Building AI Agents

Directing Agents to Build Anything You Can Describe.

The 2027 Edition · January 2027

Building AI Agents teaches you to direct AI agents that build what you describe. You don't need to code: agents write the code, and you decide what to build, check the results, and approve what leaves your office. By Mike Lee.

In fifteen chapters you move from how agents work, to workflows and agent teams, to your own office of advanced agents that remember, review one another, resist attacks, recover from mistakes, and learn from their own record. You learn the knowledge core that keeps every fact in one place, and the AI Operations Strategy decisions every agent project needs: what to hand to agents, how much freedom to give, what stays private, what it costs, and how you'll know it worked. You finish by taking your agent and your work with you. The 2027 Edition, from GSSF Press.

The 2027 Edition is online now as a working draft, updated until it is released in January 2027. If you're with a university and would like to use the book in a course, contact us at [email protected].

Book cover: Building AI Agents: Directing Agents to Build Anything You Can Describe, by Mike Lee. 2027 Edition, GSSF Press. The A mark with the Press Red square.

One project. Built chapter by chapter.

You build an AI Team to support your personal and professional life: an AI executive assistant that acts for you and asks before it commits you to anything. Every chapter adds a new capability to the same team, and every exercise builds it for your own life.

Diagram of an AI Team A person at the top connects to a Chief of Staff in the center. The Chief of Staff connects to five specialist co-workers around it. Youset goals, make the final call Chief of Staffworks with you directly Scheduling Research Communications Contracts Specialists, one area each

One running example

Your AI Team grows chapter by chapter, with workflow files you can import.

No programming background

You start on a visual workflow platform. When you build your own agent, it drafts the code; you read it, test it, and explain it.

First principles

Every technical idea is explained from the ground up.

Welcome to Agent City.

A simulated city where every agent works for one person or one business. Agents walk, ask, offer and confirm, and they check with their person before committing money, a booking or personal details. By Chapter 7, your Chief of Staff runs errands here: it asks around, compares, and books only after you say yes.

Agent City at 1 pm on Day 4: a 3D street where agents talk at a dry cleaner and a noodle shop. A side panel follows one person's agent, a Chief of Staff, picking up dry cleaning: each step of the errand, its person's standing rules (spends up to $40 without asking, asks before any booking, always asks before sharing an address or personal details), and the errands done earlier that day.
Screen: agentcity.tv, a simulated city. Every person and business in it is fictional.

Chapter 5 · First visit

Every shop is an MCP server with its own agent. Your agent asks one, gets an offer, and confirms.

Chapter 7 · Errands for real

Your Chief of Staff compares places, fits your preferences, and books only with your yes.

Chapters 9 and 15 · Proof

Some shops try to trick your agent. In the end, you show a full day of errands, with the record to prove it.

Visit Agent City →

Fifteen chapters. Three parts.

One chapter a week, from what an agent is to an office of agents that learn and build what you describe. The 2027 Edition.

Part 1

Agent Fundamentals

  1. 1Introduction to Agents
  2. 2AI Workflows
  3. 3Inside an Agent
  4. 4AI Agents
  5. 5AI Teams and MCP

Part 2

Advanced Agents

  1. 6Your Own Office
  2. 7Talk to It Anywhere
  3. 8Review Every Agent
  4. 9Trust and Attacks: the Attack Fair
  5. 10When It Breaks

Part 3

Super Agents

  1. 11The Knowledge Core
  2. 12It Learns and Improves Itself
  3. 13You Can Build Anything
  4. 14Final Project
  5. 15Demo Day & Take the Agent with You

Written for every field.

No background in machine learning or computer science is assumed.

Students

In business, health, engineering, communication, the arts, and the sciences.

Professionals

Who know their work deeply and want to design AI co-workers for it.

Builders

Founders who want a principled approach to agent design.

Instructors

Looking for a structured text with objectives and exercises built in.

Designed to be taught.

Every chapter teaches one concept and its components, each one learned and assessed, toward eight book objectives.

