Skip to Content
DocumentationBuild with AIRead what AI built

Read what AI built

AI does not hand its work over somewhere else — it writes into your own project. Every result lands where it would have landed if you had built it, in the editors you already use.

Start from the per-turn summary

At the end of each turn, AI Copilot posts an N items created summary, grouped by table, Actionflow, API and AI Agent. Click any entry and the editor jumps straight to it. That is the fastest way in: see what changed this turn, then decide what deserves a closer look.

While it is building, you can also switch on Follow AI changes in the panel header. The editor view jumps to each change as AI makes it, so you can watch the work happen.

Where each kind of result lives

ResultWhere to lookHow to read it
Tables and fieldsData tabField names, types, and whether they are required
Dependencies across tables, APIs and ActionflowsOpen the bird’s-eye view from the data model, Actionflow, or AI Agent editorA node-and-edge graph centered on the current object, for seeing what one change touches — see Bird’s-eye view
ActionflowsAction → Actionflow canvasRead the nodes top to bottom; that is the logic, step by step
PagesDesign tabThe component tree shows structure; select a component to see its data binding
Roles and permissionsSettings → PermissionsWhich tables each role can read or write, and which Actionflows it can call
Third-party APIsAction → APIRequest method, path, parameters, and response shape

Worth checking after each turn

  • Are the tables and fields what you asked for — names, types, required or not. Once a field is deployed to the backend, you cannot switch between a normal field and a formula field; you have to delete and recreate it. See Set Up the Database.
  • Is the Actionflow in the right order — especially anything that charges money, decrements stock, or sends a notification. Node order is business order.
  • Are permissions too open — who can see a newly created table? Did a generated role expose sensitive data to anonymous users? See Manage app permissions.
  • Is the page bound to the right data — a component bound to the wrong query looks fine and shows the wrong numbers.

When you cannot tell, make it run

If reading does not settle it, have it run for real. A debug run shows every node’s inputs, outputs, duration and errors; repeating a read as a specific identity shows what that identity can actually see. See Verify against the live app.

⚠️

Verification runs against the deployed backend. If the draft has unsynced changes, the result does not describe the draft.

Last updated on