assembl · intuitive agentic customer journeys
The agent schema
The structure of an assembl agent — one specialist in an agentic customer journey. Each agent has a single job, six declared parts, and a written authority level. Nothing that reaches a customer or commits money proceeds past draft without a named person approving it.
Machine-readable version: /agent.schema.json. Walk it in 3D on the homepage gallery, or see what AI makes of your own website.
The six parts
- knowledge
- What it may read: the written record of the business — services, prices, voice, policies. Declared, versioned, and the agent reads it before it does anything.
- signals
- What it watches for: work coming due, things lapsing, customers gone quiet. Signals trigger drafts, never sends.
- ability
- What it can do, and the written list of what it may touch.
- boundary
- Where it stops: the things it will never do on its own, in writing. No sending, no spending, no promises a person has not seen.
- approval
- Whose yes it needs: a named person, and nothing consequential without them.
- flightLog
- What it did, kept: what it read, what it wrote, who approved it, how long it took. Measured numbers and calculated ones are kept apart, and labelled.
The authority ladder
- observe — watches, says nothing
- draft — writes it, holds it
- recommend — puts options to a person
- act-with-approval — acts only after a named person clicks yes
Invariants
- One agent, one job.
- Every part is declared in writing before the agent runs.
- Nothing that reaches a customer or commits money proceeds past draft without a named person.
- The flight log keeps measured numbers and calculated ones apart, and labels which is which.
- A second agent reuses the same knowledge and boundaries — the journey is a team, not a monolith.