A typical book contains around 70,000 words. Most of those are common: “the,” “and,” “if,” “but.” But what gives a book its meaning isn’t the individual words. It’s how they’re arranged: what they refer to, imply, or reveal when placed in sequence.
Large language models work the same way. They don’t “know” things in the traditional sense. Instead, they operate on probability, stitching together patterns of words that sound plausible based on vast amounts of training data. They’re sensitive to phrasing, punctuation, and even spacing - a missing comma can shift the outcome.
So when we talk about “designing with AI,” what we’re really talking about is designing around structure: how information gets shaped, delivered, and interpreted.
That’s the premise behind my book, Designing AI Interfaces, which will be published by O’Reilly in early 2026. I started writing it because I couldn’t find a practical, well-reasoned guide to designing with AI that wasn’t too technical, too fluffy, or entirely centered on chatbots. And because like a lot of designers working in AI, I kept running into the same problems - vague inputs, unpredictable outputs, users blaming themselves when the model got it wrong.
At some point, I realized that designers needed a rosetta stone to understand the implications of AI on the products they're creating; giving them grounded examples and delineating clearly what's actually net new.
So what is the book about?
Designing AI Interfaces is a guide for product designers who want to build AI features that are actually usable. It’s not a technical manual (though you’ll learn a lot about how models work). And it’s not an ethical manifesto (though ethics absolutely matter). It’s a book about interfaces — about the places where human intentions meet machine predictions, and what happens in between.
Some chapters focus on structure: how to design input patterns that help the model succeed, how to organize outputs for different task types, how to handle latency, errors, and model weirdness in ways that support user trust.
Others dig into newer territory: How do we support exploration in open-ended systems? What should “feedback” mean when your AI product doesn’t improve with it? How do we design for subjective questions, where there’s no right answer?
There are examples from consumer tools, enterprise workflows, medical interfaces, coding assistants, and a lot of odd little edge cases that I’ve come to love. There are also diagrams. So many diagrams.
Who is it for?
This book is for designers who are working with — or about to work with — AI. If you're comfortable with Figma but fuzzy on what an embedding is, you're in the right place. It’s written with the assumption that you know how to build products, but want to better understand the moving parts of an AI-powered one: where the design ends and the model begins (hint: it’s blurry), what your team should be asking at each stage, and how to build things that actually help users get things done.
It’s also for PMs, researchers, and curious engineers — anyone who’s trying to navigate the weird and wonderful world of AI product design without reinventing every wheel (or adding a chatbot to every feature).
Why now?
Because the tools are new, but the challenges are familiar. Users still want clarity, control, context. They want systems that work — and when they don’t, they want to know why.
The rise of large language models has introduced some new magic, but also new uncertainty. The job of the interface is to hold those things together — to support exploration without confusion, automation without overconfidence, intelligence without illusion. That’s not something we can copy-paste from last year’s design patterns. We have to build new ones.
This book is my attempt to do that: to document what’s emerging, offer a set of shared principles, and give designers the language and tools to work confidently in this space.
I’ll be sharing more here in the coming weeks — chapter previews, process thoughts, design patterns that didn’t make it into the book, and a few hot takes about “magic” buttons.
Thanks for reading. If you're curious about this stuff too, I’d love to have you along for the ride.
—Louise