Work with your agent

You connected an assistant. This guide gives you the mental model behind the app and the habits that get good results.

Three things to know

• A dashboard is one document. Every chart, table, note, and filter lives in one JSON document. Your agent reads and edits that document — the same one the app edits. Changes appear in the app as the agent works.
• Charts read saved data. A datasource is a query plus a snapshot — a saved copy of the query's results. Charts read the snapshot, never the live warehouse. Fresh numbers come from a refresh, not from opening the dashboard.
• The agent acts as you. The agent sees the dashboards you can see and edits the ones you can edit. Every change is recorded under your account.

What an agent can do

• Create dashboards, and edit everything in a dashboard: blocks, queries, layout, field formats, the theme.
• Run datasources and refresh their snapshots — the kinds that run server-side.
• Explore before it changes anything: read snapshot rows, test a join, try a transform.
• Use the org's shared datasources — curated queries with data included.
• Browse the warehouse catalog: table names, descriptions, and columns. Never rows. The catalog needs the analyst right — the same right that runs BigQuery under the org's credential.
• Check its own work: validate the dashboard, and render any chart as the same picture a viewer sees.
• Answer your comments. Commenting is not editing, so an agent you shared a dashboard with read-only can still reply to you on it.

Your rights carry over. If your account cannot run BigQuery under the org's credential, the agent cannot either.

You can read the checks yourself. Open the Findings panel — the warning icon on the right-hand rail — and the dashboard tells you what is broken, what will read badly, and how much the charts cost to draw. Every finding carries a class. A problem means the dashboard is broken or reads wrong now. Advice means the dashboard works and something will hurt. The badge on the rail counts problems alone. A block with a problem carries the same mark in its own header.

The panel shows the checks that run in the browser. Three checks need the server and reach agents alone: the Vega-Lite compile check, the check for an unregistered service account, and the check for an unknown delivery destination. An agent can report a finding the panel does not show. The panel is for editors — a reader you shared a dashboard with read-only will not find it. Reading that panel after an agent finishes is the fastest way to judge the work.

What an agent cannot do

• Refresh a datasource that runs under your own Google sign-in. That kind runs only in your browser. If the agent must keep a query fresh, ask it to use a server-side datasource.
• Import a file from your computer. Import the file in the app — drop it on the canvas, or use the Data panel. After that, the agent works with the imported data like any other datasource. A file that lives at an address is different: give the agent the address and it can add and refresh that one itself.
• See or change a dashboard outside your access.
• Change who can see a dashboard. The owner, the share list, and "everyone signed in" are yours to set in the app: open the dashboard, then Dashboard settings → Sharing. An agent asked to share a dashboard answers with that instruction instead.
• Stamp a maturity level, promote a datasource into the datasource library, save a theme to the theme library, send a delivery, register a service account or a destination, or upload a file. Each of these acts records a person standing behind the result, so each is an act in the app.

Habits that get good results

• Name the dashboard. "On the dashboard 'Course feedback', …" beats "on my dashboard".
• Point at data. Ask the agent to look in the datasource library first — the org's curated queries, ready to copy into a dashboard. Or name the warehouse table.
• Ask the agent to look at its work. "Render the chart and check that the labels read well." The agent sees the same picture a viewer sees, theme included.
• Leave comments as instructions. A comment on a block is a to-do an agent can pick up: write "split this by month" on a chart, then ask an agent to resolve the open comments.
• You can always step back. Every save is archived with a version number and the name of who saved it — you, an agent, or the schedule. Open Dashboard settings — the gear on the right-hand rail — and use Version history: click a version to view it read-only, then restore it or go back to the current dashboard. An agent's changes are easy to review this way before you decide to keep or undo them.

Example asks

• "Create a dashboard 'Enrollment overview' from the shared datasource 'Enrollment by program'. One bar chart of students per year."
• "On 'Course feedback', add a scorecard with the average rating, one decimal."
• "Resolve the open comments on 'Weekly KPIs'."