Dated thesis, not measured displacement — October 11, 2026: This February 26 essay and its existing October 6 reader note separate market repricing from a proved loss of economic value. The signal-log figures below lack a linked price-series calculation and cannot isolate AI as a cause. Builder warnings and my five-layer argument are forecasts or interpretations, not measured universal outcomes. Employment, professional accountability, deployment limits and business adaptation can change how capability becomes an economic result.
For all of modern economic history, human intelligence was the scarce input. That was the whole game.
You knew something. You charged for knowing it. You analyzed, reviewed, compared, translated, processed. The knowing was the value. The fee was the premium.
Lawyers charged it. Accountants charged it. Consultants charged it. Developers charged it. Analysts, researchers, translators, underwriters, financial advisors — all of them built careers on the same foundation: I understand something you don't, and that understanding is worth money.
I charged it too. For years. My entire business was knowing how systems worked and helping people navigate them.
That premium is dying. And I watched a piece of it die in real time over three weeks in February 2026.
Three Weeks
I keep a signal log. Every day I track what's happening — not stock prices, but structural shifts. Things that tell you what the economy is becoming, not what it did yesterday.
In early February, my signal log went from interesting to alarming.
Feb 5.
Anthropic launches integrations that handle finance, legal, sales, marketing, and customer support workflows. Software stocks crater. Two trillion dollars in market cap gone. Worst stretch for the software index since 2008.
Feb 9.
A startup in Madrid launches an AI insurance application. Willis Towers Watson — one of the largest insurance brokers on earth — has its worst trading day since 2008.
Feb 10.
An AI tax tool hits the wealth management space. Raymond James posts its worst day since COVID. Schwab drops 7%.
Feb 12.
A former karaoke machine company. Six million dollar market cap. They demo an AI freight logistics tool. Twenty-three billion dollars wiped from the trucking sector in a single session.
A karaoke company.
Let that sit for a second.
Feb 20.
Anthropic launches a tool that finds security vulnerabilities in code. Fifteen billion gone from cybersecurity stocks in a day.
Feb 23.
A 7,000-word blog post on Substack contributes to an 800-point Dow drop. IBM falls 13% — worst day since 2000.
Michael Burry shares the post and writes: "And you think I'm bearish."
What I'm Not Saying
I want to be careful here. Because it would be easy to turn this into a clean narrative — "AI killed seven industries in three weeks" — and that's not quite right.
Markets move for lots of reasons. The software selloff had DeepSeek and valuation concerns mixed in.
The Dow drop wasn't caused by one Substack post — it was a confluence of fear. Some of these moves will partially reverse. Some of these companies will recover.
I'm not saying the market got it perfectly right in three weeks.
I'm saying the market is telling you something about direction. And the direction is: the premium that companies charged for human intelligence — for knowing things, understanding things, processing things — is getting repriced. Fast.
Not every company in that list will go to zero. But the business model they share — charging a premium for human cognition — is compressing. And markets are forward-looking enough to start pricing that in, even before the actual displacement hits at scale.
The specifics and direction remained uncertain; those price moves did not prove that expertise had lost all economic value.
What Died
I grouped several affected companies under one thesis, but they had different business models and market exposures.
Someone who understood something you didn't. Someone who could analyze what you couldn't. Someone whose brain was the product.
Insurance brokers who understood risk. Cybersecurity firms that understood vulnerabilities. Wealth managers who understood tax optimization. Software companies that understood workflow. Trucking logistics firms that understood supply chains.
AI made that understanding available for the cost of a subscription.
Not perfect understanding. Not complete understanding. But good-enough understanding at a price point that makes the human premium indefensible for most use cases.
This is what I lived through personally. My business was built on understanding systems that my clients didn't understand. When AI could explain those systems — imperfectly, but well enough — the gap I was charging for collapsed.
I decided to sunset that strategy before I thought its economics would become worse. The companies in my signal log had different circumstances; the log does not prove they needed to make the same choice.
Author chronology clarification — October 11, 2026: The clarified Chapter 6 account says the decision to sunset SimpleDirect’s fintech business was made around December 2025, the official pivot happened around June 2026, customers were given two years to transition, and the app remains available and online. The strategic decision described here is not an immediate December service shutdown or an independently measured guarantee of operational health.
The Builders Are Warning You
Here's the part that should genuinely unsettle you.
The CEO of Anthropic — the company building Claude — said publicly that AI could eliminate half of entry-level white-collar jobs. Push unemployment to 20%. Within five years.
The person who created Claude Code said "software engineer" as a job title starts disappearing this year.
These aren't critics. These aren't bears looking for attention. These are the people building the technology, telling you what it does.
When the person who made the weapon tells you to take cover, you should probably listen.
But most people aren't listening. Because the displacement hasn't hit them personally yet. Their job still exists. Their paycheck still clears. The disruption is something that happened to software stocks and trucking companies and insurance brokers — not to them.
That's what Phase 1 always feels like. It's happening to other people. Until it isn't.
What Can't Be Compressed
I've spent months thinking about what survives this.
Not as an intellectual exercise. As a survival question. When the thing you built your career on — knowing things, understanding systems — becomes worthless, you have to figure out what you have left.
Here's what I landed on. Five things AI can't do. I use these as personal responsibility categories, not permanent technical limits established for every AI system or deployment.
Identity.
AI can't be you. It can generate content in your style, mimic your voice, approximate your perspective. But it can't be George Pu sitting across from someone who needs to hear a hard truth from a person who's been through it. The being-there is the value. Not the information.
Relationships.
AI can't call your lawyer at midnight because you've known each other for ten years and he trusts your judgment. It can't sit at a dinner in Dubai and build the kind of trust that leads to a handshake deal two years later. Relationships are built on shared time, shared risk, and demonstrated integrity over years. There's no shortcut. There's no prompt.
Stakes.
AI doesn't take risk. It doesn't put its money, its reputation, or its future on the line. When I decided to move away from those business strategies, that was a real decision with real consequences that I have to live with. AI doesn't make those decisions. It generates options. The human who picks one and bets on it — that person has something AI never will.
Selection.
AI generates a hundred options in a second. It can't tell you which one matters. That requires judgment. Judgment requires experience. Experience requires scars. The ability to look at infinite possibilities and say "this one, not that one" — that's not intelligence. That's taste built from living.
Accountability.
AI isn't responsible for outcomes. When things go wrong — and they will — someone has to own it. Say "that was my call." Stand behind it. In a world full of people hiding behind algorithms and committees and disclaimers, the willingness to be accountable is becoming the rarest thing there is.
That's what survives.
Not what you know. Not what you can analyze. Not what you can process.
Who you are. Who trusts you. What you're willing to risk. What you choose. And whether you stand behind it.
The Question
I'm not writing this to scare you. I'm writing it because I've been through the thing that's coming for a lot of people, and I'd rather say it clearly now than watch people figure it out the hard way.
My thesis was that some fees for information processing would compress; I had not established that all expertise premiums would reach zero. Not for everything. Not overnight. But the direction is unmistakable, and the speed is faster than anyone expected. Three weeks in February made that visible to the markets. I did not have evidence for a three-year timetable covering everyone.
The people who are already rebuilding around the things AI can't replace — their identity, their relationships, their willingness to take real stakes — will navigate this.
The people who are still charging a premium for knowing things will watch that premium evaporate, and it will feel like the ground opened up, because the thing they spent their career building was never as solid as it seemed. It was solid because intelligence was scarce. Intelligence isn't scarce anymore.
The question isn't whether this is happening.
The question is whether you're still charging for something that's becoming free.