Your hourly rate is about to get compared to a model that costs less than a vending machine soda.
Google dropped Gemini inference pricing again last month. DeepSeek is running margins so thin the Chinese hyperscalers are openly subsidizing tokens. Anthropic is matching. The race to zero on the model layer is no longer a forecast. It's the Tuesday news cycle.
Here's the question every expert with twenty years of pattern recognition needs to answer this quarter: when a competitor can run your synthesis at $0.10 per million tokens, what is the buyer actually paying you for?
I'm Matt Cretzman, and I've been building the answer for the last two years. The short version: the price war makes expert-owned knowledge more valuable, not less. But only if you move before the commoditization wave hits your category.
The Margin Collapse Nobody Wants to Name
Follow the money. In 2023, GPT-4 cost roughly $30 per million input tokens. Today the equivalent capability runs under a dollar at most providers, and the open-weight class is approaching ten cents. That's a 99% price collapse in 24 months.
This is not a temporary discount. It's the structural endgame of any commodity. When three hyperscalers and a Chinese lab are all racing the same curve, no one holds pricing power. The model layer is going to look like long-distance phone calls in 2005.
The consulting world is already feeling the second-order effect. I've talked to three partners at top-five firms this year who admitted their associates are running first-draft analyses through Claude before any human touches them. The cost of generic synthesis is approaching zero. The margin on generic expertise is following it down.
If your service offer is "I'll think about your problem and write up what I conclude," you're competing with a $0.10 token bill plus a junior who knows how to prompt. That's the cost anchor. Be honest about it.
What Actually Gets More Valuable
Here's the part most experts miss because they're too busy being anxious about the model layer.
When inference commoditizes, the bottleneck shifts. The scarce resource is no longer compute or capability. It's context — the proprietary judgment, the edge cases, the "I've seen this fail seventeen times in the following specific way" pattern library that lives in your head and nowhere else on the internet.
A frontier model with generic context produces generic output. A mid-tier model with your context produces output the frontier model cannot match, because the frontier model has never sat across the table from your 400 clients.
This is the inversion. The price war is not coming for experts. It's coming for the middlemen who were renting access to generic intelligence and marking it up. The expert who owns refined context becomes the only non-substitutable input in the stack.
But — and this is where most experts lose the decade — context that lives in your head is not an asset. It's a liability with a retirement date.
The Three Extraction Economies
I wrote a chapter in my book about this. There are exactly three ways your expert knowledge ends up in the AI economy, and you are already in one of them whether you chose it or not.
Extraction One: The LLMs already took it. Your blog posts, your conference talks, your published case studies, your podcast transcripts — ingested without consent into training corpora. The UK Parliament debate in 2024 was a tell. 88% of surveyed experts wanted protection. The government punted. Your IP is in the weights. You were not asked. Those were their terms.
Extraction Two: Mercor and the hourly dump. I watched the Foody interview with Mercor's CEO last summer. The pitch is clean and the money is real: experts log in, answer questions, label edge cases, get paid hourly. What's actually happening is a one-time conversion of decades of judgment into training data that will be sold back to the market — minus you — forever. Jacob selling his birthright for a bowl of stew. Their terms.
Extraction Three: Expert-owned. You build the system that captures, refines, and monetizes your judgment under your name, your distribution, your revenue, your terms. The frontier-model price war is what makes this finally viable. Inference is cheap enough that a solo expert can run the same stack a $50M consultancy ran in 2022.
You're in one of these three. The first two are happening to you. The third is the one you choose.
What I Built and Why It Matters Now
Skill Refinery is the system I built for Extraction Three. It's a Knowledge Delivery System — a stack that takes an expert's accumulated judgment and turns it into a productized asset the expert owns end to end.
The architecture is boring on purpose. Capture mechanism for tacit knowledge. Refinement layer that structures it into reusable patterns. Delivery layer that meets buyers where they are — async, on-demand, agent-mediated. Distribution that doesn't depend on a platform's algorithm to feed you. Revenue that doesn't get clipped by a marketplace's 20%.
The reason it works now and didn't work in 2022 is exactly the price war we started with. When inference was expensive, building expert-owned systems required venture money. Today I can spin up the same infrastructure for an individual practitioner for less than they spend on coffee. The collapse in model pricing is the unlock for the people the price war is supposed to be threatening.
That's the inversion most coverage misses. The hyperscalers think they're winning by racing to zero. What they're actually doing is funding the largest decentralization of expert IP in economic history. They just don't know it yet.
The Numbers That Should Set Your Strategy
A few anchors to sit with this week:
Inference cost per million tokens has dropped from $30 to under $1 in 24 months, with open-weight models approaching $0.10.
A productized expert offer with proprietary context typically commands 3-10x the price of generic AI-assisted consulting, because the buyer is paying for the judgment delta the model cannot replicate.
The window to convert tacit expertise into owned IP is roughly 18-36 months, based on the rate at which generic synthesis is closing the gap on mid-tier knowledge work.
Experts who started building owned systems in 2023 are now charging premium retainers while their peers are watching their hourly rates compress. I've watched this happen in three categories already: legal research, financial analysis, technical writing. The pattern is identical each time.
The Stewardship Question
There's a line in my book I keep coming back to. Your expertise is not yours in the sense that you generated it from nothing. It was built on mentors, mistakes, clients who trusted you before you deserved it, and time you can't get back. That makes it a gift, and gifts come with stewardship attached.
Letting it get extracted on someone else's terms — silently, hourly, or by default — is not neutral. It's a choice about what compounds and what gets clipped. I think the people who built something worth extracting owe it to the people who taught them to make sure the asset outlives the career.
I'll leave it there. You know what I mean.
What To Do This Quarter
Three moves, in order:
First, audit which extraction economy you're currently in. If your content is being scraped and you're not building owned distribution, you're in #1. If you're trading hours for labels on someone's platform, you're in #2. Neither is bad. Both are someone else's terms.
Second, identify the three to five judgment patterns in your work that no generic model could replicate without your specific history. That's your context moat. Write them down this week. Not for a blog. For your own audit.
Third, decide whether you want to be a buyer or a builder in the expert economy that's coming. The price war makes both viable. Only one of them compounds.
I'm writing a book about exactly this — On Whose Terms: The New Expert Economy and the Fight for What You Know. If the thesis resonates, join the launch list here.
The rest of the system, the playbook, and what I'm building at Skill Refinery and Stormbreaker Digital lives at mattcretzman.com.
The price war is real. So is the inversion underneath it. Move while the cost curve is still your friend.
Keep Building,
— Matt