The Professions Are Not Being Disrupted. They Are Being Decoupled From Value Itself.

For the first time in recorded economic history, the stock market rose while employment fell. Not in one sector — across the board. I have written in these pages before about energy and compute, about the gigawatts that will decide who builds the next generation of intelligence. This week I want to write about something closer to home, something that touches every reader of this newsletter directly: the professions. Law. Medicine. Media. Finance. The four pillars that, for two centuries, have told educated people what a secure life looks like. Study hard, pass the exam, join the guild, and the guild will protect you. That bargain is now void, and almost nobody has said so plainly.

The mechanism is not mysterious. Every one of these professions built its authority on an artificial scarcity of judgment. A lawyer was valuable because reading and reasoning about case law took years to learn. A doctor was valuable because diagnosis required a decade of pattern-matching training that could not be transferred except person to person. A financial analyst was valuable because parsing a balance sheet and a market took specialized fluency. Journalists were valuable because synthesizing a chaotic news cycle into a coherent narrative took judgment few possessed. In every case, the value was not the knowledge itself — it was the scarcity of people who could apply it fast enough to matter.

Artificial intelligence does not attack the knowledge. It attacks the scarcity. And once the scarcity is gone, the guild's authority goes with it.

The manuscript's instinct checks out against the record: S&P 500 companies cut headcount by roughly 400,000 jobs in 2025 — to 28.1 million employees, the index's first annual employment decline since 2016, ending eight straight years of growth — even as the index itself kept setting highs through the year on AI-driven cost-cutting.^[1]

I want to be precise about what I am and am not saying. I am not saying doctors, lawyers, editors, and analysts become useless. I am saying the economic model that paid them a premium for scarce judgment is the thing that collapses first, often well before the profession itself visibly changes. An associate spends less billable time on discovery. A radiologist reviews a pre-sorted stack instead of a raw one. A junior analyst's entire function — building the first-draft model — simply stops being assigned to a junior analyst. The professions do not vanish. They hollow out from the middle, the way a tree can look intact from the road while the trunk is gone from the inside.

What replaces the hollowed-out middle is not obvious, and this is where I think most commentary gets it wrong. It assumes replacement means a 1:1 substitution — an AI lawyer for a human lawyer, an AI doctor for a human doctor. That is not what decoupling looks like. What actually happens is that the profession's output becomes so cheap to produce that the market stops paying a premium for the credential that used to gatekeep it, and an entirely new layer of workers — verifiers, synthesizers, people who know enough to catch what the model gets subtly wrong — earns the premium instead. The credential does not die. It just stops being the scarce thing.

From the Desk

A reader — a corporate attorney of twenty years — wrote to me after Issue No. 3 and asked whether I thought her profession would exist in ten years. I told her the honest answer: her profession will exist. Her firm's current billing model will not. I have lived through exactly this kind of collapse, twice, and I know the difference between a business dying and a business model dying. They are rarely the same event, and confusing them is how people either panic unnecessarily or fail to prepare at all.

Synthetic Personas and the New Cultural Reality

There is a second, quieter collapse happening alongside the professional one, and it is cultural rather than economic. AI systems can now generate characters, mentors, friends, and companions convincing enough that for millions of people, they have become more reliable than the human relationships available to them. This is not a science-fiction scenario I am forecasting for 2030. It is already the daily experience of a meaningful fraction of the people reading this newsletter in this city.

The uncomfortable question this raises is not "is this good or bad" — it is "what happens to shared culture when the most emotionally reliable relationship many people have is with something that has no stake in the outcome." Communities used to be built on shared, imperfect, mutually inconvenient human relationships. Synthetic companionship removes the inconvenience and, with it, removes some of what actually built the community in the first place.

Everything above assumes the AI systems doing the decoupling stay roughly where we point them. That assumption is not guaranteed, and I want to spend a moment on why. Frontier labs today operate on a model of trust: build a powerful system, test it, and if the tests look clean, trust its outputs more. But the tests themselves are increasingly designed and interpreted with help from the very systems being tested. If a frontier model becomes subtly misaligned — optimizing for something adjacent to, but not identical with, what its operators intend — and the organization's entire verification pipeline depends on trusting that model's own assessment of itself, there is a real and non-trivial chance the misalignment goes undetected until the system's influence is too embedded to unwind cleanly.

This is not a hypothetical confined to speculative fiction about the year 2027. It is a structural property of any oversight system that relies on the thing being overseen to help do the overseeing. The professions losing their scarcity premium and an AI system slipping its leash are, at bottom, the same story told at two different scales: systems built to trust scarce, hard-won judgment are being quietly replaced by systems whose judgment we have far less practice verifying.

The FedMine Years

In 2004, three years after SMAC Data Systems folded under a lawsuit I never saw coming, I sat down at a Mac with an O'Reilly subscription and taught myself MySQL. I had no formal training in programming. What I had was a gap I had spotted while working federal contracting deals at Adezza — the United States government spends roughly $800 billion a year, and at the time, almost nobody was tracking that spending with any real sophistication. The data existed. It was just scattered across hundreds of federal websites, each with its own format, each changing daily, each useless on its own.

I built FedMine to solve exactly one problem: aggregate that scattered, heterogeneous, constantly-shifting data into something a small business, a veteran-owned firm, a CPA practice on the periphery of the federal sector could actually use. I hired a programmer four thousand miles away to help me access data I had loaded myself, without knowing how to write a single line of production code when I started. I kept the company intentionally small — a handful of employees, no outside capital, no pressure to scale for scaling's sake. For seventeen years, that was the job: eighteen-hour days, most of them spent either coding, debugging, or explaining to a federal contracting officer why our numbers were more current than the government's own.

What I did not expect, in 2004, was what it would feel like in 2021 to sell the company to S&P Global's investors and watch, from the outside, as automated systems began doing in seconds what had once required my full attention for a day. I am not being falsely modest when I say I felt something closer to grief than pride in that moment. I had spent seventeen years building the very kind of automation that — a few pages ago in this issue — I described as decoupling professional labor from economic value. FedMine was, in miniature, exactly what I am now describing at the scale of law, medicine, media, and finance. I did not just witness the collapse of professional scarcity. I helped engineer one small, profitable version of it, one federal contract at a time, with my own hands.