Introduction: The Age of One-Person Startups Powered by AI#

A founder working alone, surrounded by orbiting tool agents

AI Went from a Conversation Tool to “Something That Finishes the Work”#

The progress of artificial intelligence (AI) technology has been nothing short of dazzling. The AI field, which had stagnated until the early 2000s, finally broke through around 2006 when deep learning research opened a path forward, and in 2016 it captured the world’s attention when Lee Sedol, a 9-dan Go professional, was defeated by AlphaGo. Then, in November 2022, OpenAI released ChatGPT, and AI went from a technology confined to research labs to a tool that anyone uses every day.

Yet looking back at the few years since ChatGPT, you can see that the direction of change has shifted one more step. At first, it was “a tool that answers when you ask questions, as if talking to a person.” Then came the ability to recognize and generate images, audio, and video, followed by the ability to call external tools to look up real data or carry out tasks. And now it has evolved into the agent form: give it a goal, and the AI makes its own plan, calls tools repeatedly, checks the results, and pushes on until the work is done. Agents that write code are already an everyday tool for developers, and agents that operate browsers and business tools are following close behind.

Each time the way we use AI changed, the work left for humans changed with it.

This shift matters because “the work left for humans” has changed. In the past, the ability to build things by hand was itself the competitive edge. Now, deciding what to build, setting the standards for how far automation may go, and reviewing and taking responsibility for the results have moved up to become the human’s job. A large share of the work once done by hand can now be handed to agents.

That Is Why: One-Person Startups#

Companies will change shape along with this. Companies that already have a stable product will minimize their maintenance headcount and grow the number of small teams exploring new directions, while startups building their first product will try to launch an initial version with minimal staff using AI before hiring people for each function. The “one-person startup” in this book’s title points exactly there. It means we have entered an age in which it is realistically possible to launch an initial product and scale it to a certain level alone, or with a very small team.

But do not misread the word “alone.” The “one” in one-person startup is not a person who does everything by hand; it refers to the human side of an organization made up of one decision-maker and multiple agent workers. The founder sets the goals, priorities, constraints, and quality standards, and reviews the results. Writing code, researching, drafting documents, and routine operational tasks are performed by agents. Making this structure run is the core skill of a one-person startup, and the Execution section of this book covers how to build it.

The "one" in one-person startup is not someone who does it all alone, but the decision-maker of an organization of agents.

Who Wrote This Book, and Why#

I am an application developer, currently building services that run on AI. At work, I am rebuilding the development process around coding agents, putting AI features into products, and managing their cost and safety. On my own time, I build and operate a few projects together with agents, including this blog, as an ongoing experiment in “how far can one person go alone?”

This book is an interim report on that experiment. It contains the key AI concepts you need to know, the services and tools you can use, and the problems I actually ran into while building, operating, and promoting products on my own, along with my current conclusions about them. When I first started writing, I focused on organizing concepts and tools, but in the meantime the way agents work has changed dramatically, and based on that experience I added a new Execution section.

What This Book Covers, and What It Does Not#

This book has six parts. The first three parts are the Foundations, which organize the key concepts, services, and tools of AI. The last three parts are the Execution section, which covers how to work with AI and everything from finding an idea to building a product, adding AI features, operating, marketing, and dividing up work when you are on your own. The overall structure is summarized in the diagram on the table of contents page.

Let me also be clear about what it does not cover. The “general procedures for setting up a company,” such as incorporation, taxes, fundraising, and HR, are outside the scope of this book. There are already plenty of good resources on those topics, and they are not directly related to AI. This book focuses on “how to use AI technology to build and run a product on your own.”

Target Audience#

Because this book was written by a developer and describes how AI technology can be put to practical use, it is aimed at developers or those with a reasonable technical understanding of software development. Understanding server-client architecture and the concept of an API is enough. The Foundations were written so that non-developers can read them too, but the Execution section mixes in talk of code and infrastructure.

If you already use AI actively in your work, you can skip the Foundations and start from Part 4. Conversely, if you are just getting started with AI, I recommend reading the Concepts in Part 2 first and then dipping into the Tools in Part 3 only for the chapters you need.

One more thing: in the AI field, models and products change every few months. This book avoids information that goes stale quickly in the main text wherever possible, and marks the places where it is essential with an “As of” note. The principles and structure of the book should last, but please keep in mind that discussions of specific products may have changed by the time you read them.