Build Your Own AI System

Shift from writing single prompts to designing automated processes you can delegate. Understand how agents differ from chat interfaces, why persistence and structure change everything, and what kinds of work agents handle well that chat alone can't.

Set up the Cowork desktop app and learn its building blocks – instructions, context, files, and live tasks. Run your first end-to-end agent task in Cowork before the session ends. This is your home base for the next eight sessions.

Select a real recurring task from your job – a report, a screening pass, a multi-step workflow – and break it down into the steps, decisions, inputs, and outputs an agent can manage independently. Leave this session with the build brief for the agent you'll spend the rest of the program building.

Build the first working version of your agent in Cowork. Wire up instructions, attach your own files, and create reusable Skills the agent can call on to handle recurring sub-tasks reliably.

Connect your agent directly to the software you use daily – Slack, Gmail, Notion, Google Drive – using Model Context Protocol (MCP). Move from an isolated agent to one that lives inside your real working environment.

Where Cowork's UI hits its ceiling, Claude Code takes over. Get comfortable with the "code without coding" workflow – you describe what you want, Claude writes and runs the script, you review the diff. Use Claude Code for the things Cowork can't do as elegantly: deeper analytical work, custom scripts, multi-file logic, complex data handling.

Transform your agent from a reactive assistant into a proactive system. Design workflows that run on schedule, chain multiple steps together, and operate without you needing to start them – using both Cowork's scheduling and Claude Code's deeper logic where you need it.

A full session dedicated to building. Take your agent further, add the capabilities it actually needs, and bring it from a working prototype to something polished and reliable.

Break your agent on purpose. Find the edge cases, reduce unpredictable outputs, and learn the techniques that take an agent from "works on a good day" to "works every day."

Present your finished agent to the cohort. Three-minute structured demo: the pain you solved, what your agent does, a live demonstration, and what you'd build next. Walk away with an agent that handles a real part of your week – and the skills to keep building.