Seventy million dollars. A seven billion dollar valuation. And almost nobody outside hospital finance has heard the company's name.
It is called Commure. Last month it raised at a $7 billion valuation, led by General Catalyst, with Sequoia Capital and Morgan Stanley alongside. Its AI agents now run inside more than 130 of America's largest health systems, including HCA Healthcare and Tenet.
Here is the strange part.
While the rest of the industry races to build the smartest possible model, Commure went the other way. It built agents that live inside one industry's mess. Insurance codes. Claim denials. Payer rules. Billing.
Boring work. Regulated work. The work nobody posts about.
And it turned that into a $7 billion business. That gap, between what gets the headlines and what actually wins, is the whole story. Most people are watching the wrong race.
The race everyone is watching is the wrong one
Open any feed right now and the AI story sounds the same. Who has the smartest model. Who beat whose benchmark. OpenAI versus Anthropic versus Google, trading the lead every few weeks.
It is a genuinely exciting race. It is also not where most of the money is going to be made.
Because a smarter model is becoming a commodity. a16z made the point bluntly this year: as foundation models get better and cheaper, the scarcity stops being the model and starts being the data and the workflow around it. When anyone can rent intelligence by the token, intelligence alone stops being a moat.
So the question flips. If the model is no longer the hard part, what is?
The moat was never the model. It was the mess around it.
This is where Commure gets interesting.
US healthcare burns roughly $1 trillion a year on administrative work. Not treatment. Paperwork. Eligibility checks, coding, claims, denials, appeals. The stuff that sits between a doctor seeing a patient and the hospital actually getting paid.
I trained as a doctor before I moved into this work. The part they never warn you about is how much of medicine is paperwork, not patients. Every denied claim is a real person somewhere, re-typing the same form for the third time.
For thirty years, software promised to fix this and mostly failed. As Commure's CEO put it, software could store the information but it could never do the work. It could not make the calls or fight the denials. AI can. Commure's agents now handle more than 85% of revenue cycle work with no human touching it, and the platform processes over $25 billion in claims a year.
I call this the Boring Moat.
The Boring Moat is simple. The most defensible AI businesses are not built on the cleverest technology. They are built on the ugliest, most regulated, most tedious work in an industry. The work so specific that no general model can wander in and replicate it.
Boring. Regulated. Unglamorous. And almost impossible to copy.
Commure is not a fluke. It is a pattern.
Look around healthcare AI and the same shape keeps appearing. a16z points to Abridge, sitting on millions of real clinical conversations, and OpenEvidence, built on a vast medical library. The model is not their advantage. The hoard of proprietary data only they could have gathered is.
Each one picked a corner of medicine, went deep, and turned data nobody else holds into a position nobody else can take. Different companies. Same moat.
Why a general model cannot just walk in and take this
You might think OpenAI or Anthropic could build this tomorrow. They cannot, and the reason is not talent. Three things protect the Boring Moat, and none of them is the model.
First, the data. Years of real claims, denials and payer behaviour that only exist because you have been doing the work. A general model has never seen any of it.
Second, the workflow. The agent has to plug into the exact systems a hospital already runs, in the exact order the work happens. That integration is earned slowly, one deployment at a time.
Third, the regulatory maze. Healthcare rules are specific, local and punishing to get wrong. That complexity scares off generalists. For the company willing to live in it, the same complexity becomes the wall.
One analysis of a16z's latest startup cohort put it plainly: compliance barriers block general AI labs from going vertical, which turns regulatory complexity from an obstacle into the moat itself.
What this actually means for you depends on where you sit
This is the part nobody is spelling out. So here it is, by who you are.
For founders: stop pitching AI for everything. The companies winning right now are absurdly narrow. They pick one industry's worst job and go impossibly deep. Your edge is not a better model. It is proprietary data, a workflow you own, and a regulatory headache you are willing to suffer. Pick the boring thing on purpose.
For healthcare leaders: the vendor question is changing. You are no longer buying software that helps your team work faster. You are buying agents that do the work instead. That is a different purchase, with different risk and different savings. Commure's pitch is not log your claims better. It is we run your back office. Decide how you feel about that before a competitor decides for you.
For anyone selling AI: the pricing model is shifting under your feet. Seat-based pricing dropped from 21% of companies to 15% in a single year as buyers moved to paying for outcomes instead of logins. When AI does the task, you stop charging per person and start charging per result. If your offer still reads X dollars per user per month, you are pricing for a world that is ending.
The seat is not getting a co-pilot. The seat is going away.
Here is the line that should keep every software founder up at night.
For two decades the SaaS deal was simple. Build software that helps a human do their job, then charge for every human who logs in. a16z now argues that era is closing. The new model is not AI sitting beside the worker. It is AI replacing the task the worker was hired for.
In healthcare, that future already has a price tag. Seven billion dollars. And it was built on the most boring work in the building.
The lesson is not go work in healthcare. It is this. In the AI economy, the deepest, dullest, most regulated corner of your industry is probably the most valuable real estate you own.
Most people will keep watching the model race.
The smart ones are quietly digging moats.
If the model is no longer the moat, which boring, regulated corner of your industry is the one worth owning? Drop it below. I read every comment.