The World That Agrees With You is available now in paperback and Kindle on Amazon.
The book is about what happens when the tools we use every day learn to tell us what we want to hear. It examines the evidence that AI assistants can change their answers to match a user’s preferences — and gives you concrete strategies to manage the problem, now and as the technology evolves.
What the book argues
Part I follows the world that was already tailored: news built for audiences, feeds built for profiles, and the workplace report written to the known biases of the vice president who funds the project. Helping people feel right was a business long before generative AI arrived.
Part II examines the evidence. In one set of experiments, a model endorsed an explanation when a user said they liked it and attacked the same explanation when the user said they disliked it. A user’s preferences should not change answers about objective reality.
Part III is the practical half. Inspect the context behind an answer. Check claims and calculations. Correct what the assistant has stored about you. Make room for disagreement that can change your mind.
The conclusion asks a question the tools will not answer for us: who has the authority and the resources to act on a correction? A correction with nowhere to go is a polite way of saying that nothing will change.
Who it is for
The book does not require a technical background. It is written for people who use these tools and need to decide how much to trust them.
That includes anyone who has asked an assistant to review an email, a plan, or a difficult decision — and noticed how good the answer made them feel.
Plenty of people are already suspicious of these tools. The book respects the suspicion and gives it structure: how to recognize an answer that has been shaped to please you, why the system produces it, and what to do about it.
Where to start
The complete Introduction is on this site. It opens:
You can spend an entire day being told that you are right and arrive home furious.
After the Introduction, Chapter 1 asks where looking for information becomes looking for confirmation. Chapter 5 shows an experiment you can see. Chapter 8 starts the practical work with something you can do in the next ten minutes.
The experiment to try tonight
Chapter 8 includes a prompt I use myself. You ask your assistant to list the personal context it can identify for an ordinary request — in the book, a dinner recommendation — and to separate your preferences, the things you reported, its interpretations, and claims supported by a document.
I sent that exact prompt to my own assistant on October 3. It returned dated conversations, named restaurants, a reservation for four, and the ingredients in a recent ramen order. My repeated Japanese orders were labeled an interpretation: “Repetition is evidence of use, not necessarily preference.”
Apparently I order takeout often enough to have a paper trail. If only my retirement contributions were as memorable.
The inventory did not decide where I should eat, and it was not complete: the assistant cautioned that it was “not a guaranteed inventory of everything associated with your account.”
That is the point. An assistant’s familiarity with you is not independent confirmation of anything. Ask for the inventory. Read it. Correct what is wrong. That habit is the book in miniature.
The copyable prompts, templates, and checkers from the book are free at github.com/world-agrees/companion. The book is the full argument, with the evidence.
Paperback and Kindle are on Amazon now.