OpenAI’s new Workspace Agents live inside Slack, Gmail, and your calendar. They don’t wait for prompts. They just work.
For the past two years, “AI agent” has been one of the most overused phrases in tech. Everyone promised autonomous AI that would do real work. Most of what shipped was a smarter chatbot with a fancier interface.
That’s changing now.
On April 22, 2026, OpenAI launched Workspace Agents in ChatGPT — and this time, the demo actually matches the pitch. These aren’t agents you open in a separate tab and prompt manually. They live inside Slack. They read your Gmail. They check your calendar. You tag them like a colleague. They do the work and come back with output.
This guide breaks down exactly what Workspace Agents are, how to build one from scratch, what they can actually do in the real world, and how they compare to what Claude and other competitors have shipped.
What Are ChatGPT Workspace Agents?
Workspace Agents are an evolution of OpenAI’s earlier Custom GPTs. Powered by Codex, they can take on many of the tasks people already do at work — from preparing reports to writing code to responding to messages. They run in the cloud, so they can keep working even when you’re not logged in.
The key shift from a chatbot to an agent comes down to one word: autonomy.
A chatbot waits. You prompt it, it responds, you copy the output somewhere, and the loop ends. An agent is different — instead of treating GPT as a chatbot, you wire it up as an autonomous decision-maker that can read your Gmail inbox, post to Slack, create Linear tickets, and move between those systems based on what it finds. It acts, not just responds.
Called “Workspace Agents,” OpenAI’s new offering allows users on its ChatGPT Business and Enterprise plans to design or select from pre-existing agent templates that can take on work tasks across third-party apps and data sources including Slack, Google Drive, Microsoft apps, Salesforce, Notion, Atlassian, and other popular enterprise applications.
How to Build Your First Workspace Agent (Step-by-Step)
Here’s exactly how to go from zero to a working agent — no coding required.
Step 1: Find the Agents Tab
Open ChatGPT. In the left sidebar, you’ll see a new option called Agents. Click the + button that says “Create Agent.”
Step 2: Choose Your Starting Point
You have two options:
- Describe it in plain English — type “Build me an agent that does X” and ChatGPT builds it for you
- Start from a template — the faster, smarter option for most people
OpenAI offers templates for a range of roles including a Software Reviewer, Product Feedback Router, Weekly Metrics Reporter, and Lead Outreach Agent. There are over 20 pre-built templates in the library, covering everything from Chief of Staff to Data Analyst to Design Partner.
Step 3: Connect Your Apps
Once you pick a template, ChatGPT asks which apps to connect. Current integrations include Slack for real-time communication and task updates, Gmail for automating email drafting and follow-ups, Google Calendar for scheduling and event management, Google Drive for document sharing and storage, and HubSpot for CRM updates and lead management.
There’s also an MCP (Model Context Protocol) option — this is where it gets powerful. MCP lets you connect virtually any tool that has an MCP connector: Notion, Linear, your own CRM, or custom internal tools.
Step 4: Watch It Build Itself
ChatGPT helps turn your description into an agent by defining the steps, connecting the right tools, adding skills, and testing it until it works the way you expect. You can literally watch it happen in real time — the agent reads the template, updates files, adds skills, and configures the workflow.
Step 5: Preview Before You Save
Always hit Preview before saving. Run a test prompt — for a Chief of Staff agent, try “Prepare today’s brief.” Watch the working log on the left as it reads your calendar, scans your email, and pulls Slack messages. The output is only as good as the test, so run a real one before deploying to your team.
Step 6: Deploy to Slack
Teams can put agents on a schedule or hook them into Slack, where they pick up requests on their own. To do this:
- In the agent builder, go to the Slack integration section
- Create a new Slack channel (keep it public)
- Copy the channel ID from the channel settings
- Paste it into ChatGPT and add the agent bot to the channel
Now the agent lives in two places: inside ChatGPT for private use, and inside Slack where your whole team can tag it.
Real-World Demo: What a Chief of Staff Agent Actually Does
The Chief of Staff template is the best starting point for most teams — if you can automate figuring out what needs to happen today, that alone justifies the setup.
Here’s what happens when you ask it to “Prepare today’s brief”:
The agent spends about 5 minutes working through your connected tools. It evaluates which tools to use, reads the calendar, searches through recent emails, and pulls Slack context from both public and private channels. Nobody tells it which emails to check or which Slack threads matter — it decides that on its own.
