Private course for your team

AI Agent Training with Claude Cowork

One agent doing everything is the mistake everyone makes first. Build small agent teams with clear roles, handoffs and approval rules, and watch what breaks when you do not.

Length

1 day

Format

Remote or on-site

Tools

Claude Cowork

Audience

Builders, technical leads and hands-on operators

Request information

The problem

A single agent loses the goal, grades its own homework, and when it fails you cannot tell which part failed. The fix is not a bigger prompt. It is a shape.

What your team walks away with

  • Working agent teams built from plain, readable role files.
  • Three proven patterns, and the failure mode of each.
  • A pattern you chose and built for a real recurring decision at your company.
  • The same team running in Cowork on a schedule.

The labs

1

Your first agent role

Roles as files, and tool limits as the permission boundary.

2

The assembly line

A fixed pipeline, and how an upstream error arrives downstream looking confident.

3

The council

Independent reviewers who must agree before anything is released, and why independence matters.

4

The lead and specialists

An agent that plans and routes work, and what that costs.

5

Build your own

Pick the pattern that fits a real judgment call from your work, and build it.

6

Cowork on a schedule

The same team as a skill and a daily scheduled task.

Cover of Design Patterns for AI Agents by James Harmon

Built on the book Design Patterns for AI Agents

The patterns taught in this course are the ones catalogued in the book.

More about the book

Who it's for

  • Builders and technical leads who will own the agents afterwards.
  • Hands-on operators who want to shape their own work, no coding required.
  • Teams with a recurring judgment call they would like to shape into a pipeline.

What you need

  • Claude Desktop on a Claude Team or Enterprise plan, with Cowork enabled.
  • A setup check goes out a week before.

Taught by James Harmon

Independent consultant and trainer based in the Chicago area, specializing in the Anthropic Claude ecosystem. Author of Design Patterns for AI Agents. Over 20 years building software for companies including JPMorgan Chase, State Farm and Foot Locker. Runs XoSelf.

XoSelf also takes on a small number of Claude consulting engagements: rollout, agents and MCP integrations. Mention it in your request.

About the book

Request information

Tell us a little about your team and James will come back with dates, pricing and what the day would look like for you.

No newsletter, no sales sequence. James reads these himself.