6 min lesson
Agent playbooks: turn a procedure into something every agent can run
Build Frontline agent playbooks with instructions, tools, and data access. How Max writes playbooks and how agents run them.
An agent playbook is a reusable procedure for a Frontline agent. It holds the instructions for a job, plus the tools, records, custom tools, and integrations that job needs. You assign it to the agents that should be able to run it.
This is the playbook that lives on a Studio agent, the one that talks to people on WhatsApp, live chat, Messenger, or Instagram, or that you use on internal threads. Max has its own playbooks for chat and email. Those are a different library. Agent playbooks scale the agent.
Harrington’s qualifying agent should not carry every procedure in its standing instructions. The instructions say who it is and how it speaks. The playbooks hold “Qualify a mandate,” “Explain fees,” “Request KYC documents,” and “Book a review.” Each one is a complete job.
What does a playbook contain?
Name is how you and the agent find it.
Description is when to use it. That sentence is what the agent sees in its playbook library. “Use when a prospect asks about advisory fees, custody fees, or the all-in cost of a mandate.” The agent matches the conversation to that line, then runs the playbook. Write it as a trigger, not as a summary of the steps.
Instructions are the procedure. Who to look up. What to ask. What to write back to People or Deals. When to call a custom tool. When to hand the thread to a person. Keep this specific and professional. Harrington’s fee playbook might say: look up the Person by phone, read the latest Deal, if a mandate size is on the record use the fee schedule from the knowledge base, explain the fee in one short message, offer a review with the advisor.
System tools are Frontline’s own actions the playbook may use for that job.
Data access is which objects and record types the playbook can read or write, and whether it can log an activity. Qualify a mandate might create a Deal and update the Person. Explain fees might only read.
Custom tools are the API calls that belong to this procedure. Request KYC can call the vendor. Qualify a mandate does not need that endpoint.
Integrations are connected accounts the playbook may use, such as a calendar for booking the review.
A playbook can be active or inactive. It can be visible to the whole account or kept private to the person who created it. Account-wide is the default when the team should share the same SOP.
How does an agent decide to run a playbook?
Assigned playbooks appear in the agent’s library as a name and a when-to-use line. The agent reads the conversation, picks the playbook whose description matches, and executes it. Execution loads the instructions and unlocks that playbook’s tools, records, custom tools, and integrations for the rest of the job.
That is why the description matters more than people expect. Two playbooks that both say “client questions” will compete. “Use when the person asks for a statement or a performance report” and “Use when the person wants to change an address or a beneficiary” will not.
You can assign several playbooks to one agent. The qualifying agent on WhatsApp might hold Qualify a mandate and Explain fees. The onboarding agent might hold Request KYC documents and Book a review. Add a playbook when the procedure is stable. Leave the standing instructions short.
How do playbooks work with knowledge bases and custom tools?
They stack.
The agent already has its instructions, its model, and the knowledge bases you assigned. Those bases are how it answers from policy. A playbook is how it follows a procedure. A custom tool is how it talks to another system during that procedure.
James writes in on WhatsApp about a $12M mandate. The agent matches him by phone and opens a Person. The conversation sounds like qualification, so it runs Qualify a mandate. That playbook asks for timeline and entity type, writes a Deal, and may call Get household snapshot if an account number appears. Fee questions later in the same thread run Explain fees, which searches the “Investment policy and fees” knowledge base and stays read-only on the CRM.
Can Max write an agent playbook?
Yes. Describe the job the way you would brief a senior associate. Who the playbook is for, when it should run, which objects it may touch, which custom tools it needs, and what a good outcome looks like.
Max drafts the name, the when-to-use description, and the instructions. You assign the playbook to the agent. Then send one WhatsApp as James and read the conversation. If the agent picked the wrong playbook, rewrite the description. If it picked the right one and skipped a step, tighten the instructions.
A coding agent on the CLI can create and update the same playbooks, which is useful when the SOP should be reviewed like code.
Where should I start?
Write one playbook for the conversation you already have every day.
For Harrington, that is Qualify a mandate. Description: use when a prospect is exploring a new advisory relationship or names a mandate size. Instructions: greet them by name if the Person exists, confirm the mandate size and the entity, ask timeline and who else is involved, create or update the Deal, offer a review. Give it People and Deals access. Leave custom tools off until the first week of conversations looks right.
Then add Explain fees, with the fee knowledge base already assigned to the agent. Then Request KYC, with the vendor custom tool only on that playbook.
The pattern that holds is a short agent brief, a few playbooks with clear when-to-use lines, knowledge bases for facts, and custom tools for the systems those procedures must call.