Don’t Sell the Tool. Buy the Work.
Why AgentAI raised $72M to buy medical billing companies and train agents on how experts work.
By Zorik Gordon, CEO · 7 min read
For forty years, software had one job: help people do their work. Now it has a new one: be the place where agents learn how the work is done—then do the routine part alongside the experts.
AI models are becoming a commodity. Expert judgment isn’t. Out of the box, AI doesn’t come with the instincts of a biller who has spent twenty years getting practices paid: which payer auto-denies which claims, which clinical note needs provider clarification, how to unstick a stalled claim. That judgment lives inside service businesses—exactly the real-world decision data AI labs are spending billions to recreate.
The opportunity isn’t another tool for people. It’s agents trained on real judgment. So we buy billing companies, put their billers in our platform, and let their daily work show us every gap we must close before agents can run the routine work on their own.
History records. Watching teaches.
Connect an AI agent to a billing database and it sees that Claim 1042 was paid and Claim 1043 was denied. It doesn’t see the biller open a clinical note, check an unwritten payer rule, spot a missing modifier, and fix one field before hitting submit. The database records the outcome. To learn the work, an agent has to see the decisions that produced it.
Interviews and documentation can’t capture that sequence. After twenty years, a biller’s deepest expertise is muscle memory. Across our acquisitions, we have found countless critical steps only by watching billers work—instinctive checks, subtle code adjustments, payer-specific habits they would never think to mention in an interview or write in an SOP.
An outside software vendor can’t capture those habits: it sees a narrow slice of the workflow and depends on someone else’s systems. Owning the service business gives us the whole view—the customers, the data, the veteran billers—and the authority to bring the entire workflow inside our own software.
Put the workers in the platform. Close the gaps. Train the agents.
Three years ago, we started AgentAI with a simple premise: rather than sell AI tools to billing companies, acquire them and rebuild their operations from the inside. We’ve raised $72 million from Transformation Capital, Eir Partners, Slow Ventures, Oceans Ventures, and others, and acquired five medical billing businesses.
Buying the businesses gets us in the building. But what happens inside our software is what turns their expertise into agents that work:
- Billers work inside our platform. Experienced billers work live claims in our software. Instead of relying on interviews or manuals, the platform learns from what they actually do—their actions, the clinical and payer context they check, and the judgment calls they make from experience.
- Their work exposes the gaps. Our test: can a biller do the entire job without leaving the platform? Every exit—to a payer portal, a spreadsheet, a phone call—is a gap, and any work done outside the system is a blind spot the agent can’t learn from.
- We close every gap with code. Our engineers sit beside our billers, so when a biller leaves the platform, we see it immediately and build the missing tool, integration, or portal bridge into the environment. More and more, our own coding agents close the gaps automatically.
- Billers teach the agents. Once a workflow runs entirely inside the platform, agents handle the routine cases on their own. When an agent hits a case it can’t solve, it hands it to a human-in-the-loop biller—and learns from how the biller resolves it. Today’s exception becomes tomorrow’s routine.
Capturing clicks isn’t enough. Closing the gaps and running a live human-in-the-loop captures the full chain of judgment—including instincts veteran billers don’t know they have—and checks it against whether the claim got paid.
Agents perform like experienced billers on Day One
Today our agents run around the clock, turning messy clinical paperwork into clean claims and managing submissions. Where we have closed the gaps, we see 75%+ lower agent cost per claim and more than 80% of claims handled autonomously in our most developed specialties. Quality and speed have improved as well: agent-prepared bills were 95.3% first-pass clean versus 92.9% for bills hand-keyed by experienced billers, and the median rejection is corrected and resent in 23 minutes. And the learning curve is accelerating: at one of our billing companies, weekly volume through our agents and system grew from 1,469 bills in July to 6,008 in October. And we are deploying agents not just in our own billing companies, but for PE-backed groups and MSOs too.
Because our agents learn the reasoning behind the work rather than brittle scripts, they handle new work the way seasoned billers do. Hand an experienced orthopedic biller a new orthopedic practice on a billing system they already use, and they don’t need months of retraining. They know the software, the codes, the modifiers, and how payers adjudicate the specialty; they adapt to the practice’s nuances and start immediately. Our agents work the same way.
We saw it with Atlas, our charge-capture agent. When four new practices came onto the system, we didn’t configure custom rules or train a new model. Atlas matched their specialty and workflows to practices it already knew and began processing their superbills. Agents handled 59% of the first 1,952 bills at those practices—with no new rules, training round or engineering work from our team.
That changes the economics of onboarding and M&A. A traditional billing company serves more doctors by hiring more people for repetitive work. In our model, every acquisition makes the system smarter: its billers help us expose and close new specialty gaps, while our trained agents take on the routine volume in specialties they already know. For MSOs, PE-backed groups and large practices that keep their own billing teams, we build custom agents for each practice or portfolio company—trained on its specialty, rules and payers, and taught by its billers inside our platform.
Better economics. Better jobs.
What happens to the human team? The biller’s job gets better, not smaller. A router sends routine claims—known specialty, gaps closed—to agents, and sends novel exceptions and hard calls to experienced human-in-the-loop billers. Instead of data entry, portal logins, and copying fields between screens, billers focus where expertise matters: quality, exceptions, and complex payer problems. Every hard case they close trains the agent.
Each biller can now oversee far more claims and practices. Because we own service delivery, that leverage flows directly to our P&L. And because billing fees are tied to collections, preventing rejections and recovering underpayments grows revenue for both the practice and AgentAI.
The hard part is the moat
Buying service businesses doesn’t automatically create an AI company. Every acquisition brings integration costs, legacy habits, and strict healthcare compliance requirements. Disrupting collections during a migration can cost you the customer, and an agent completing a task isn’t proof it made a compliant clinical or financial decision.
That’s why we built one unified platform instead of bolting third-party AI tools onto separate subsidiaries. We started with one billing company. Now we’re deploying the same approach across all of them: put the billers in the platform, close the gaps, and let every hard case they resolve teach the agents. Judgment calls still go to a biller, not an agent.
AI models will keep getting cheaper for everyone. The model isn’t the moat—owning the platform where human expertise is learned, validated, and scaled is. Every gap we close, every exception we resolve, and every billing company we acquire compounds our agents’ performance. That’s the business we’re building at AgentAI.
If you own a billing or RCM company and are thinking about a sale, we’d love to talk—or if you run a large MSO, PE-backed group or practice, we’d love to deploy custom agents into your billing operation—backed by our specialty billers and accountable for the result. Reach out at hello@agentai.app.