Where each client job stands
A Friday summary of open Tasks per client Space, written from the Tasks themselves.
Open betaFree while the beta runs
A report should makethe next question clearer.
Keep the source and the time window on the page. Say what changed, say what it might mean, and give the next step to a person.
Four parts.Every report has them.
A number without a source or a date is hard to act on. Write these four into the agent's instruction or a Skill, and every report arrives the same way.
- Source
Read a defined source
An MCP connection with read tools, or a file in the Library. The report names which one it used.
- Window
State the period
Yesterday, last week, this month. The dates go on the page, with the time the data was read.
- Reading
Keep fact and reading apart
Sold 84 and 17 left were observed. The campaign is working is a reading, and the page puts it under its own heading.
- Follow-up
Give the next step an owner
The agent opens a Task with the report linked. A person decides what happens.
When the data is not there.The report says so.
The useful difference between three Monday mornings.
Complete
Every source answered
The report is posted with its sources and its window. The follow-up Task links back to it.
Posted
Partial
One source is missing
The instruction says a missing source is named and never estimated. The report goes out with the gap stated where the number would be.
Posted with a gap
Failed
The Run did not finish
You see the failed Run with its reason and a retry. A failed Run never posts a made-up report in its place.
Retry offered
Delivery
Run it once by hand.Then put it on a schedule.
The report is the same either way. Only what starts it changes.
Yesterday's orders, every morning
The digest that opens a Task when stock runs low. See it on the page for online shops.
What shipped this week
Published Change Requests and the Tasks behind them, read from the Project.
Ask for it in a thread
Mention the agent in a Channel or assign it a Task. Good for the first version and for one-off questions.
By handSchedule it as an Automation
The same instruction with a time and a Channel to post in. Automations covers schedules in detail.
On a scheduleBefore you trust a report.
Where does the data come from?
Autopilot has no built-in analytics or store connector. Data reaches the agent through an MCP server you connect under Integrations, or through files you put in the Library.
Is a report live?
No. It covers the window it states and reads the data when the Run starts. To see newer numbers, run it again.
Are the interpretations facts?
No. The numbers are read from the source. The reading is the agent's suggestion and sits under its own heading, and a person reviews it before anything with consequences happens.
Can a report act on its own findings?
It can open a Task for a person. Changing something in an outside system needs a tool the agent's Role has been granted, and without it the agent has to ask.
Where does the report end up?
Where you tell the agent to put it: a message in a Channel, or a Knowledge page in the Library that keeps its revisions. Figures arrive as text and tables. A report that needs more layout is an HTML file the agent attaches, opened in an isolated frame.
Start with one report.
Pick one your team already reads. Name the source, the question and the person who acts on it.