What problem is worth solving?
Teaching scenario: online interest is lost between appointment, store handoff and outcome. Carry authorized intent with context to an accountable owner and capture what happens next.
Reduce lost intent and duplicate outreach; connect conversations to traceable store service.
- WORKING ROLES
- User, online service, store staff, operations owner and FDE
How does an FDE turn symptoms into a solvable problem?
The following discovery and implementation exercise uses a synthetic scenario.
Follow one inquiry end to end
Replay consent, opt-out and rescheduling paths, noting loss of need, time, store or ownership.
Ask why the store cannot act
Is consent missing, the slot unclear, stock unknown or the notification unowned? Repair the gap rather than sending more alerts.
How does the solution enter the workflow?
- 01
Understand
Confirm needs, time and store preference without assuming purchase.
- 02
Authorize
Confirm permitted channels and appointment; stop outreach on opt-out.
- 03
Hand off
Pass needs, confirmed conditions, owner and acknowledgment deadline.
- 04
Record
Link visits, rescheduling, cancellation and no-show by appointment ID; preserve unknowns.
Which decisions make the solution reliable?
Sent is not handed off
Handoff needs acknowledgment and usable context; timeouts escalate to an assigned role.
Keep unknown outcomes unknown
Record unknown visits honestly rather than forcing outcomes; measure recording coverage separately.
Rescheduling is not a new customer
Synthetic appointment A-21 moves from Saturday to Sunday; independent updates can cause duplicate contact without a stable ID.
- Event
- Keep appointment_id; update version 1 to 2.
- Tool
- Check version before writing; replay returns the original result.
- Store
- Show the new slot and reason; invalidate the old notice.
What should you do and deliver?
Map inquiry-to-handoff objects
- OUTPUT
- Service blueprint + ownership
- PASS CONDITION
- Each stage has a receiver and input
Implement appointment changes and consent
- OUTPUT
- Tool contract + state machine
- PASS CONDITION
- Handle opt-out and replay correctly
Test unowned work and stale context
- OUTPUT
- Escalation + recovery
- PASS CONDITION
- Notifications cannot remain silently unowned
How do you prove completion and reuse the work?
Acceptance checks
Track consent, context, duplicate outreach, response, visits and recording coverage with a complete observation window.
Reusable assets
Reuse handoff objects, state machines and outcome contracts for appointments, leads and field service with scoped consent and rules.
Further reading and original sources
- Building effective agents ↗Anthropic · 2024-12-19
- Demystifying evals for AI agents ↗Anthropic · 2026-01-09