From the Desk of Ashok Mehan
I woke before six and did what I do every morning now — reached for the phone before my eyes had fully adjusted to the dark.
WhatsApp first. The alumni group: sixty-three people I have known for decades, some since the classrooms of the late 1980s where we argued about the future of computing when computing was still something that happened in dedicated rooms with raised floors and controlled humidity. Smart people. Accomplished people. People whose judgment I have respected across a long span of professional life.
Three new messages since midnight. A photograph: someone's granddaughter at a birthday party, icing on her nose, pure joy. A response chain of fifteen heart emojis and laughing faces. And then a link — someone sharing a news story about AI with the caption: "Interesting stuff — AI might be getting smarter than us! 😂"
Four people liked it. One replied: "Ha! We'll always have the edge on common sense." And that was that. The thread closed. The group moved on to debating whether the reunion should be in Lisbon or Barcelona.
I lay there in the half-dark, phone on my chest, thinking about those sixty-three people. Not with frustration — with something more complicated. These are not unintelligent people. They built careers, companies, hospitals, practices. One ran a federal agency. Several have published books. Their children are themselves professionals now, carrying the family expertise forward into a world that is quietly being reconstructed beneath their feet.
And almost none of them feel the urgency.
I have been trying to write this dispatch for three weeks. Every time I sit down, I feel the peculiar loneliness of knowing something you cannot quite transfer to the people you care about. The wave is not forming. It is here. And the beach is full of people discussing Barcelona.
So this week I write to my classmates. Not to them specifically — they have not subscribed to this newsletter, which is a separate problem I am choosing not to examine too closely. I write to the type they represent: the educated, experienced, well-connected people of a certain generation who built their careers on the premise that expertise is hard to acquire and therefore valuable, and who have not yet been forced to confront what happens when that premise is suddenly and permanently wrong.
The people in my classmate group are not uninformed. They read. They follow the news. They have opinions about AI — measured opinions, calibrated by decades of watching technology overpromise and underdeliver. They say things like "it's a tool, not a replacement" and "there's no substitute for human judgment." They cite the dotcom bust. They cite the hype around big data in 2012. They are not wrong, exactly. They are wrong about the time frame — and that is a much more expensive kind of wrong.
Here is the structural problem. The people most qualified to understand technological disruption are, paradoxically, the ones most psychologically protected from feeling it. A partner who has spent thirty years building a practice, a physician whose diagnostic intuition has saved lives, a financial advisor who navigated clients through 2008 and 2020 — these people have genuine expertise. Their expertise is genuinely valuable. The markets confirmed it, year after year, with compensation and status and the particular satisfaction of being the person in the room who knows the answer.
That confirmation has become a kind of trap. It is very difficult, from inside a career that has worked, to feel the urgency of a shift that has not yet touched you directly. The first-year associate whose research work is being replaced by Claude is not a senior partner's peer. She is invisible to the senior partner. The GP whose diagnostic time is being compressed by AI triage tools does not appear at the conferences the senior specialist attends. The disruption is happening below the waterline, and the people above it are dry and comfortable and discussing reunion venues.
"The most dangerous kind of wrong is not ignorance. It is expertise applied to a world that no longer exists."— Ashok Mehan
AI researchers have a phrase for this that I keep returning to: inference on outdated weights. A model — any model — generates outputs based on the parameters it learned during training. If those parameters were fixed at a particular point in time, the model continues to produce confident, fluent outputs even as the world it was trained to represent drifts further from what it currently looks like. The outputs are not lies. They are simply stale. And they arrive with exactly the same authority as outputs that are accurate.
I know what this feels like from the inside because I did it at FedMine. I would answer competitive questions with the particular ease of a man who has been in many rooms — fluently, quickly, with apparent conviction — from knowledge that was sometimes two years out of date. The machinery of expertise had kept running well past its information's expiration date. This is not a character flaw. It is a structural feature of biological memory. The problem now is that the gap between when a mental model was formed and what the world currently looks like is growing at a rate our nervous systems were not designed to register.
The Anthropic Economic Index — published in early 2026 and representing one of the first serious attempts to measure how AI is actually being used in professional work — contains numbers that should be in every WhatsApp group that includes a physician, a lawyer, a financial advisor, or a senior executive.
Forty-three percent of conversations with large language models are work-related. Not personal curiosity. Not entertainment. Actual professional tasks, being delegated to AI by someone, right now. Of those work conversations, the top occupational categories include legal services, financial activities, and management consulting — the exact domains my classmates have spent their careers building.
Fifty-one percent of all AI interactions involve what researchers call "augmentation" — a human using AI to enhance and extend their own judgment. This sounds benign. It is benign, for the person holding the prompt. The question is what it means for the professional whose billable hour was previously the only route to that judgment.
And then there is the number that I think about most: the cost of AI reasoning has fallen by a factor of 390 over the past two years. Not thirty-nine percent. Not threefold. Three hundred and ninety times cheaper. The economic barrier that once protected professional expertise — the sheer cost and difficulty of providing it — is not eroding. It is collapsing. And it will keep collapsing.
