Skip to main content

Chapter 3: Working with General Agents: Claude Code and Cowork

You know why AI-driven development matters—and that it's happening right now, in 2026. But understanding transformation and experiencing it are two different things.

This chapter introduces you to Claude's General Agents—Claude Code and Cowork. These aren't just AI assistants; they're agentic AI systems that can reason through problems, make plans, and take action across domains.

What's a General Agent? An AI that observes, orients, decides, and acts—the OODA loop—executing actions rather than just generating text. Claude Code uses the terminal interface (for developers), while Claude Cowork uses the desktop interface (for knowledge workers). Both are powered by the same Claude Agent SDK.

The goal is not to build a supercomputer, but to establish Claude as your collaborative thinking partner for everything—and to build Skills that can become products.

📚 Teaching Aid​

What You'll Learn​

By the end of this chapter, you'll have:

  • General Agents Explained — Understanding why Claude's agentic AI is fundamentally different from passive assistants: the OODA reasoning loop, the Product Overhang insight, and filesystem access as the universal interface
  • Two Interfaces, One AI — Claude Code (terminal-based for developers) and Cowork (desktop-based for knowledge workers), both running on the Claude Agent SDK
  • Claude Code Installed & Working — Complete setup with multiple paths: Official Claude Pro/Max subscription (Lesson 2), OR free backends via OpenRouter, Gemini, or DeepSeek (Lesson 3)
  • Hello World — Mastering the CLI interface (Lesson 4): slash commands, permission loops, and safe usage patterns
  • Persistent Project Context — Creating CLAUDE.md files (Lesson 5) that eliminate repetitive context-sharing and establish team memory
  • Teach Claude Your Way — Mastering custom instructions (Lesson 6) to align Claude's behavior with your personal or team standards
  • Autonomous Expertise — Agent Skills (Lessons 7-8) with prompt patterns (Persona + Questions + Principles)
  • Subagent Orchestration — Recognizing when Claude automatically delegates to specialized agents (Lesson 9) for complex tasks
  • External Integration — MCP servers (Lesson 10) for safe access to external systems
  • Token Optimization — Compiling MCP servers to Skills (Lesson 11) for 80-98% token reduction
  • Configuration Mastery — Settings hierarchy (Lesson 12) and precedence rules for team collaboration
  • Event-Driven Automation — Hooks (Lesson 13) for custom behaviors triggered by events
  • Complete Extensibility — Discovering and using plugins (Lesson 14)
  • Autonomous Iteration — The Ralph Wiggum Loop (Lesson 15) for self-correcting workflows
  • The Creator's Workflow — How Boris Cherny (Lesson 16) uses Claude Code effectively
  • Claude Cowork Fundamentals — From terminal to desktop (Lesson 17), getting started (Lesson 18), and practical workflows (Lesson 19)
  • Browser Integration — Claude in Chrome (Lesson 20) for web-based automation
  • Connectors — MCP for everyone (Lesson 21): pre-built integrations with Google Workspace, Notion, Slack, and more
  • Built-in Document Skills — Working with docx, xlsx, pptx, and pdf files (Lesson 23)
  • Decision Framework — Choosing between Code and Cowork (Lesson 24)
  • From Skills to Business — Understanding how Skills become monetizable products (Lesson 25), the Digital FTE model, and the path to revenue