AI Agents: Through the Company of Zero

AI Agents: Through the Company of ZeroThe first edition of this book was written when AI agents were mostly a promise. This second edition is written from inside the revolution. In the months since, Anthropic released the Model Context Protocol. Google shipped A2A. Claude Agent SDK, OpenAI Agents SDK, CrewAI and LangGraph went from weekend experiments to production infrastructure. Klarna quietly walked back its "AI replaces humans" announcement. Air Canada lost a lawsuit because its

The first edition of this book was written when AI agents were mostly a promise. This second edition is written from inside the revolution.

In the months since, Anthropic released the Model Context Protocol. Google shipped A2A. Claude Agent SDK, OpenAI Agents SDK, CrewAI and LangGraph went from weekend experiments to production infrastructure. Klarna quietly walked back its "AI replaces humans" announcement. Air Canada lost a lawsuit because its chatbot invented a refund policy. The EU passed the AI Act. And a small but growing group of founders started running real businesses with almost no employees.

This is a field guide for the people building in that world.

Inside you will find:

• What an AI agent actually is - and what it isn't - in terms you can use tomorrow morning
• The protocols (MCP, A2A) that turned isolated demos into a real ecosystem, and why they matter more than any single model
• Context Engineering: the discipline that replaced prompt engineering as the skill that separates working agents from expensive toys
• The design patterns from Anthropic's "Building Effective Agents" research, explained with real code and honest tradeoffs
• A 30-minute walkthrough to build your first working agent with the Claude Agent SDK
• A frank comparison of today's agent frameworks, including which one to pick for which job
• The Company of Zero blueprint: the architecture behind one-person operations that deliver the output of a fifty-person team
• Failure case studies - Air Canada, Klarna, and others - with the lessons most post-mortems miss
• Everything you need to take an agent from prototype to production: observability, evals, cost control, and the team structures that actually work
• A clear-eyed look at ethics, liability, the EU AI Act, and where the human override still belongs
• Scaling patterns for going from one agent to a thousand without losing your mind or your margin
• A 30/60/90 day action plan tailored to founders, engineers, managers, and the professionally curious

This is not a book about what AI agents might do someday. It is a book about what they are doing right now, what it takes to build them well, and what it means for your work, your company, and the shape of the economy you are about to be operating in.

If you are wondering whether to build, what to build, or how to build it - start here.

About the author: Daniel Locke writes about the intersection of AI, work, and the tools we use to shape both. He has been building with Claude since the first Agent SDK release and runs his own experiments in what he calls the Company of Zero.