When I talk to my classmates — individually, in the ways that get past the social surface — I notice that they cluster around four positions. I am not caricaturing. I have heard each of these articulated, in substance, by people I respect.
The first is the Tool Argument: AI is a productivity tool, like a spreadsheet or a search engine. It amplifies what professionals do; it doesn't replace them. This is true for the first wave of adoption and wrong for the second. Spreadsheets did not replace accountants — but they reduced the number of accountants required per unit of accounting work by roughly seventy percent over two decades. The first wave of AI looks like a tool. The second wave looks like a structural reorganization of the labor market for professional services.
The second is the Judgment Argument: there is no substitute for human judgment, especially in high-stakes situations. Also true. The senior lawyer's strategic counsel, the physician's management of a genuinely complex case, the advisor's behavioral coaching through a market panic — these will command a premium for years, perhaps decades. But they cannot sustain the current pyramid. Law schools produce 35,000 graduates a year in the United States alone. The profession can no longer absorb them at anything like current terms. Judgment at the top does not save the pipeline below.
The third is the History Argument: technology has always created more jobs than it destroyed, and this will be no different. This is the most seductive of the four, because it has been true — repeatedly, over two centuries. The mechanization of agriculture, the electrification of manufacturing, the computerization of clerical work — all of these displaced workers and all of them, eventually, were followed by net employment growth. The honest version of this argument acknowledges two things its proponents usually omit: first, "eventually" can mean decades of genuine hardship for real people; and second, this acceleration is faster than any previous one by an order of magnitude.
The fourth is silence — the position of the person who sent a single laughing emoji and went back to discussing Lisbon. This is the most common response. It is not stupidity. It is the rational behavior of someone who has a full life, a functioning career, and no immediate evidence that anything is wrong.
I have not forwarded this dispatch to my classmates. Not yet. But if I could distill what I want them to know into something small enough to read between the laughing emojis, it would be this:
In 1987, when we were sitting in the same rooms, I took AI seriously and most of you had already made up your minds about where your careers were going. Those were good decisions, well-executed. You were right. The market confirmed it across thirty-five years of professional life.
The difference now is not that you were wrong then. The difference is that the world those decisions were optimized for is being reorganized at a speed that has no historical precedent. The transitions that took decades before are taking years. Some are taking months. The people who will navigate this best are not the ones who panic — they are the ones who start paying attention before the evidence becomes impossible to ignore.
The wave is not coming. It is here. It is in your hospitals, your law firms, your advisory practices. The question is not whether you will need to adapt. The question is whether you adapt on your own timeline or on the market's timeline. The market's timeline is less forgiving.
"Nobody on the beach notices a tsunami until the water pulls back. By then, the time for a measured response has already passed."— Mehan Dispatch
I am not writing this from a position of superiority. I sold FedMine five months before ChatGPT made everything I had built look like a rehearsal. I spent a year wondering whether I had missed the most important opportunity of my professional life. That discomfort — the particular sting of watching the future arrive just after you left the room — is not something I would wish on anyone. It is, however, something that sharpened my attention in ways that being comfortable does not.
Working Table — The Classmate Risk Matrix (Professional Disruption, 2026)
| The Archetype | What They Believe Right Now | What They Are Missing | The Window for Adaptation |
|---|---|---|---|
| The Law Partner | "AI is a junior associate tool — it helps with research and drafting. Strategy is still mine." | First-year associate work down 30% since 2024. Clients using AI before calling the firm. The pipeline that creates future partners is contracting. | 3–5 years before client expectations on cost and speed permanently reprice strategic counsel |
| The Senior Physician | "Medicine is about relationships and clinical judgment. AI can't sit with a patient." | AI diagnostics for common presentations now match specialist accuracy. JAMA research: patients rate AI responses as more empathetic in 64% of cases. The relationship argument is eroding. | 5–7 years before institutional adoption of AI triage and diagnostics reaches mainstream practices |
| The Financial Advisor | "Clients trust a human face. Robo-advisors are for simple portfolios." | Robo-advisors now manage $3.7T. AI generating CFP-level financial plans in minutes. The complexity threshold above which a human is needed is rising fast. | 2–4 years; fee compression already visible; move up the value chain now or reprice later |
| The Senior Executive | "I use AI tools. I'm already adapting. This is a productivity story, not a threat story." | Organizational decision-making being augmented at every level. The executives who thrive will be the ones who can evaluate and govern AI outputs — not just delegate to people who can. | Ongoing; the executive who cannot interrogate an AI recommendation is already at a disadvantage |
| The Comfortable Retiree | "I'm out of the game. This doesn't affect me." | Pension funds, healthcare systems, estate planning, and the financial infrastructure of retirement are all being reorganized by AI. Your children's professions are directly in the path. | N/A for career — but the conversation with adult children about their trajectories is overdue |
From the Book · Memoir Passage
In 1987, I sat in a graduate seminar on artificial intelligence at a time when artificial intelligence was, to most of the world, either science fiction or a joke. The field had survived one AI Winter already — the period in the 1970s when the gap between what researchers promised and what the technology could deliver became impossible to ignore, and funding dried up and careers migrated elsewhere. The second Winter was coming, though we did not know it yet. We were in the brief warm interval between failures, convinced that this time the tools were finally serious.