The output includes:
- Recommended focus for today — what actually matters right now
- Slack priorities from the last 24 hours, distilled to action items
- To-do list pulled from actual conversations
- Gmail priorities sorted by urgency
- Calendar summary with meeting prep notes for each session
- Key decisions and blockers
That’s roughly 45 minutes of human coordination work done in 5 minutes. And unlike a human, the agent doesn’t get slower when you have 500 Slack messages instead of 50.
One setup, whole team access. Once you build the agent and add it to a Slack channel, everyone in that channel can use it — on a single subscription.
Real-World Demo: The Design Partner Agent
For teams that ship design work, the Design Partner template shows just how far these agents can go.
The template connects to Adobe Acrobat, Photoshop, Canva, Figma, Asana, and Slack. Its capabilities include creating structured design briefs, auditing a full user journey, critiquing screens, refining interfaces, and packaging design handoffs.
In a live test, the agent was given a set of app screenshots and asked to critique the product flow. The instructions told it to research the brand first, understand their audience and constraints, and then deliver a critique based on that context — the kind of brief a senior designer would give a junior.
What the agent did next was surprising: it ran web searches for the brand’s website, wrote and executed Python code to crop and analyze individual screens, and delivered a seven-page design document covering flow summary, implementation targets, key design decisions, flow architecture, and a code mapping section for engineers.
Nobody told it to write Python. It figured out that cropping and analyzing the screenshots was the right approach to understand the UI — and did it autonomously.
How to Use Agents Day-to-Day (The Shortcut)
You don’t need to navigate to the Agents tab every time you want to use one. Inside any ChatGPT chat, just type @ and a menu pops up with all your saved agents. Select the one you want, and it’s active in that chat with all its tools and context loaded.
You can also update an agent’s behavior without touching any settings. Just type what you want changed directly in the chat — for example, “Stop showing me meetings before 10 AM” — and the agent updates itself.
Workspace Agents vs Claude Agents: What’s the Difference?
Claude has had managed agents for a while, and they’re genuinely capable — particularly for complex, long-horizon reasoning tasks and coding. But there are real differences in the day-to-day experience right now:
| ChatGPT Workspace Agents | Claude Agents | |
|---|---|---|
| Setup effort | Low — templates, no coding | Higher — more configuration required |
| Slack integration | Native, one-click | Not built-in natively |
| Template library | 20+ pre-built templates | Limited templates |
| Team sharing | Built-in, one setup for whole team | More manual |
| Reasoning depth | Strong | Stronger on complex tasks |
| MCP support | Yes | Yes |
The honest summary: ChatGPT Workspace Agents win on ease of deployment and team collaboration right now. Claude agents win on reasoning depth and complex task handling. For most teams that want something working this week without engineering support, ChatGPT’s templates are the faster path.
Availability and Pricing
Workspace Agents are currently in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. The Business plan starts at $20 per user per month.
On Enterprise and Edu plans, admins can configure role-based controls to set which user groups can create and share agents and which tools they can access. The Compliance API gives admins visibility into every agent’s configuration and the ability to pause agents if needed.
Custom GPTs are not going away immediately — GPTs will remain available while teams test Workspace Agents with their workflows, with plans to make it easy to convert GPTs into Workspace Agents.
The 5 Best Templates to Start With
If you’re on a ChatGPT Business or Enterprise plan today, here are the five templates worth trying first:
1. Chief of Staff — Daily brief, Slack triage, email priorities, meeting prep. The best first agent for anyone.
2. Sales Assistant — Researches inbound leads, scores them against your criteria, drafts follow-up emails, updates your CRM.
3. Product Feedback Router — Monitors Slack, support channels, and public forums, turns feedback into prioritized tickets and weekly summaries.
4. Weekly Metrics Reporter — Pulls data every Friday, creates charts, writes the narrative, delivers a report automatically.
5. Design Partner — Critiques screens, audits user journeys, creates structured design briefs, packages handoffs for engineers.
Should You Set One Up?
If your team already uses Slack and Gmail and you’re on a ChatGPT Business plan, the answer is yes — and you should start this week, not next quarter.
The Chief of Staff agent alone recovers 30–45 minutes of coordination work per person per day. At team scale, that compounds fast. And unlike most “AI productivity” tools, the setup genuinely takes under 10 minutes with the templates.
The more important reason to start now: the teams building familiarity with agent architecture today are going to have a significant head start when this infrastructure matures. The workflow patterns you establish now — how agents are prompted, what triggers them, how they hand off to humans — will be the foundation of how your team operates in two years.
The chatbot era is ending. The agent era is already here.
Further reading:
