This Is Your AI

That was my moment

This book began with a walk.

A research paper, an idea for a trading system, and GPT-4 producing a couple of thousand lines of code in about three minutes while I was out with the dog. I went home, tested it, made a few small changes, and it worked. I knew something about my working life had changed. My days as a coder were over.

I want to be careful about what that moment was. It was not proof that programming had been solved. It did not show that the machine was infallible; it did not even show that the trading idea was any good. It was the moment I recognised that a different way of working had become possible for me — and decided to act on it.

Everything since has been an attempt to understand that decision and make it work. I have shown you the methods I use, the systems I have built, the mistakes I have made and the evidence for and against my own claims. Now I want to give the decision back to you. Not my decision. Yours.

Don’t take my word for it

These are my ideas. They are not your instructions.

I believe the machine increasingly does the coding and the human increasingly does something else. I believe you can talk software into existence, judge it by using it, think with the machine as well as build with it, choose different intelligences for different jobs and, one day, build software that helps maintain itself. I believe those things because of what I have seen. You have not seen what I have seen, and you will see things I never will.

So challenge me. Find the weak points in my arguments — there are some, and the evidence notes behind this book record several of them honestly. Try my methods and keep the parts that work for you. Throw the rest away. Use different tools. Prefer a different kind of AI. Decide that some of what I call obsolete is still the best way for you to work; for some work it may be. Build an entirely different working environment from mine. Invent things I would never have thought of.

The technology is changing too fast for any method in this book to remain the best for long, mine included. The measure of whether the book worked is not whether you agree with me. It is whether you are now more able to explore and realise ideas of your own.

What makes you different matters

Throughout this book I have kept saying, in one form or another, don’t copy me. Here is why it matters enough to say one last time.

If you follow my tools, my methods and my choices exactly, you will learn something. That is a fine way to start. But you will also tend to make things that look like the things I make, and there is no shortage of those. You carry a different history. You have run into different problems, care about different things, know different corners of the world well, and have a different sense of what is good. You will notice opportunities that I will walk straight past.

Those differences are not decoration. For most of computing’s history, they could not get into software unless their owner also learned to program, or found someone who had. That is changing. AI gives you a way to act on your particular understanding of the world directly.

Being different does not guarantee anything. An original idea can still be a bad one, and nobody owes your work an audience. But becoming a faithful copy of someone else’s working method was never the reason to learn this. The reason is to make things that could only have come from you.

You can begin now

The barrier between having an idea and having working software has changed more in the last few years than in the whole of my career before them. Not every barrier has gone. People still differ in time, money, knowledge and circumstance. Some systems still need deep specialist engineering, and not everything can be built reliably by talking to a machine today. But the barriers have moved far enough that people who were shut out can now take part.

A teacher can try an educational tool. A historian can build a research instrument for the archive only they understand. An artist can make something interactive. An inventor can explore a technology they had only read about. Someone who runs a small business can build software shaped around how their business actually works. Someone with an unusual hobby can make something for the few hundred people in the world who share it. A programmer can become far more ambitious than before.

And you do not have to earn anything from it. You might build a small tool that makes your days easier, or a game with your children, or something that helps your street, or an experiment you are simply curious about. Those are real achievements whether or not anyone pays for them. If you do want to earn a living from what you make, you will need to find the people who value it, and that is its own work. But the ability to make something worth having for yourself is already a great deal.

You also do not need everything in the later chapters to start. You do not need premium subscriptions, a rack of hardware, an elaborate memory system or a team of agents. You need an idea and a machine that can help you explore it. Describe what you want. Ask what it would take. Build something small. Use it. See what it teaches you. Add context, memory, tools and collaborators when your work actually needs them. Chapter 13 began with one small game for exactly this reason: that was enough to start.

This is your time

I do not know which tools will matter in five years, or how fast the capabilities will grow. Nobody does. The large models and the small ones, local and remote, coding agents and research tools, teams of machines and software that looks after itself — all of it will look different soon.

That is why the most durable thing in this book is not any tool. It is a way of approaching technology: notice a possibility, investigate it, try something, look honestly at the result, and change direction when the evidence says so. You can keep doing that however the machines change. It is something more lasting than knowing any one model or interface, and it is yours to keep.

The last chapter described my ambition: software that lives in the sense of continuing to improve. That is what I want to build. It is not the limit of what you can imagine, and you need not believe it is the future at all. You may have a completely different idea of what the next generation of computing should be. Good. Go and find out.

What strikes me most, after thirty years of this, is how short the distance has become between imagining something and holding a first version of it. You can talk your dreams into reality — not every dream, not perfectly, not on the first attempt, and sometimes only to discover the dream was wrong. But you can begin far sooner than anyone could before, and beginning is where everything else comes from.

This is your AI

You do not have to become a programmer like me. You do not have to build my applications or follow my process. You can begin with an idea, talk to a machine, see what it makes, and keep improving it. You can learn what you need as you go. You can build for yourself, for your family, for your community, for people you will never meet, or simply because something interests you.

You may make something nobody else has imagined. You may find out the idea does not work. Either way you will have learned something, and you can try again.

For the first time in my working life, I can imagine a world in which the ability to create software is no longer reserved for the people who learned to program. I do not know where that leads. I certainly do not know what you will choose to build. That is what makes it interesting.

When I came home from that walk, I made a decision about my own work. Now it is your turn to look at what is possible and decide for yourself.

This is your AI.

This is your time.

Go and build something that’s yours.