Foreword

Demos of complex applications built in minutes. Conference stages feature engineers describing how they shipped features that once took weeks in a single afternoon. People on social media telling how many tens of thousands of lines of code they have generated with AI.

Many of them have a bridge to sell? You bet. My experience, shared by many of my colleagues, has made me quite sceptical about the boldest statements. Yes, wonderful tools, but still hopelessly unpredictable and unreliable unless you keep a very close eye on them or you're just after quick wins. Larger scale, reliable, maintainable software? That's a different story.

There was no single moment of failure that made me write this book. It was more of a progression. As our ambition level grew beyond "fix this" tasks, the standard off-the-shelf features started to fall short in a team setting. Agents tried too hard or gave up too easily, made shortcuts, or produced uneven output. The single instruction files, MCP servers, and bolt-on tools added to confusion rather than reducing it, and at times it felt like things got worse rather than better. What I gradually realized was that the gap between "impressive demo" and "reliable delivery" is not a tooling problem. It's a governance problem.

It's more complicated than that. AI-assisted development can indeed make you dramatically more productive, but it is not a magic wand some seem to believe. Getting the right things right requires a lot of hard work, and the organizational challenges are significant. In some cases they might even outweigh the benefits of the AI-powered approach.

The basic question I set out to answer with this book is: what does it take to get your AI-generated app beyond demo/PoC quality in a sustainable way?

Who this book is for

I wrote this for the people who have to make AI-assisted development work in practice: engineering leaders, delivery leads managing scope and timelines, architects concerned with system integrity, and senior developers who understand that tools are only as good as the processes surrounding them.

If you are looking for a tutorial on how to use Cursor, Copilot, Claude Code, or any other specific tool, this is not that book. The tools change too quickly, and the real challenges are not in the tools themselves but in how we organize around them. I won't be handing out specific prompts that will make AI generate perfect code. I won't be making you more assured that AI will solve your software delivery problems.

Why this book, why now

There has never been a better time to write this book and to start taking things into practice. Looking back my 40+ years of programming with all hypes, new things coming and going, I have never seen a technology with such a potential to change how we work. If somebody had told me 30 years ago that one day we would tell the computer to write very complex software for us and it would do it, I would have thought that was pure science fiction. And yet here we are.

This book was inspired by my previous work on the topic, including a book together with an amazing co-author, JP Heimonen.

Another motivation were the needs of a real project on the horizon and the long-forgotten draft table of contents I had pieced together. For the real application, and considering myself relatively well informed on this topic, I realized we need to build something on top of the tools to make them work for us in a sustainable way. The standard ones, even the best ones, were frankly not good enough for the kind of work we needed to do. How come they are still stuck in a single developer, single session paradigm with naive approaches to context management? None of this is gonna work in bigger setups.

This book, however, has been my personal ambition.

I've used the AI intensively for pretty much everything ever since the beginning. I think that only in the last few months both models and tooling have reached a state of being good enough for more ambitious applications, and I certainly feel that the time is right to share the insights and practical guidance I've gathered along the way to help colleagues on track with this with a realistic and responsible approach.

Hope you find some useful insights in the book.

And for those who know me, expect whatever you find inside to be littered with (bad) jokes, complicated sentences and a healthy dose of skepticism.

About the author and acknowledgements

I started programming over 40 years ago and have been doing it professionally for almost 30. I've been an architect, a developer, a team lead, and occasionally all three at once. I've survived waterfall, survived Agile, survived SAFe, and now I'm trying to survive AI. Along the way I've worked on everything from tiny startups to large enterprise systems, and I've collected enough scar tissue to fill a book. This one, apparently. Much of what's in here I learned from colleagues and customers over the years. They are far too many to name, but you know who you are.

Throughout, I have tried not to oversell what I think is possible right now. AI-assisted development is moving fast, and some of what I write here will need revision as the field evolves. But I argue that the underlying principles about accountability, traceability, and the necessary boundaries between your judgment and automation will last.

Many thanks to all colleagues and customers along the way for their patience.

Tornio, Finland March 2026 Antti Koivisto

Want to reach out? Find me on LinkedIn or send an email.

E-mail: [email protected] LinkedIn: linkedin.com/in/antti-koivisto-24646a

P.S. In case you disagree on my points, or think they're a load of BS, I'll respond by quoting a former politician. There was a rumour of him saying, 'I once thought that I was wrong. But I was mistaken.' So bit like him, I'm gonna stand by my work but reserve my right to be wrong.