The Thinking Recession

In late 2024, Oxford University Press surveyed a year of human language and chose two words to define it: brain rot. Usage of the phrase had climbed 230 percent in twelve months. Oxford defined it as the supposed deterioration of a person's mental state from over-consuming trivial, unchallenging content. It was, on the surface, an internet joke. Beneath the surface, it was a diagnosis.

What makes the term striking is its age. "Brain rot" was not coined by a teenager on TikTok; it was coined by Henry David Thoreau in Walden, in 1854. He used it to lament a society that preferred the easy and the shallow over the difficult and the deep. We have been worried about this for 170 years. The difference is that we now have the data — and the data is worse than the worry.

For nearly two decades, the informatics researcher Gloria Mark has measured how long people stay focused on a single screen before their attention moves. In 2004, the average was about two and a half minutes. By 2012 it had fallen to 75 seconds. In recent years it has settled at roughly 47 seconds, with a median closer to 40. Half the time, in other words, a knowledge worker's attention survives less than the time it takes to read this paragraph.

This is not a story about distraction as a personal failing. It is a story about the slow erosion of the substrate that all knowledge work runs on: sustained, deliberate attention. The capacity is not gone — but the habit, the muscle, the infrastructure for thinking, is quietly wasting away.

In 2011, researchers at Columbia published a finding in Science that named a phenomenon many had felt but few had measured. When people know information can be easily retrieved later, they no longer bother to remember it. They remember where to find it instead of what it is. The researchers called it the "Google effect"; the popular term became digital amnesia.

For twenty years, that was the bargain of the internet age, and it seemed harmless enough. We offloaded our memory — phone numbers, facts, directions — to machines, and freed our minds for higher work. But memory was the small thing to outsource. The large thing — the thing that actually distinguishes one professional from another — was never memory. It was judgment. Synthesis. The ability to sit with a hard problem and reason your way through it.

"Memory was always the small thing to outsource. We are now outsourcing the large thing: the act of thinking itself."

And that is precisely what is now happening — faster, and at a scale, that no one has fully reckoned with.

In early 2025, researchers from Microsoft and Carnegie Mellon University published one of the first large studies of what generative AI does to the minds of the people who use it. They surveyed 319 knowledge workers across 936 real workplace examples. The headline finding should give every leader pause.

The more a worker trusted the AI, the less critical thinking they applied. Confidence in the tool was associated with reduced cognitive effort; confidence in oneself was associated with more. As AI took over the work, effort shifted away from analysis and synthesis and toward something thinner — verifying and integrating whatever the machine produced. The researchers warned of long-term over-reliance and a "diminished" capacity for independent problem-solving.

Companies have spent extraordinary sums deploying AI that produces answers. They have deployed almost nothing that helps their people think better. They bought intelligence that replaces cognition and called it transformation — without noticing that, for the workforce, it may be doing the opposite.

"We have given every employee an intelligence that answers. We have given almost none of them one that asks."

Step back and look at what the enterprise has actually built over the last thirty years. It built physical infrastructure — offices, networks, devices. It built digital infrastructure — cloud, SaaS, the modern software stack. It built data infrastructure — warehouses, pipelines, analytics. And now it is building AI infrastructure — copilots and assistants in every workflow.

Each layer made the organization faster. Not one of them made it wiser. There is a layer no organization has deliberately built, because no one has named it.

This is not the same as skills training. Skills training teaches people what to do. Thinking infrastructure shapes how people think before any specific task ever arrives. One gives a person a fish. The other slowly changes how they see water.

Thinking infrastructure is ambient rather than scheduled, conversational rather than passive, and — crucially — compounding. It does not expire when the course ends. It accrues.

If thinking infrastructure has a raw material, it is not information. The internet gave us infinite information and almost no wisdom; the two are not the same substance, and confusing them is the central error of the digital age.

The world's best non-fiction books are something far rarer than information. They are humanity's most refined thinking — pressure-tested over decades, distilled by its sharpest minds, and arranged into frameworks a reader can actually use. A great book on leadership, decision-making, or psychology is not a pile of facts. It is a way of seeing. It rewires the reader's defaults.

The problem was never access to these books. The problem is that the modern professional has no time to read them, no one to discuss them with, and no way to summon them in the moment a real decision is being made. The reading list grows. The wisdom stays locked on the shelf, while the mind that needed it scrolls.

"We are surrounded by the best thinking ever recorded — and starved of the time, and the companion, to absorb any of it."

Imagine an AI that is the opposite of a generic chatbot. A general-purpose model is a brilliant, ungrounded improviser: it knows a little about almost everything and is anchored to nothing in particular. Ask it a question and it will produce an answer — and, as the research now shows, the more you trust that answer, the less you tend to think.

The emerging category is the inverse. It is an AI with a brain of its own — not a blank model waiting for prompts, but one grounded in the accumulated thinking of the world's best books, and designed not to do your thinking but to coach it.

The distinction is the whole point. One kind of AI makes the answer cheaper. The other makes the thinker stronger. Only one of them is infrastructure.

Organizations no longer compete on what their people know. Knowledge has become a commodity any model can recite on demand. They compete, increasingly, on how their people think — the quality of judgment under uncertainty, the originality of an idea, the capacity to lead, to persuade, to read a room and a moment.

These are the only advantages artificial intelligence cannot commoditize — for the simple reason that they are the very faculties AI is, for most people, quietly eroding. The organizations that pull ahead in the next decade will not be the ones that deployed AI fastest. They will be the ones that deployed it wisely — pairing the tools that produce answers with the infrastructure that keeps their people able to question those answers.

That is what it means to treat thinking as infrastructure: to invest in the development of judgment and perspective as deliberately as the enterprise once invested in cloud and data — and to recognize that in an age of abundant intelligence, the scarcest and most defensible asset a company has is a workforce that can still think for itself.