Welcome

Thanks for coming. This is a book about the next ten years, published here in full and free to read.

The argument is short enough to state in a paragraph. Artificial intelligence has begun to automate the one thing that has always limited the pace of progress — the supply of capable minds working on hard problems. Where a field is bottlenecked mainly by thinking, that bottleneck is being removed. Where it is bottlenecked by physics, biology, capital, or politics, it is not. Sorting the first case from the second, domain by domain, is what these sixty-four chapters are for.

Start anywhere. Each chapter is written to stand on its own, with the same five moves in the same order: where the field actually stands in 2026, who is working on it, what the next decade plausibly holds, the second-order effects most analyses skip, and what could go wrong.

---

What the book argues

Sixty-four chapters across nine domains — biology and medicine, energy, transportation, space, AI and robotics, education, government, media, and quantum technology — plus cross-cutting chapters on materials, food and water, cybersecurity, finance, and climate, and a final section on work, longevity, risk, and what to do about any of it.

Each domain gets a historical chapter first. Before asking what the next decade holds for medicine, the book spends a chapter on what happened to medicine between 1926 and 2026, when a scratch could still kill you. That structure is deliberate. The claim that AI might compress a century of progress into a decade is meaningless until you have looked squarely at what an ordinary century of progress actually delivered.

Three confidence tiers run through every forecast, and they are used strictly:

• Near-term likely — the technology exists and is scaling. The question is adoption speed, not feasibility.
• Plausible — the science is understood, the engineering is hard, and it may not arrive.
• Wild — speculative boundary markers, included to show where the edges are, not as predictions.

Where a chapter cannot honestly claim a tier, it says so. Chapter 21 spends its length explaining why faster-than-light communication is almost certainly impossible, which is the opposite of what a book like this is supposed to do.

---

The essay that started it

In October 2024, Dario Amodei — CEO of Anthropic, and one of the people who built the systems in question — published an essay called "Machines of Loving Grace."

It was not about AI risk, which he has written about at length elsewhere. It was an argument that most people are badly underestimating the upside: that powerful AI could compress fifty to a hundred years of scientific progress into five to ten, through what he called a country of geniuses in a datacenter.

What made the essay unusual was not the optimism. Techno-utopian writing is cheap and abundant. It was the specificity, and the source — an engineering judgment from someone who understands the architecture from the inside, and who left OpenAI precisely because he took the risks seriously.

The essay also supplied the analytical tool this book leans on hardest: marginal returns to intelligence. Some fields are bottlenecked by cognition, and adding more of it produces proportionally more progress. Others are bottlenecked by clinical trial timelines, construction schedules, capital formation, or political consent, and no amount of additional intelligence moves them much. Knowing which is which is the whole game.

Progression is an attempt to take that essay seriously and extend it — across more domains, in more detail, with the second-order effects traced out and the failure modes named.

---

How wrong will this be?

Fairly wrong, and the book says so on its first pages rather than burying it.

The specific predictions here will miss on timing, sequence, and mechanism. The history of technological forecasting is a history of confident, well-argued, comprehensively incorrect predictions, and nothing about this book exempts it from that tradition. What it aims at instead is a useful map — a way of thinking about the change that helps you update as evidence arrives. Less wrong over time, rather than right in advance.

---

How to read it

Short on time. Chapter 1 for the thesis, then whichever domain you care about. Finish with Chapter 64.

Policy. Chapters 1, 27, 32–36, 57–60, and 63.

Investing or strategy. Chapter 1, the domains you hold exposure to, and Chapter 62.

Technical. The domain chapters, and the 1926–2026 historical chapters for context on how each field got where it is.

No particular agenda. Start at the front and read it through. It is built to work that way.