In the Age of Abundance, What's Missing?

For most of the last two years, the argument about AI and people has been an argument about jobs — how many disappear, how fast, whether the machines are coming for the work. It is the wrong argument, or at least the incomplete one. The number that matters is not how many jobs AI removes. It is what happens to the human being who stays.

Two things are happening to that person at once, from opposite directions, and almost no one is looking at both.

The pipeline, from one side

Start with what the data actually shows, because the headline version is misleading in a way that discredits anyone who repeats it.

Aggregate unemployment has not moved. Three independent efforts reach the same place using different data and different methods. Anthropic's economists, working with U.S. survey data, found no systematic rise in unemployment for the most AI-exposed workers since late 2022. An IMF analysis of Danish payroll records, covering tens of thousands of workers, found no significant effect on wages or hours. And Stanford's AI Index reports that large-scale aggregate job losses have not materialized in the overall employment data. If you lead with "AI is destroying jobs," a serious room will stop listening — because the best available evidence says it hasn't, not in the aggregate.

But the aggregate is hiding the story. Look at the edge, and the picture sharpens.

In Canaries in the Coal Mine?, Erik Brynjolfsson and his co-authors used payroll records from ADP — millions of workers, month by month — and found that early-career workers, ages 22 to 25, in the most AI-exposed occupations experienced a sharp relative decline in employment, controlling for firm-level shocks, while employment for experienced workers in the same occupations held steady. Overall employment kept growing. Employment for the youngest workers in exposed jobs did not.

The metaphor is the whole point. Young workers are the canary — not because they matter less, but because they register the gas first. And the reason they register it first is the reason that should stop every reader cold: their value comes disproportionately from codified knowledge — the book-learning of formal education, the procedures that can be written down, the tasks that can be checked. That is precisely the knowledge AI closes the distance on soonest. Experienced workers are protected, for now, by tacit knowledge — the judgment, the client sense, the process instinct that was never written down and cannot yet be automated.

This is the mechanism, stated by the researchers themselves: AI is automating the codifiable, checkable tasks that historically justified an entry-level headcount, while complementing the judgment-intensive work of people who have been there longer.

The bottom rung of the ladder is being sawn off. Not the ladder — the bottom rung.

And there is a second finding in the same paper that matters more than it first appears. The declines show up in the applications of AI that automate work — that substitute for the person. They do not show up where AI augments — where it makes the person better. Same technology, two surfaces, opposite effects on the human. Which surface an organization builds toward is not a technical detail. It is the entire question.

The mind, from the other side

Now the pressure that tends to get overlooked — or under-reported.

While displacement squeezes the entry pipeline, something quieter is happening to the people who keep their jobs and use the tools every day. The early evidence suggests the tools may be eroding the very capacities the era most demands.

The most concrete finding is clinical, and it was published in The Lancet. Researchers tracked endoscopists at four centers that had adopted AI assistance for colonoscopy, then measured how those same doctors performed on standard colonoscopies without the AI. Their adenoma detection rate — the core measure of whether they catch precancerous growths — fell six percentage points after AI exposure compared with before it. When the machine was there, they detected well. When it was taken away, they detected worse than they had before they ever used it.

Measured decline — in exactly the domain where a human is supposed to be the last line of judgment.

Other work points the same direction. A Microsoft Research and Carnegie Mellon study, published at a peer-reviewed venue, surveyed several hundred knowledge workers about their AI-assisted work and found a revealing pattern: the more a person trusted the AI, the less critical thinking they brought to its output — while the more they trusted their own judgment, the more they engaged. The effect is self-reported, so hold it accordingly. But the mechanism is the one that matters: reliance is a dial, and turning it up turns your own scrutiny down.

The paper's own phrase for what's at risk is a person's "cognitive musculature" — left atrophied when the thinking is handed off.

Source: Lee et al., "The Impact of Generative AI on Critical Thinking" — Microsoft Research & Carnegie Mellon, CHI 2025. 319 knowledge workers, 936 task examples.

Softer still, but pointing the same way: an early study from MIT's Media Lab used EEG to compare people writing essays with an AI, with a search engine, and with nothing. The AI group showed the weakest neural connectivity and struggled afterward to quote work they had just produced. It is a preprint, the sample is small, and its own lead author has publicly pushed back on the "brain rot" headlines it generated — so it is a signal, not a verdict. I include it only because it points the same direction as the sturdier evidence, not because it proves anything on its own.