My classmates — some of the same people who are now in the WhatsApp group debating Lisbon — were in adjacent rooms making adjacent decisions. Computer science, yes. Engineering, yes. Business. Law. Medicine. The decisions were not wrong. They were rational optimizations given the information available. AI in 1987 could play chess adequately and recognize a limited set of spoken words under controlled conditions. The case for treating it as the dominant technology of the next century was not, at that moment, obviously compelling.
I was the unusual one for taking it seriously, and I knew it, and it made me somewhat unpopular at certain dinner tables.
What I remember from those rooms is not the technical content — the expert systems, the backward chaining, the symbolic reasoning frameworks that would spend the next decade producing impressive demonstrations and disappointing real-world results. What I remember is the quality of the disagreement. These were people who argued from evidence, who changed their minds when presented with better data, who had the intellectual honesty to say "I don't know" when they didn't know. I trusted their judgment in 1987. I still trust it.
"The irony is not that my classmates don't understand AI. The irony is that they understand it too well — they remember when it didn't work, and that memory is now the thing standing between them and the urgency of the present."— From the manuscript
Here is what has changed and what has not. The quality of their reasoning has not changed. The intellectual honesty has not changed. What has changed is the information environment — and specifically, the signal-to-noise ratio in the AI coverage that most educated people encounter. For every sober analysis of what AI actually does and what it costs and who is deploying it and with what results, there are a hundred breathless predictions and a hundred skeptical dismissals. The coverage is calibrated to produce strong feelings, not updated beliefs. And strong feelings, in busy professionals with full lives, tend to resolve into the path of least resistance: not paying close attention.
I built FedMine on a version of the same assumption my classmates are now making — that the cost and complexity of a particular kind of knowledge work formed a durable moat around the business. I was right, until I wasn't. The moat didn't collapse slowly. It didn't give me a graceful exit ramp. It drained in months, not years, and the timing made it invisible to me until after it had happened.
What I want to say to my classmates — the ones who are physicians and lawyers and senior executives and comfortable retirees — is not: you are wrong. It is: the world you are correctly reasoning about is no longer the world that exists. The weights you are running on were formed in a different environment. The inference is fluent and confident and stale.
Forty years ago, I earned a postgraduate degree in information systems with a major in artificial intelligence, when AI was a punchline. I am spending my retirement watching it eat the world. Forty years, phew — and the most astonishing part is not that it finally arrived. It is that the people who most deserve to understand what is happening are the ones least likely to be paying attention in the way the moment requires.
The classmate group is still debating Lisbon versus Barcelona. I have booked neither. I am here, writing dispatches about what is coming, hoping that some of these words eventually find their way into that group and land with the weight I intend.
What I'm Reading This Week · For the Classmates Who Are Not Yet Reading It
Mollick is a professor at Wharton who has done more than almost anyone to translate AI capability into practical, grounded guidance for working professionals. This is the book I would put in every classmate's hands first — not because it is alarmist (it isn't) but because it is honest about what adaptation actually requires. He introduces the concept of the "jagged frontier": AI is remarkably capable at some tasks and surprisingly poor at adjacent ones, and the frontier is not where you expect it to be. Required reading before you form an opinion about what AI can and cannot do in your specific field.
Acemoglu is an MIT economist and winner of the 2024 Nobel Prize in Economics. This book is his answer to the "technology always creates more jobs" argument — and it is a rigorous, historically grounded one. His finding: the gains from technological progress are not automatically distributed to workers. They go to whoever holds the power to define what the technology is for. This is not a pessimistic book. It is a precise one. It asks the question my classmates should be asking: not "will AI be good or bad?" but "good and bad for whom, and on whose terms?"
Frey is the Oxford economist who, in 2013, published the paper estimating that 47% of U.S. jobs were at high risk of computerization. The paper was criticized — often by people who had not read it — for being too alarmist. This book is the long-form version of the argument, grounded in the full sweep of technological history from the Industrial Revolution forward. His core insight is that transitions which benefit society in aggregate can be devastating for particular workers and particular communities — and that the speed of the current transition is unlike any previous one. For the classmates who lean on the "technology always creates more jobs" history: read this first.
An unexpected entry on this list, and I include it deliberately. Oprah's book is not about technology. It is about paying attention — to evidence, to discomfort, to the signals that the world is trying to give you before those signals become impossible to ignore. I read it for the first time last year, in a strange period of post-FedMine reflection when I was trying to understand why I had not seen things that in retrospect seemed obvious. The discipline she describes — of holding your certainties loosely, of remaining genuinely curious in the face of disconfirming information — is exactly the discipline that the current moment demands from every one of my classmates. Some books find you at the right time. This one found me.
Same argument from the other side: why the people who learned to search badly — and paid for it — are the ones best equipped for what AI now amplifies.