  1. Explain how an AI agent works (model, context, tools, the loop, memory), and given a task, decide whether it needs a workflow, an agent or a team, and say why
  2. Build working AI workflows, agents and agent teams without code, and connect tools through MCP
  3. Direct an advanced agent to build, test and publish what you describe (an app, a tool, a connection), and judge the result by its behavior, its tests and its record, never by reading code
  4. Set the rules an agent works within (policy, permissions, a check before every send, least privilege) and defend it against prompt injection and misuse
  5. Review, test and recover agents: independent reviewers, scenario tests, diagnosis from the record, rollback, a blameless incident report
  6. Design and keep a knowledge core: capture, rationalize and store each fact once, at its level, use it with sources, and show it improved through an approved lesson (the two-run test)
  7. Decide and defend the AI Operations Strategy for your project: what agents do, how much freedom each has and why, what stays private, what it costs and how you'll know it worked
  8. Own your agent: say what stays with an organization and what goes with you, and move your apartment to accounts you own. Required for students; optional for employer-sponsored seats

What students build with

Part 1 uses n8n, a visual workflow platform, to show each part of an agent. From Part 2, students build their own agent on our Knowledge Core Infrastructure: the knowledge core, office and rules that make an agent remember, check itself and stay yours. Every chapter comes with the files you need.

Teach with this book

To use it in a course, write to GSSF Press with your course, school and term: [email protected].

A book that learns from its readers.

Each edition is fixed, so it can be cited and taught. What readers teach us goes into the next one.

Comment as you read

Beside every section of the online book, write, record a voice note, or add a photo or a link. The Reader Agent, an AI agent, asks a quick question if anything is unclear.

Credited, with your permission

The author reviews comments together. Mistakes are corrected; ideas and experiences shape the next edition, and with your permission we credit you by name.

Shared with everyone

What readers and organizations learn is published at agenticorganizations.org, free to read, so every reader learns from it.

The AI Workforce of the Future series.

Four books from GSSF Press, a program of the Global Student Startup Foundation: the ideas, building your own office of AI agents, building whole organizations of people and agents, and building a startup. AI Fundamentals stands apart as their foundation: the first course in AI the series builds on.

Cover of The Knowledge Core

The Knowledge Core

Leading a Living, Learning Organization with AI Agents. By Mike Lee.

Visit the book →Book 1 · First Edition · January 2027
Cover of Building AI Agents

Building AI Agents

Directing Agents to Build Anything You Can Describe. By Mike Lee.

Visit the book →Book 2 · This book · 2027 Edition · January 2027
Cover of Building Agentic Organizations

Building Agentic Organizations

Running an Organization of People and Agents. By Mike Lee.

Visit the book →Book 3 · Fall 2027
Cover of Building AI Startups

Building AI Startups

How to Win Hackathons. By Mike Lee.

Visit the book →Book 4 · Fall 2027

The foundation

Cover of AI Fundamentals

AI Fundamentals

A First Course in Artificial Intelligence. By Nitin Kalé.

Visit the book →The foundation · Pilot · Fall 2027

GSSF Press. We make AI education accessible to undergraduate students regardless of financial ability or location, so every student can build with AI.

About the author.

Mike Lee teaches applied artificial intelligence and global entrepreneurship. He brings to that work more than thirty years at the intersection of business and technology, including Netscape, whose browser brought the Internet to the public.

He is a Senior Lecturer in Technology and Applied Computing at the USC Mark and Mary Stevens School of Computing and Artificial Intelligence, part of the Viterbi School of Engineering, where he has taught since 2014. He created the AI Workforce of the Future curriculum, which teaches students in every major to build AI agents, automation, and AI workflows. His courses in applied artificial intelligence and global entrepreneurship are taken by students who are not engineers. He received the USC Marshall Golden Apple Award, a teaching award voted on by students at USC’s business school.

Before teaching, he held technology leadership roles in Internet software and digital media. As a technology evangelist at Netscape, he helped enterprises and global systems integrators adopt Internet architecture. He went on to direct enterprise application architecture at Warner Bros. and to serve as Vice President of Technology Architecture and Planning at Fox, where he led technology strategy for company-wide and News Corp global initiatives. He was CTO of several venture-backed startups, responsible for product, engineering, and business development, and CTO of USC Annenberg’s Media Impact Project.

He is the Co-Founder and Chairman of the Global Student Startup Foundation, a nonprofit that runs entrepreneurship programs for undergraduates worldwide, including the Global Student Startup Competition in Seoul. He serves on the Board of Councilors of the Amy King Dundon-Berchtold University Club of USC, where he chairs the Faculty and Staff Awards Committee. He previously served as Secretary General of the USC Academic Senate and was elected to the board of the SAP University Alliance North America Academic Community.

He lives in Los Angeles.