But the direction is consistent, and the logic underneath it is not mysterious. Offload a capacity and it weakens — this was true of navigation before it was true of judgment. The concern is only that we are now offloading the capacities that don't grow back easily, and that we most need to keep.

The vise

Put the two together and see the shape of it.

From one side, displacement is pressing on the pipeline — foreclosing the entry-level path where people used to accumulate the tacit knowledge that later protects them. From the other side, cognitive offloading may be dulling the attention, memory, and judgment of the people already inside. The person in the middle is being pressed toward reactivity by the threat, and possibly toward diminished capacity by the tools meant to help.

Notice what this does to the ladder. The bottom rung is being cut by automation. And the middle rungs — the years where codified knowledge slowly ripens into tacit judgment — are exactly the years the tools may be hollowing out. If you can't get onto the ladder, and the ladder itself erodes the faculties that climbing was supposed to build, the promotion the whole system depends on — the shift from executing work to governing it — starts to look less certain than everyone is assuming.

Because that promotion is the thing the entire agentic transition is built on. The story every organization is telling itself right now is that AI absorbs the execution and humans move up — into orchestration, judgment, governance, the work that only a person can do. That story is correct. But it contains a condition almost no one is stating out loud: the move up only happens if the human is ready to make it.

An unprepared person does not rise into the governance layer. They get stranded at an execution layer the agents now own.

What's missing

Which reframes the whole problem.

For a century, organizations were built on an assumption: intelligence is scarce and coordination is expensive. AI has broken both. Intelligence is becoming abundant and cheap. And when the abundant thing floods in, the scarce thing is whatever it can't replace.

That scarce thing is not intelligence. It is human readiness — the regulated attention, the durable judgment, the capacity to think clearly and adapt under acceleration, the standing to govern a system rather than be quietly governed by it. The organizations that win the agentic transition will not be the ones with the best models. Everyone will have the models. They will be the ones whose people were ready to move up when the execution moved down.

This is the part the readiness conversation keeps missing. We talk about readiness as an organizational property — decision rights, governance, absorption capacity — and it is that. But underneath the org chart is a more fundamental layer, and it is human before it is structural. You can redesign the operating model perfectly and still fail, because a redesigned structure staffed by people who can't think clearly under pressure is just a better-drawn diagram of the same problem. Structure cannot calm a threatened nervous system. Structure cannot restore an atrophied judgment. Those have to be built in the person.

That is the work almost everyone skips — because it is the hardest to build and the easiest to fake. It doesn't show up in a tool procurement or an org-chart redraw. It shows up only in whether a specific human, standing at the helm of more intelligence than any generation has ever commanded, can hold the wheel steady.

The scarce thing in the age of abundance was never going to be intelligence. We are building intelligence by the acre. What's missing — what's actually running short while everything else floods in — is the human capable of governing it. And we are, right now, doing two things at once that make that human rarer instead of more common.

So the practical question is how you build and protect that person. But sit with it a moment longer and a harder question opens underneath it — the one the data can point to but can't answer.

What is readiness actually made of? When two people face the same acceleration and one collapses into reactivity while the other holds the wheel steady, the difference isn't intelligence — both may be brilliant, and the endoscopy data suggests brilliance is exactly what erodes first. It isn't information; they have the same information. What separates them is something we don't have a metric for: the quality of attention they can sustain, the ground they operate from, whether they can stay clear while everything speeds up. We have been calling it judgment, but judgment is the output. The thing underneath it doesn't have a clean name in the language of org charts and payroll data.

That is not a gap in the research. It is the point where the research runs out — where you leave the territory that data can measure and enter the one that has always been the real subject: not how capable the human is, but how present. The whole AI age, underneath the displacement charts and the deskilling studies, is quietly forcing a question the culture has spent a century avoiding — what human capability actually rests on, and whether the people at the wheel can see clearly enough to steer.

This piece is about the floor: the measurable ways the human is being pressed from both sides. The floor beneath the floor — what readiness ultimately rests on, and why intelligence and consciousness are not the same axis — is the through-line of everything else I write. If the evidence here left you asking not how many jobs but what it takes to stay clear while the ground moves — to stay present, and to keep judgment intact, when everything around you is accelerating — then you've already found the door.

Reggie Britt is a technologist and executive who has spent decades at the intersection of enterprise systems, consumer finance, and emerging technology. He writes about AI, human judgment, and what it actually means to lead through transformation.