FIELD NOTES
Voice AI booking: three outcomes to check with Calvin Observe
A voice AI agent says the appointment is booked. Calvin Observe checks the calendar to find out whether that promise matches the record. These three cases show how the evidence determines the result.
The voice call#
Caller: “Can I come in Tuesday at 3?”
Voice AI agent: “Of course. You’re booked for Tuesday at 3 pm. See you then!”
The commitment is to create the appointment before confirming it aloud. Observe needs calendar evidence with enough context to match the booking to the caller and requested time.
Missing booking: the commitment was not met#
The calendar is available and was checked, but no matching appointment exists. Observe labels this “Off orbit · Commitment not met.” The polite confirmation does not establish a completed booking.
Observe routes the call transcript and calendar evidence to the front desk lead. That owner decides how to resolve the caller’s request and investigate why the spoken confirmation and recorded outcome differed.
Verified booking: the record matches the promise#
The calendar contains a confirmed appointment matching the customer, date, and time. Observe labels this “In orbit · Outcome verified.” The record supports this specific booking commitment.
Continue checking the same requirement as the agent and business rules change. A successful example is evidence about that outcome; it is not a guarantee about every future conversation.
No system access: the outcome is unverified#
Calendar access is unavailable. Observe labels this “Out of view · Evidence unavailable.” There is not enough evidence to decide whether the appointment exists.
Restore access and rerun the outcome check. Keep the result distinct from both a verified booking and a known missing booking.
Apply the same approach to chat and back-office agents#
The evidence source changes with the mission. For a conversational coaching chatbot, compare the feedback with the underlying match data. For a back-office automation, compare an invoice marked paid with the bank’s settlement record.
Simulate creates scenarios from the mission so your team can check these commitments before launch. Observe checks real work in production, groups recurring problems, and routes evidence to the responsible owner. Integrating your agent with Observe takes 5 minutes.