An AI agent that works through a big task
Let an AI agent plan a task, work it step by step with its tools and write a report.
By the xMap team · Updated
Some work is more than one question: compare the neighbourhoods of a city, check a list of sites one by one, or prepare a weekly brief. Set the AI agent’s How it works to Work through a big task. It brings its own item, so it can follow a Schedule with nothing else before it.
- Write the task in Instructions: what to find out and what the report should say.
- Link the blocks it may use to its Tools port (the purple dot under it), for example Nearby places with its point written
{{agent: the neighbourhood to look at}}. - Pick Most steps: the longest to-do list it may work through (3 to 30).
- Put an Email or Make a report after it, with
{{report}}in the message.
When it runs, the agent writes itself a to-do list, works the steps one at a time, hands a self-contained step to a helper when that is simpler, keeps notes between steps, and ends with a report. While it works, the run shows the step it is on ("Step 4 of 9: ..."). A long task carries on in the next parts of the run, about an hour at most.
It stops at Most steps, or when the run’s credits run out, and still reports what it has. A test run stops after 3 steps and says so. Each thinking step costs what an AI agent step costs; its tools cost what they cost in a flow.
To see how it worked, open the run and press See how it answered on the AI agent’s line. The plan is a checklist (done, nothing found, not reached); open a step to see each tool it used, what it gave the tool and what came back, a helper’s work folded inside the step it did, its notes, and the time and credits of each step. Then comes the report, and why it stopped, such as Stopped: credits or Stopped: Most steps. Its notes are shown only to the flow’s owner and organization admins; keys and tokens are always hidden.