You've spent twenty years building judgment that takes a junior a decade to even recognize. Right now, that judgment is worth somewhere between $40 and $90 an hour to a knowledge-extraction platform — and zero dollars to the LLM that already ingested your blog, your podcast appearances, and the PDF of your framework somebody uploaded to Scribd in 2019.
That's the actual price of "staying human" in 2026. Not a philosophical price. A line item.
So the question isn't whether to resist AI. The question is which parts of your expertise you keep human-owned, and which parts you let the machine swallow on its terms instead of yours.
I'm Matt Cretzman. I build systems for experts who refuse to be raw material. Here's the frame I use, and the stack underneath it.
Mollick Asked the Identity Question. The Economic One Is Harder.
Ethan Mollick's "Choosing to Stay Human" is the most honest piece written about the felt experience of working alongside AI. It names the unease. It refuses the easy answer. Good.
But identity is downstream of economics. You don't get to choose to stay human if the market has already priced your human contribution at zero. You don't get to "do the work yourself for the satisfaction of it" when the work no longer pays rent.
The people writing thoughtful essays about whether to use AI are mostly tenured. The expert in the wilderness — the consultant whose retainer just got cut, the coach whose course platform is suddenly a ghost town, the analyst whose firm is quietly running pilots — that person needs a different question.
The question is: what specifically do I refine into IP I own, before someone else refines it for me?
That's the skill refinery. Not nostalgia. Not a vibe. A discipline.
The Three Extraction Economies (Pick One — You're Already in One)
In the book I'm writing, I lay out three extraction economies. You are in one of them right now. There is no fourth option called "opting out."
Extraction #1 — LLMs took it without asking. The UK copyright saga is the cleanest example. 88% of creators wanted protection. The government punted. Your work is in the training data. The terms were not yours. The compensation is not yours. The attribution is not yours. This already happened.
Extraction #2 — Platforms pay you hourly to dump it. Mercor's CEO has been admirably blunt about the business model: pay domain experts $40-$200/hour to externalize the judgment that took them careers to build, package it, sell it to frontier labs as training data. You get an hourly rate. They get a permanent asset. Jacob selling his birthright for stew, except Jacob at least got a hot meal.
Extraction #3 — You refine and own it. Same raw material. Same judgment. Different terms. You define the cards, the system, the distribution, the price. The machine still helps — it just helps you, not someone monetizing you.
Mollick's essay is really a meditation on Extraction #1 and #2 without naming them as extractions. "Choosing to stay human" inside someone else's extraction economy is just choosing to be raw material with feelings about it.
What the Skill Refinery Actually Is
I named the company Skill Refinery on purpose. A refinery is not a content factory. A refinery takes crude — messy, mixed, low-value in raw form — and outputs distinct, named, priced products.
For an expert, the crude is your experience: every client engagement, every diagnostic call, every framework you've redrawn on a napkin, every pattern you recognize in the first ninety seconds of a conversation. Most experts die with that crude un-refined. Their firm captures some of it. Their team inherits a sliver. The rest evaporates.
The refinery does four things, in order:
1. Extract the judgment from the engagement. Not the deliverable. The decision tree behind the deliverable. The five questions you actually ask. The three signals you watch for. The one trap you've learned to avoid.
2. Encode it into transferable artifacts. Cards, frameworks, prompts, diagnostics, scoring rubrics. Objects, not vibes.
3. Stack the artifacts into a Knowledge Delivery System. A sequence a client or learner can move through with or without you in the room.
4. Own the distribution rails. Your domain. Your list. Your payment processor. Not a platform that can reprice you Tuesday morning.
Notice what's not in that list: "produce more content." The refinery is the opposite of the content treadmill. The treadmill is Extraction #1 with extra steps — you generate raw material faster, the models eat better, you stay broke.
The Part Mollick Got Right (And the Part Worth Stewarding)
Mollick is right that some things are worth doing yourself even when a machine can do them faster. I'd push the language. It's not about doing them yourself. It's about stewarding them.
There's a thread in how I think about this work that I won't argue for, but I'll name: the expertise you've built is a gift. Twenty years of pattern recognition is not a commodity input to someone else's model. It's something you were entrusted with. Stewardship means you don't sell it for stew, and you don't let it evaporate either. You refine it, you compound it, you pass it on with integrity.
That's the part of "staying human" worth keeping. Not the romance of the manual. The responsibility for what you know.
The machine isn't the enemy of stewardship. Cheap extraction is.
The Numbers That Force the Choice
Here's the math that ended the debate for me, and for most of the experts I work with at Skill Refinery and Stormbreaker Digital.
- Mercor-style platforms: $40-$200/hour, capped, with the asset transferring to the buyer in perpetuity.
- A refined Knowledge Delivery System priced at $2,400 with a 6% conversion off a 4,000-person list: $576,000/year, asset retained, distribution owned.
- An LLM that ingested your work without consent: $0 to you, indefinitely, with your phrasing now showing up in someone else's mouth.
You do not need three of those to be true. You need one. The asymmetry between refined and raw is the entire game.
And the asymmetry is widening every quarter. Every model release makes raw expert hours cheaper and refined expert systems more valuable. The middle — "I'll just keep doing it the old way and hope" — is the position getting crushed.
What This Looks Like on a Tuesday
Concretely, for an expert reading this with a calendar full of calls and a vague sense the floor is shifting:
This week, pick one engagement type you've run more than ten times. Open a doc. Write down the five questions you ask in the first call, the three signals that tell you which path the engagement takes, and the one decision the client almost always gets wrong without you. That's a card. That's the first product off the refinery line.
Next week, do it again with a second engagement type. Now you have two cards. Stack them. That's a diagnostic.
In ninety days, you have a system. In six months, you have an offer that runs whether or not your calendar is open. In twelve, you have a moat that the next model release strengthens instead of erodes — because the model is helping you deliver your refined judgment, not replacing it with averaged judgment from a thousand consultants who never bothered to refine theirs.
That is what choosing to stay human looks like when you take the economics seriously. It's not standing athwart the machine yelling stop. It's building the refinery before someone builds it around you.
On Whose Terms
I'm writing a book about this — On Whose Terms: The New Expert Economy and the Fight for What You Know. It lays out the three extractions in full, the playbook for the third one, and the specific moves experts are making right now to stop being training data and start being category owners. If the thesis resonates, join the launch list.
Mollick asked whether to stay human. The better question is on whose terms you stay human, and what you're going to do this quarter to make the answer yours. I work with experts on exactly that at mattcretzman.com.
Keep Building,
— Matt