You'll want an Autonomous Agent whenever the work is too big to type back and forth in a chat — a report you'd rather receive finished, an integration you'd normally hand to a developer, a research pass that needs to run while you're doing something else.
Setting one up is quick. What matters more than the setup is how you brief it — but we'll get there. First, the basics.
Create a new Agent
In the left sidebar, click Agents
In the top-right corner, click + New Agent
Give the Agent a Name - a short label so you can tell your agents apart
Write the Instructions - what you want this Agent to do. These are the ground rules that apply to every task it takes on.
Create Agent
From there it will add Agent to the list and open side panel to create your first Task inside that agent.
⚠️ Once created, Agent instructions can't be changed. Take a minute to get them right.
Agents work through Tasks
Every actual piece of work happens inside a Task.
Think of the Agent as a group of related Tasks that share a purpose. The Instructions you wrote at the Agent level are the ground rules that apply to everything inside. The Task is where you write the specific brief for the specific job, and the more detail you put in, the better the result.
A single "Content Assistant" Agent might work through Tasks like:
"Write a 900-word blog post about how small teams manage research. Pull ideas from my saved Topic 'Team Workflows', match the tone of my last three published posts, add a hero image, and save the draft to my workspace."
The Agent learns as it works
Autonomous Agents pay attention while they run. If a Task's brief is missing something important, the Agent can update the instructions itself and keep going with the improved version.
You can edit Task instructions too, at any time. Just don't do it mid-run — start a fresh Task if the direction has really changed.