Most AI tools make you do the orchestrating. You paste in context, remind it what happened last time, manually trigger each step. ChatGPT Workspace Agents flip that dynamic — you describe a recurring job, and the agent figures out which apps it needs, builds its own skills, sets a schedule, and maintains persistent memory across runs.
Here’s what that actually looks like in practice and how to get one running fast.
What Workspace Agents Actually Do
Think of an agent as a standing employee with a defined role. You give it a job description, connect it to your tools, and it handles the workflow on autopilot — on demand or on a schedule.
What separates these from a standard ChatGPT chat:
- App integrations — an agent can read your calendar, pull tasks from a project manager, and check a communication tool, all in one run
- Auto-generated skills — when you build via chat, the agent writes its own skill files (structured instruction sets that shape how it handles specific subtasks)
- Persistent memory — a dedicated folder where the agent stores notes, drafts, and outputs it can reference on future runs
- Scheduling — set it to fire at 8 a.m. every weekday and it will, without you opening ChatGPT
Who Can Access Them Right Now
Workspace Agents currently require a Business, Enterprise, or Education plan. Standard paid ChatGPT subscriptions don’t include them yet. If you’re on a regular Plus plan and don’t see the Agents section in your left sidebar, that’s why — not a bug.
Business plan starts at two seats, so even a solo operator can qualify by purchasing the minimum.
How to Build Your First Agent (The Fast Way)
Skip the manual builder. Instead, press the + icon in the Agents sidebar and describe what you want in plain language. The more specific you are about the workflow, the better the output.
A concrete example: say you want to start every morning knowing exactly what to focus on. You’d describe it roughly like this:
Every weekday morning, check my Google Calendar for today’s meetings, pull my open tasks from Todoist, and summarize any unanswered messages in Slack. Give me three priorities, flag any scheduling conflicts, and suggest a time-blocked plan for the day.
After you send that, the builder will:
- Name the agent and write its role description
- Identify which apps it needs (Calendar, Todoist, Slack in this case)
- Generate any skills it needs to do the job well
- Set a daily schedule
- Prompt you to connect any apps you haven’t authorized yet
All of that happens automatically. You’d have done every one of those steps manually in older agent platforms.
Connecting Apps
Head to Browse Apps inside ChatGPT to see what’s available — tools like Google Calendar, Notion, Asana, HubSpot, Dropbox, and Intercom are already there. For apps not on the native list, MCP (model context protocol) connections are available and expanding.
Connect the apps once and every future agent you build can draw on them.
Useful Agent Ideas Beyond Scheduling
The morning planner is the easiest entry point, but the pattern applies to almost any repeatable knowledge-work task:
- Content review agent — pulls a draft from Google Docs, checks it against a brand style guide stored in the agent’s memory, flags issues, and posts a summary to a Slack channel
- Support triage agent — monitors an Intercom inbox, tags tickets by urgency and topic, drafts first responses for common questions, and escalates edge cases
- Weekly reporting agent — aggregates task completions from Asana, meeting notes from a Notion database, and key metrics from a connected source, then writes a structured weekly summary
- Design brief agent — takes a project brief submitted via form, cross-references past project files in Dropbox, and generates a structured creative brief with relevant references
Every one of these can be built through the same chat-based flow. Describe the workflow, connect the apps, let it build.
Skills: The Part Most People Miss
When you build an agent via chat, it may create skills on its own — markdown files that define structured behaviors for specific subtasks. A content agent might create a content-reviewer-skill.md that spells out tone rules, a checklist of common errors, and output format requirements.
You don’t have to write these. But you can edit them, add your own, or upload existing ones. Skills are what make an agent reliably consistent rather than improvising each run. Once you’ve used an agent for a few weeks and noticed where it drifts, tuning the relevant skill file is usually the fastest fix.
Sharing Agents Across a Team
On Business and Enterprise plans, agents are shareable. Go to the agent’s edit view, hit the three-dot menu, and choose Share. You can list it in your workspace’s agent directory so teammates can find it, or just copy a direct link.
Individual chat runs are also shareable — useful when you want to walk a colleague through what the agent produced without giving them full agent access.
The Real Value Here
The leverage isn’t in any single agent. It’s in building a small library of them — each one handling a recurring, brain-draining task — so that every morning you’re working on decisions instead of assembling information to make them. Start with one workflow that currently eats 20 minutes a day. Build the agent, run it for a week, tune the skills where it misses. Then add the next one.
