A recorded walkthrough of an actual build from kickoff to first live call. Discovery from real recordings, the knowledge base being assembled, prompt tuning, the eval suite catching a genuine failure, the fix, and then the agent taking its first real call. The edit keeps the parts that went wrong.
The eval failure is the most instructive minute in the recording. It is a booking-conflict case that only appears when a caller changes their mind after a slot is offered — exactly the kind of bug that never shows up in a scripted demo and always shows up in week three of production.
Buyers who want to know what they are paying for, and in-house teams deciding whether to build it themselves. If you are the second, the recording is a fair map of the work: the model is the easy part and the evaluation loop is most of the calendar.