Your most valuable conversation this week happened in a hallway. Forty seconds. A junior asked a question, you answered, and that answer compressed fifteen years of pattern recognition into a paragraph nobody wrote down.
Now imagine that paragraph was captured. Not by you. By the pendant on their collar, the earbuds in their ears, the glasses on their face. Synced to a cloud you don't own, under terms you didn't sign, feeding a model that will sell your pattern back to the market at $0.003 per token.
That is the 2027 baseline. The question is whether you'll have built your own container before it arrives, or whether you'll spend the next decade watching your expertise show up in someone else's product.
I'm Matt Cretzman. I build AI systems for experts who refuse to be raw material. Here's why the hardware supercycle changes the math.
The capture layer is moving closer to the human
For the last three years, the extraction debate has been about software. Did OpenAI scrape your blog. Did Anthropic train on your podcast. Did Meta torrent your book. The fight was about content that already existed in digital form, sitting on servers, waiting to be ingested.
That fight is almost over. The experts lost it. The UK government punted on copyright protection even though 88% of respondents wanted it. The training sets are baked. The models shipped.
The new fight is about content that doesn't exist yet. The conversation you're about to have. The diagnosis you're about to deliver. The deal structure you're about to sketch on a napkin. The hardware supercycle Kalinowski is calling — wearables, ambient capture devices, always-on context engines — is a bet that the next training corpus is the one being spoken into existence right now in conference rooms, clinics, job sites, and Zoom calls.
The capture layer used to live behind a login screen. Soon it lives on a lapel. The friction to extract your expertise drops to zero.
What this costs you, in actual dollars
Let me put a number on this. A senior consultant bills somewhere between $400 and $1,200 an hour. The IP density of that hour — the frameworks, the judgment calls, the pattern matches — is what justifies the rate. Strip the IP density out, commoditize it inside a model, and the rate compresses to whatever a junior with a good prompt can deliver. Call it $80 an hour on the high end.
That's not a future problem. Mercor is already paying experts $50 to $200 an hour to dump structured knowledge into training pipelines. The CEO has said the quiet part out loud on multiple podcasts: they pay hourly because hourly is cheap relative to the lifetime value of the IP. Experts are selling a birthright for a bowl of stew because nobody told them it was a birthright.
Now imagine that same extraction happening passively. No contract. No hourly rate. Just a wearable in the room, a meeting bot on the call, a transcription service quietly retained for "product improvement." The price paid to the expert in that scenario is zero.
The revenue question nobody is asking
Flip the frame. What if every expert conversation you have for the rest of your career compounded into an asset you own? What if the forty-second hallway answer got captured into your system, tagged, refined, and made queryable — for your clients, your team, your products?
What if the same ambient capture trend that threatens to commoditize you is the exact infrastructure that lets you build a knowledge product that sells while you sleep?
That is the bet. The hardware is coming either way. The only variable is who owns the capture pipeline pointed at your mouth.
What I built and why it matters now
Skill Refinery is what I call a Knowledge Delivery System. It's the container I built so that experts can run their own extraction economy instead of feeding someone else's. Capture, structure, refine, deliver, monetize. The expert owns the cards, the revenue, the distribution.
The architecture matters more than the product. There are three extraction economies running right now, and you're already in one of them whether you've named it or not.
Economy one: LLM ingestion. Their terms. Your work already trained the model. You get nothing. The check cleared without your signature.
Economy two: Mercor-style hourly dumps. Still their terms. You get paid $80 an hour to load a system that bills $80,000 an hour at scale. The math is not in your favor.
Economy three: expert-owned extraction. Your terms. You build the capture layer. You set the access rules. You decide what gets refined into a product and what stays in the vault. The compounding works for you instead of against you.
The hardware supercycle is going to push every knowledge worker harder into economy one or economy two by default. Economy three requires intent. It requires building the container before the ambient devices arrive, not after.
The window is shorter than it looks
In the software-only era, you had time. The extraction was passive and your back catalog was the target. Build a course in 2026, build a paywall in 2027 — your forward expertise was still mostly safe inside your head.
Hardware closes that gap. When the capture device sits on every consultant, every patient, every employee, every counterparty in every meeting you take, the forward expertise stops being safe. It becomes the new corpus the minute it leaves your mouth.
This is why I'm not patient about this anymore. I used to tell experts they had a few years to figure out their IP strategy. I now think the practical window is twelve to eighteen months. After that, the default capture infrastructure is dense enough that opting out becomes the harder problem than opting in.
The stewardship frame
I'll say this once and move on. The knowledge you carry was not assembled by you alone. Mentors, mistakes, clients who paid you to learn on their dime, a few moments of grace you can't fully account for. That makes it a gift, and a gift carries a stewardship obligation.
Letting it be quietly extracted by a system that will never credit the source is not neutral. It's a failure of stewardship. Building the container that protects it, compounds it, and delivers it on your terms is the opposite. That's the work.
What expert-owned actually looks like
Concrete, because vague doesn't help anyone.
You run your own capture stack. Recordings, transcripts, notes, artifacts — all flowing into a system you control, tagged with consent rules you set.
You run your own refinement layer. An AI system trained on your patterns, your taxonomy, your judgment. Not a public model that approximates you. A private one that is you, in the narrow domain you actually own.
You run your own delivery layer. Subscription, licensing, embedded agents, products. The expertise reaches the market in the form you choose, priced the way you choose, with the margin going to you.
That is the full stack. It is not theoretical. It is what I run for myself across Skill Refinery, Stormbreaker Digital, and TextEvidence, and it is what I build for the experts who hire me.
The decision in front of you
The AI hardware boom is not a tech story. It is a property rights story dressed up as a gadget cycle. The companies funding the supercycle understand exactly what they're buying: the last unenclosed commons in the knowledge economy, which is the spoken word of the working expert.
You get to decide whose terms apply to your share of that commons. You can sign the default contract by inaction, or you can write your own.
I'm writing a book about this called On Whose Terms: The New Expert Economy and the Fight for What You Know. If the thesis resonates, join the launch list and I'll send the chapter on the three extractions when it ships.
If you want to see how I'm building expert-owned systems and where to start with your own, everything lives at mattcretzman.com.
The hardware is coming. Build the container first.
Keep Building,
— Matt