The Year We All Got Scared (And Started Reading)
I need to be honest with you from the start: 2025 was the year I stopped pretending to understand what people mean when they talk about AI, and I started actually reading about it. Not the Twitter explainers. Not the think-piece headlines that promise to make sense of everything in 800 words. Real books. The kind that sit heavy in your hands and make you think, “Oh. Oh no. Oh wow, I did not know that.”
And it turns out I was not alone in this minor crisis of comprehension. A January 2026 Gallup poll found that 61 percent of American adults wanted to understand AI better but had no idea where to start. That number stuck with me because it felt like permission to admit confusion. An invitation to stop nodding along in conversations and actually dig in.
The publishing industry noticed this moment too. Oxford University Press reported a 67 percent increase in AI-themed nonfiction submissions between 2023 and 2025. Publishers saw demand tracking our collective anxiety, and they started acquiring books at a pace that felt almost frantic. The 2025 Frankfurt Book Fair, that annual pilgrimage where editors and booksellers track where culture is going, identified artificial intelligence and society as its single fastest-growing nonfiction category. Over 400 titles in that space alone. Four hundred.
So I read a lot of them. Some were revelations. Some were deeply, disappointingly mediocre. And I want to tell you which ones actually made the cut and why some books that should have been essential turned out to be kind of a letdown.
The Unexpected Comeback: Why Suleyman’s Predictions Matter More Now
Here is where my disappointment narrative starts getting complicated. Mustafa Suleyman published “The Coming Wave” in 2023, so technically it should not be on a 2025 list. Except it is. Because he became the CEO of Microsoft AI, and suddenly everyone wanted to reread what he predicted, as if his new position gave his words retroactive authority.
I reread it this year expecting to be underwhelmed by a book I remembered as solid but not revelatory. I was wrong. The book works better now than it did then, which is both encouraging and unsettling. Suleyman’s core argument about the gap between technological capability and our ability to govern it feels less speculative and more like journalism. He walks you through scenarios that seemed theoretical two years ago and now look prescient. The book does not give you easy answers, which disappointed me initially. I wanted him to solve the problem for us. Instead, he maps the problem with such clarity that you understand why nobody has solved it yet.
The disappointment, then, is not with the book. It is with reality for making the book more relevant than we would want. There is something almost cruel about returning to a text and realizing that the author’s warnings have held up better than you hoped they would not.
Kate Crawford and the Data That Changes Everything
Kate Crawford’s updated essays on artificial intelligence and power, published through MIT Press in 2025, did something that felt almost aggressive in its competence. She expanded on her “Atlas of AI” framework with new data showing that five companies controlled over 80 percent of large language model compute capacity globally. Five companies. Eighty percent.
I came to this book expecting the same dense, theoretical analysis that made “Atlas of AI” such a challenge to read the first time around. Crawford does not make concessions to accessibility. She assumes you are willing to sit with complicated ideas about labor, power, extraction, and infrastructure. She does not hand-hold.
But this time the data hit different. The specificity of her ownership analysis, the way she traces compute capacity back to physical locations and material resources and human workers, makes the abstraction of AI suddenly feel very concrete. You cannot read Crawford’s work and pretend that AI is this neutral technological phenomenon. You read it and see the scaffolding that holds it up, and that scaffolding is made of inequality.
My frustration with this book was that it reinforced how little structural change has happened. Crawford is arguing essentially the same thing she argued in 2021, just with updated numbers showing the concentration got worse. The book is excellent. The situation it describes is bleak. MIT Press AI and Society catalog 2025 has multiple entries trying to grapple with these questions, and they are all worth your time if Crawford’s work speaks to you.
The Frankfurt Fair Phenomenon: 400 Books and Diminishing Returns
I spent time browsing the Frankfurt Book Fair 2025 trends report, looking at the 400 titles in the AI and society category, and I felt something I did not expect: fatigue. Not from reading them all, obviously. But from the sheer repetition of themes, the way so many books were making similar arguments in slightly different registers.
There were books for the technologist. Books for the ethicist. Books for the person worried about their job. Books for the parent worried about their kids. Books for the policymaker. This is not necessarily bad. Different people need different entry points. But after reading fifteen of them, you start noticing the pattern. A lot of these books are playing slight variations on a central anxiety without offering much that is genuinely new.
That gap between the sheer quantity of books being published and the quality of distinct thinking happening across them is real, and it is frustrating. Some of these 400 titles are original and brilliant. Many are competent repackagings of arguments that have already been made. The market flood makes it harder, not easier, to find the books that actually matter.
What Actually Worked (And What Did Not)
The books that stayed with me after 2025 were the ones that either gave me information I genuinely did not have or made me feel something real about the stakes. Crawford delivered both. Suleyman, on rereading, delivered on clarity even when clarity just confirmed the worst. The books that disappointed me were the ones that seemed to exist primarily because there was market demand for AI books, not because the author had something specific to say.
I keep thinking about that 61 percent of people who want to understand AI better. They deserve actual guidance, not just volume. They deserve books that respect their intelligence and their time. Which is why I kept reading.
Tell me what you have been reading. Have you found the books that clarified things for you, or are you still wading through the noise? I want to know what stuck with you and what felt like a waste of time. That conversation, the one where readers actually talk to each other about what worked and what did not, that feels like the real book community to me.
