full credit pricing table yet. Budget accordingly, especially for agents that run multiple times per day across large teams.
They cannot browse the live web by default. Unless you explicitly give the agent web browsing access and your plan supports it, the agent works within the data sources you connect. It cannot pull real-time competitor prices or news unless you set up a specific data feed.
Complex multi-step decisions still need human review gates. For tasks where a wrong output could cause a real business problem, such as sending a client email or logging a sales entry, build an approver step into your workflow rather than letting the agent fire autonomously. OpenAI's documentation recommends this pattern explicitly.
Enterprise rollout requires IT pre-approval. Admins must enable Workspace Agents in the ChatGPT admin console before team members can build or use them. If your company has a centralised IT review process, factor in that lead time before planning any workflow launches.
Try This: A Ready-to-Copy Prompt to Build Your First Agent
Most practitioners waste their first attempt by being too vague. Here is a complete, copy-paste-ready prompt you can drop into the Workspace Agent builder to create a weekly summary agent — one of the most immediately useful automations for any content or ops team:
Try This Prompt:
You are a weekly team summary agent. Every Friday at 5 PM Hong Kong time, do the following:
1. Check the #project-updates Slack channel for all messages posted this week.
2. Group the messages by project name and identify the top 3 action items mentioned most frequently.
3. Pull any tasks marked as "done this week" from our Notion workspace (Team Board).
4. Draft a weekly summary in this format: [What we shipped], [What is blocked], [Top 3 priorities for next week].
5. Post the summary to the #weekly-wrap Slack channel with a mention to @team-lead.
Constraints: Do not post unless there are at least 3 project updates in the channel. If fewer than 3, send a draft to the team lead for review instead.
Adjust the channel names, timing, and Notion board to match your setup. The constraint clause at the end is important — it prevents the agent from posting an empty summary on quiet weeks.
Why This Matters for Your Workflow Right Now
Workspace Agents mark a transition from AI as a writing assistant to AI as a working teammate. The practitioners who integrate this now are building an unfair advantage: their teams will be running automated research, reporting, and communication workflows while others are still copy-pasting ChatGPT outputs manually.
The free window closes May 6, 2026. That gives you less than two weeks to explore, build, and validate at least one real workflow with zero credit cost. Start with the smallest, most repetitive task your team does every week. If it involves Slack and Google Workspace, you can probably automate it this afternoon.
懂AI，更懂你 — UD 相伴，AI 不冷。 The technology is here. The question is whether you build the workflow that saves your team three hours per week, or keep doing it by hand.
Ready to Put AI to Work for Your Team?
Knowing the tool is step one. Building a workflow that runs reliably every week is step two. UD's team will walk you through every step — from setting up your first Workspace Agent to designing a full AI-powered operations stack for your team. We'll walk you through every step.