Building AI Agents That Actually Work

From Simple Skills to Multi-Agent Systems

What You'll Learn

Most people use AI as a chat interface. This course shows you how to build systems — reusable, composable pieces that get smarter the longer you run them.

  • Skills — Reusable instruction sets that load when you need them. Ten lines of YAML and a markdown body. Powerful out of proportion to their size.
  • Agents — Skills with tools. When your instruction set needs to read files, run bash commands, or call APIs, it becomes an agent. Same structure, more reach.
  • Pipelines — Agents that orchestrate other agents. Research, draft, review, publish — each stage handled by a specialist, quality-gated between them.
  • Ecosystem management — How to keep 50+ agents coherent. Creation guards, health checks, evals. The unglamorous work that makes the glamorous work possible.

Every module draws from a real, working system. The skills and agents named here exist and run in production. This isn't theory — it's documentation of what actually works.

Time commitment: About 3 hours total. Each module stands alone, but the progression matters — read them in order.

  1. What Skills and Agents Actually Are

    Not AI hype. A skill is a reusable instruction set. An agent is a skill with tools. Why this distinction matters and where most people go wrong.

  2. Your First Skill

    Write a skill from scratch. What goes in the YAML header, how to structure instructions, how Claude Code loads it. Worked example: think-first or log-to-daily.

    Members only

  3. Skills That Edit and Fix

    Skills that don't just advise but actually change files. The editors-not-critics principle. Use writing-quality and voice-editor as worked examples.

    Members only

  4. Building Agents

    When a skill needs tools — Read, Write, Bash, web search — it becomes an agent. Agent definitions, tool permissions, and delegation. Worked example: draft-reviewer.

    Members only

  5. Multi-Agent Pipelines

    Agents that call other agents. Subagent-driven development. The content pipeline: research → draft → review → publish. Quality gates between stages.

    Members only

  6. The Ecosystem

    Creation guard to prevent duplicates, ecosystem-health to detect drift, evals to test agent behaviour. How to maintain 50+ agents without chaos.

    Members only