Build / Playbooks

Goals instead of scripts

Some conversations can’t be drawn as a straight line. A playbook gives the agent an objective and the outcomes that count as done, and lets it find its own way there — without giving up control of when the call moves on.

When to reach for one

Three ways to get words out of an agent, from most controlled to least:

Scripted stepsSay / Ask / Confirm — you write the words, in order. Total control.

A playbookYou describe the goal and the outcomes; the AI finds its own path there.

An AI replyOne free-form turn. No goal, no outcome, no gate.

Use scripted steps for anything that must be said a specific way — disclosures, confirmations, compliance lines. Use a playbook when the caller could reasonably take the conversation in several directions and you care about the outcome, not the wording.

What a playbook is made of

a playbook

Name          lead_qualification
Description   Qualify an inbound lead and book a demo if they're a fit
              (this is what the AI reads to decide when to use it)

Instructions  You are qualifying an inbound lead. Find out what they're
              trying to solve, roughly how big their team is, and whether
              they have a timeline. Be curious, not interrogative — one
              question at a time, and react to what they say.

Tools         lookup_account, book_demo

Outcomes      Booked demo      — caller agreed to a demo
                required: email, company
              Not a fit        — wrong size or no need
              Follow up later  — interested but not now
                required: email

Description

What the AI reads to decide whether this playbook applies. Write it for the model, not for your team.

Instructions

How to pursue the goal — tone, what to ask, what to avoid. This is the prompt, so the advice in Voice & turn-taking applies: show it what good sounds like.

Tools

Which tools it may call while working. Unlike a plain AI reply, a playbook can look things up and act mid-conversation.

Outcomes

The ways this can end — each becomes an exit on the canvas, so you wire what happens next. Give each a plain-language description; that’s how the AI decides which one it reached.

Required variables

Per outcome. The playbook cannot exit through it until those are captured — see below.

🔒

Required variables are the safety rail. They’re enforced by the engine, not by asking the model nicely. If “Booked demo” requires an email and the AI tries to finish without one, it is refused and has to keep going. This is what makes a goal-driven step safe to put in a real call.

Using one in a flow

Add a Playbook step and pick your playbook by name. Its outcomes appear as exits on the card — wire each one to what should happen next. A playbook that books a demo might exit to a confirmation, while “Not a fit” exits to a polite close.

Reuse and versioning

Playbooks live under Flows → Playbooks, not inside one flow, so the same qualification logic can be used by several agents. Publishing snapshots a version, so improving a playbook doesn’t silently change a flow that was working — and you can deploy a playbook directly as its own agent when it is the whole job.

Writing a good one

Name outcomes as things that happened, not as next steps: “Booked demo”, not “Send calendar invite”.
Cover the unhappy ones. “Not a fit” and “Wants a human” are real outcomes; without them the agent will improvise an ending.
Only require what you truly need. Every required variable is a thing the caller must provide before the call can move on.
Keep it to one job. If a playbook has eight outcomes, it’s probably two playbooks with a Branch by intent in front.

🧪

A playbook is the least predictable step you can add, so it’s the one most worth testing. In a dry run try being unhelpful, changing your mind, and answering a different question than the one asked — then check it still lands on a sensible outcome. Note that a replay can word things differently: Reset and run it again a couple of times.