The Agent Builder Workbook

Free Resource

From one paragraph to a working three-agent chain.

Most CAS practitioners read about AI agents and stall on the mechanics. The Agent Builder Workbook is the five-exercise companion to the Agent Builder series — concept brief, worked example, fillable template per section, in a printable PDF plus a fillable DOCX. Build the first chain in about five hours.

Five hours, one client
Five fillable templates
PDF plus companion DOCX

The Gap

A working chat session isn't a system.

Most practitioners can get a useful answer out of ChatGPT, Claude, Copilot, or Gemini. What they don't have yet is an agent that runs again next month without rebuilding it, ports across clients without rewriting the prompt, or chains with other agents to deliver finished work end-to-end.

The gap between a working chat session and a working practice system is operational, not technical. Five exercises close it.

What's Inside

Five fillable templates. One running case study.

Each section has a concept brief, a worked example with one running client, and a fillable template you complete for a client of your own. The whole workbook ports across your book by swapping one file.

PDF

Read on screen, print to fill

A 32-page brand-quality workbook with concept briefs, worked examples, and fillable templates for each of the five exercises. Designed to read on a tablet or print and write into.

DOCX

Templates only, fill in your tool of choice

A companion file with the fillable spaces from the PDF and no commentary. Save a copy per client, fill in directly, keep alongside your client folders.

Case Study

One client, end to end

A small craft brewery — Glendale Brewing — carries through every section so you see the same artifact at every stage of the build. The case study isn't a demo. It's the worked example you reverse-engineer onto your own client.

Platform-Agnostic

Cowork, Copilot, Gemini, ChatGPT

The mechanics are tool-independent. The workbook flags the platform-specific differences only where they matter — replication, file access, lifecycle storage. Build with whichever LLM your firm already pays for.

Five Templates

From a single instruction to a three-agent chain.

1. Agent-grade instruction

The five components that turn a prompt into an agent: identity, context, rules, output format, contingency. Worked example: an engagement letter agent. Fill in for one of your own recurring tasks.

2. Workflow audit

Tag every step of one client's recurring engagement as Z1 (outside the platform), Z2 (data exported), or Z3 (inside the platform). Your first five agents are Z1.

3. Client context file

Ten fields capturing the institutional knowledge an agent — or a new staff member — needs before drafting an email, reviewing a statement, or jumping on a call. The single highest-return artifact in your AI stack.

4. Iteration log

Build the third component — the example — through use. Run, read, fix the instruction (not the output), run again. Two or three rounds and your agent runs at the standard you'd send.

5. Three-agent chain planner

Compose three agents into a working chain, with manual handoff between each. The realistic on-ramp for the next 12 to 18 months. Same chain ports to your next client by swapping one file.

The math. One agent saves roughly 15 minutes per client per month. A three-agent chain saves closer to two hours. Across 30 clients, that's 60 hours a month — most of a half-FTE returned, before adding a person.

Stay Current

Get the Workbook — plus weekly AI briefings for CAS.

Subscribe to The AI Accountant newsletter and get the Agent Builder Workbook delivered to your inbox, along with weekly analysis of the AI developments that matter for your practice.

Your Move

Five hours now. Two hours back per client, every month.

The compounding doesn't start until the first chain is running. Pick one client. Pick one cycle. Run the five exercises end-to-end. The next thirty clients port over by swapping one file.