The Economy
Knowledge was expensive. Now it isn’t.1 Everything built on that expense today is, economically, lying about what it sells.
The professions’ own certification exams are now passable by machines. The bar exam, all three steps of the USMLE, nearly half of real-world software engineering tasks. The exams the professions built to certify that someone had acquired knowledge are passable by systems with no training.
The Cost Crash
Take three questions professionals answer daily. Change the country. The answer changes.
Same patient on a ventilator. Three countries. Three answers to whether the person is dead.23 Same symptoms presenting in three systems.45 Same CRUD operation. Three completely different data models.67
Pick a profession. Pick a jurisdiction.
No single practitioner knows all nine. A competent lawyer knows three. A cross-border doctor knows two. A junior associate with an LLM knows all of them — not the literature, not the history, but the operative standard. The thing you’d bill for.
This is acquisition cost collapsing. Acquisition: the years of training, the university degree, the local practice — and the pattern recognition built as a result of that. The answer to “how do I tell if someone is dead” is different jurisdictionally. No one questions that. You would have the wrong answer if you were a lawyer in the wrong country — so you pay for local practice.
The answers are now freely available without the local practitioner. The examples are small. The pattern is obvious and growing.
The Argument
“I want advice on my complex legal structure across Singapore and Taiwan.”
The traditional firm takes 10 hours. The AI-enhanced firm takes 2.
Both firms have the same billing rate,13 but not the same output rate. The 8-hour gap is acquisition cost made visible.14 The time a human spends finding, reading, and applying what the machine already had is a significant time cost. The knowledge is in both. The willingness to accept liability exists in both. The only question is: how much labor does it take to retrieve that answer?
The moat is liability. The human, signing off; “I checked this.” The doctor signs off, the lawyer files, and the engineer rubber stamps. The argument goes that the knowledge economy survives because someone accepts responsibility when the machine is wrong.
But the example killed this and it’s clear. Both firms accept liability. The AI-enhanced firm signs off in 2 hours. The traditional firm signs off in 10. The AI-enhanced firm does its due diligence on the retrieved data, runs a quick audit, and calls it. The traditional firm tasks associates, trainees, interns, junior partners, researchers, to, in grueling fashion, search, index, and discover related cases, statutes, and legal frameworks. The end product is the same. The liability coverage is the same. The liability moat doesn’t distinguish those two firms.
And the 10-hour search isn’t more thorough. The associate in hour 7 is tired, skimming, pattern-matching on memory. The machine isn’t. Ten hours of human research is not five iterations of two-hour machine research — it’s one slow pass that feels rigorous because it hurt.
But would you pay 5x because a human did it, even if it’s identical?
That’s a nostalgia economy with a billing structure.15 Nostalgia reprices fast — once the comparison is visible. Most clients haven’t seen both outputs side by side. The ones who have aren’t paying 5x. When you look at the 2-hour output and put it next to the 10-hour output and can’t tell the difference, the 8-hour premium is outright indefensible.16 The traditional firm can say “I did it by hand.” The client says “So you’re just charging me more for being outdated?”
If acquisition cost is zero, graduates from Harvard and graduates from wherever else have the same access rates to knowledge in terms of speed and quality. You hired Harvard students because they had better filters to entry, professors, and peers. That led to better pattern exposure — fact recognition and case understanding. This premium depends on acquisition being costly and hard. The liability signature now authenticates the same thing regardless of origin or capability. The client gets two identical memos and two very different billing rates. The internal hierarchy of costing and knowledge in the profession goes first, long before “lawyer” becomes unrecognizable as a profession.
This is already observable.
What’s left? Taste, judgment, or aesthetic preference. One framing over another, one type of grammatical syntax over another, when both are factually correct. The preference is real and economically legible. But how much will you pay for work that isn’t better — just slightly more aesthetically pleasing — when you’re trying to be productive? The AI annotation to the right is doing the same work I am. Possibly better. Taste is what’s left when the productive value is gone.
What happens when knowledge and credentialing come apart? The economy moves slow. Knowledge migrates from production to consumption. You and I keep acquiring it, whether through Substacks or hefty university degrees, because we like knowing things. This preference is genuine and well-asserted. It also doesn’t have the same price tag. Knowledge-as-investment gave us $200k degrees in prestigious elite universities.8 Knowledge-as-consumption gives you $10/month Substacks and Coursera. Same content. Different wrapper.
But the pricing tells you what the value of it really is.
The False Precedent
We’ve seen this promise before.
Coursera, 2012. Harvard and MIT put their lectures online. Stanford opened its AI course to 160,000 students. The pitch: access is the bottleneck. Remove it and education democratizes.
It didn’t.9
Completion rates across HarvardX and MITx fell from 6% to 3% between 2013 and 2018. Not a plateau — a decline. 52% of registrants never opened the course.10 Six years. No improvement.
The standard explanation is motivation. True and irrelevant. The structural explanation is that MOOCs gave you the lecture without the capacity to use the lecture. A Stanford machine learning course assumes linear algebra. A Harvard contracts course assumes you can read a case. The content required the prerequisite. The prerequisite required the pipeline. The pipeline was the $200k.
LLMs break this.
Go back to the grid. The chest pain question. A MOOC on cardiology would teach you about vasospastic angina in week 8, after weeks 1–7 of prerequisites you may or may not have. An LLM gives you the applicable differential now. “If this presentation, in this system, then this.” Not the textbook chapter. The clinical reasoning. The productive component — the thing the doctor was trained for, delivered without the training.
That’s the difference between access to knowledge and access to the application of knowledge. MOOCs offered the first. LLMs offer the second. The first still needed the $200k pipeline to be useful. The second doesn’t.
So. The person who asked the LLM a medical question and received a competent, jurisdiction-aware response — they acquired the productive component. For free. They don’t need the lecture.
The person who still wants the lecture after getting the answer — that’s the consumption market. They have the output. They want the process. They want to know, not just to have known. That preference is genuine.11 It prices like a hobby.
The MOOC market is worth $22.8 billion.12 Coursera charges $10–$79/month. The word “open” now means audit-mode videos stripped of graded assessments. Knowledge-as-investment became knowledge-as-consumption. The platforms adjusted. The pricing tells you — same as the billing rates, same as the $200k degree. Look at what things cost. The price is the confession.
Close
Knowledge was expensive. Now it isn’t. Now you can press a button and get a damn good AI summary of everything I have to say. You’d learn as much there. You’d learn as much just reading the annotations on my work. What kept you reading? It wasn’t economic productivity, the strength of my argumentation, or credentialing. There’s no economic defense for spending an extra 5–10 minutes on my work when you can do it in 2. It’s what I can’t tell you. It’s something resembling aesthetic preference. If you can name it, an LLM can do it.