Preface to the Second Edition

By Tristan Kromer

It’s been nine years since the first edition of this book was published, and quite a lot has changed.

In 2026, uncertainty is the norm. There have always been black-swan events and unexpected serendipity, but even legacy businesses in stable sectors are now struggling with geopolitical, economic, environmental, and technological uncertainty. But the most obvious disruption that affects this book is the rapid adoption of AI technologies, specifically Large Language Models (LLMs).

Nine years ago, this book encouraged entrepreneurs to test, test, and test again to discover and validate demand before building a product or service. We go around the Build-Measure-Learn loop as quickly as possible to learn what’s worth building. The reason was simple — creating the wrong product costs time and money. It’s risky to commit capital before knowing you’ll get a return.

Today, LLMs have made it possible for an individual entrepreneur to create websites with little to no technical experience. Entrepreneurs can prototype physical products with a 3D printer. Drug companies are developing new treatments based on AI-driven simulations. AI has compressed the time to build by ~10x.

It used to personally take me a day to launch a landing page experiment by hand coding a Ruby-on-Rails app. Now it takes me 5 minutes to prompt and 30 minutes to go get a cup of tea while my agents do the work.

But while LLMs, 3D printing, and simulations have reduced the cost (and thus risk) of building products, two risks have increased: overbuilding and defensibility.

Overbuilding

Most products don’t fail because they have too few features. They fail because they have too many. Too many options confuse users and make products hard to use for anyone but experts. Chat-based interfaces are training users to expect everything done for them with little to no effort, while entrepreneurs are building modular dashboards with hundreds of bells and whistles that no one actually wants.

It’s easy to run multiple projects simultaneously, but toggling back and forth between ten chat windows autonomously building software doesn’t get customers in the door. It takes research and experimentation to understand what truly matters to customers, and where to focus. Without that focus you’re just building software that no one wants ten times faster.

Entrepreneurs require even more discipline now than ever to find the real pain points that get people in the door, engaged, and paying.

Defensibility

The old moats — first-mover advantage, accumulated codebase, copy-protection — are gone. Businesses can’t assume that a technological advantage exists for long, if at all. A dozen other entrepreneurs can copy your business on a whim. Years of accumulated software know-how can be rebuilt in a long weekend. Physical products can be offshored to on-demand manufacturing. Services can be wholesale copied by a stack of agents.

Today, a sustainable business has to be built on tight customer segmentation and hard-to-copy insights backed by a continuous improvement loop. The technology isn’t defensible, but the ability to quickly identify insights and stay one step ahead of competitors remains a key competency. Again, that takes discipline and focus on discovery and validation.

The research and experiment methods we first wrote about in 2015 are more valid than ever for building a real business — but the methods themselves needed a complete refresh with new techniques, new technology, and new advice.

What changed since the first edition

This edition is a refresh for the age of AI. And yes, it was done with the help of AI. AI drew the illustrations, ran research, fact-checked drafts, and argued with me about each page. Every page has been rewritten with both human and synthetic hands.

The full changelog lives in Version History, but the big changes are:

New Methods

We’ve added nearly thirty new research and experiment methods since the first edition, including AI-forward methods like:

Disposable MVPs — working software generated in hours by an AI agent.

Synthetic persona screening — LLM-generated stand-ins for customer segments to test value propositions.

AI-moderated interviews at scale — letting AI take the lead on customer interviews.

AI prompts

Every method needed to be updated with AI-native advice. That includes new lists of tools, new case studies, new reading lists, and most importantly, new copy-and-paste AI prompts to help you both plan the work and analyze results.

If you’re reading on an ebook reader and can’t find them, look for footnotes with links to the online version which is free to use.

Lists of tools

We’ve added lists of commonly used tools for each method to our online version. We don’t have any sponsorships and we didn’t list every tool out there because more are being released each week.

If you think we missed an important new tool, please send us a message and we’ll include it in the online version.

Improved navigation

The new “At a glance section” shows common tags, time commitment, costs, and which section of the Business Model Canvas the method reveals.

The “In Brief” section now has a common use case so you can quickly decide if it’s the best method for you right now.

Improved biases and tips

The common pitfalls have been updated — providing useful advice on how to avoid confirmation bias, overanalyzing data, or letting AI hallucinate insights on your behalf.

What didn’t change

Real willingness-to-pay still requires real humans putting real money on the table. Product/market fit is still validated by word-of-mouth and retention, not by an AI panel rating your pitch.

Trust, emotional resonance, and the felt sense that a product belongs in someone’s life are not simulatable (yet). A synthetic persona can guess at what a customer might say. But it will not tell you whether a stranger will hand you their credit card.

If you read the first edition

The methods in this book that worked in 2015 still work. They’re just faster to run, and the failure modes are different.

This is still an index of research and experiment techniques to pick and choose, just retooled for an environment where everyone has an AI agent on the desk next to them.

If you read the first edition, you can jump right in. The structure of the four quadrant index is still the same, but the methods are described in much more detail. Skip ahead to the Index of Methods to get started right away.

Help us improve

As always, this rewrite follows the same philosophy we take in everything — continuously improve. As we get feedback, we’ll incorporate it.

So if you see something missing, know a good case study, or spot something that’s just outright wrong, please let us know. This book is only as good as it is because it was written by real practitioners with real experience, and that includes you.

So read on… and build something amazing.