User input: “Create a segment of enterprise users who haven’t logged in for 30 days.”
Step 1: Resolve the criteria
The AI begins by looking up available traits before constructing anything. It doesn’t assume field names. It verifies them first.
The AI scans existing segments and user traits before building the payload
Tool call 1: look up “plan” traits
Tool call 2: look up “last login” traits
The AI uses
autocomplete_query to verify real field names before building the payload. It doesn’t guess.Tool call 3: create the segment shell

The AI surfaces structured options for ambiguous inputs. The user can select or type a free-text answer
Confirmation panel

The user sees the full resolved payload before anything executes. Nothing fires until they click Create Segment.
Tool call 4: apply segment definition
Step 2: Test the segment
The wizard advances automatically. At step 2 the wizard prompts: “We’ve created your segment. Let’s test it: do you have a user ID or email in mind?”Tool call 5: look up user by email
Tool call 6: check membership
Execution summary
6 actions · 1,033ms · 100% success
The four reads ran silently. The two writes paused for user confirmation before firing.
What if something fails

Connection errors are handled gracefully. The user can retry or continue without the context
