Beacon Health taught an AI to click through the chart - so the doctor doesn't have to

Walk into almost any independent primary care practice in America and you will find a doctor who did not go to medical school to become a data-entry clerk. Somewhere between the exam room and the billing office, a second job appeared: chasing quality measures, hunting for missing diagnosis codes, filing prior authorizations, calling patients who are overdue for a screening. The work is real, it is worth money, and it never ends. Beacon Health, a company in Y Combinator's Winter 2026 batch, is betting that most of it does not need a human at all.

The startup builds what it calls AI employees for primary care. Instead of another dashboard that asks overworked staff to log in and learn a new system, Beacon's agents log into the systems the practice already has. They open the electronic health record, navigate its screens, read the chart, and complete a task from start to finish - the way a new hire would on their first week, except they do not need lunch and they do not quit.

01The bet: teach AI to click, not to integrate

Most health-tech automation lives or dies by the API. If a system exposes a clean data connection, software can plug in; if it does not, the automation stops at the door. The problem is that a huge share of American healthcare runs on legacy EHRs where the real work happens in the interface - dropdowns, tabs, free-text notes, and buttons a person has to click in the right order. Beacon Health decided to meet the work where it actually is.

Its agents operate the EHR user interface directly. That single design choice is the reason the company can say it is EHR-agnostic: it does not wait for a vendor to open a pipe. According to the company, the platform works across Epic, athenahealth, eClinicalWorks, Cerner, MEDITECH, NextGen, Veradigm, Allscripts, and MEDENT.

02What the agents actually do

Beacon frames the product around the parts of primary care that generate revenue and eat time in equal measure. There are three jobs the company keeps coming back to: closing quality gaps, capturing risk adjustment, and covering after-hours patient needs. A practice can also record a workflow it already runs - a referral, a discharge follow-up, a screening reminder - and turn that recording into a reusable agent within minutes, then chain several agents together into a longer task that runs across an entire patient panel.

The oversight layer matters as much as the automation. Beacon's Mission Control dashboard lets a practice deploy agents across a population and keep a human hand on the wheel - watching status, pausing operations, and reviewing performance. In a setting where a wrong click can touch a patient's record, a visible pause button is not a nice-to-have.

It is worth being precise about what each of these jobs means, because the value hides in the detail. Closing a quality gap can be as simple as noticing a diabetic patient is overdue for an eye exam and firing off the outreach, or ordering the lab that a measure requires. Risk adjustment is the quieter one: the practice is paid based on how sick its patients are, and a condition that is real but never documented in the current year simply vanishes from the math. Surfacing those missed codes is not gaming the system - it is getting paid for care the practice is already responsible for. And after-hours triage, the third pillar, is the one that keeps patients out of the emergency room at two in the morning without burning out the person who would otherwise take the call.

03Who it is for, and why now

Beacon's customers are not hospitals with hundred-person revenue-cycle teams. They are the practices caught in the awkward middle of value-based care: independent physician associations, accountable care organizations, integrated networks, and direct-to-employer arrangements. These are groups that signed risk-based contracts because the math rewards keeping patients healthy - and then discovered the reward is gated behind a wall of documentation they do not have the staff to scale.

That is the gap Beacon is aiming at. A quality bonus a practice cannot capture is worth nothing. A diagnosis that never got coded lowers the practice's risk score and its payment. For a small group, an AI back office that chases those dollars is the difference between a value-based contract that works and one that quietly loses money.

The business model follows from that framing. Beacon is not selling a per-seat license to another piece of software that a practice will forget to open. It is selling outcomes tied to money the practice can actually see - quality bonuses captured, risk scores corrected, shared-savings dollars earned, and staff hours returned. The company's headline claim is that it can double a primary care practice's revenue while lowering the total cost of care. Both halves of that sentence point at the same lever: the administrative work that stands between a practice and the incentives already written into its contracts.

04The founders

CEO Mark Pothen did not discover this problem in a pitch deck. He grew up inside his mother's primary care practice, and before founding Beacon he spent six months embedded in a practice's operations, watching the workflows he now automates. His earlier work was in product and go-to-market at early-stage healthcare startups.

CTO Obinna Akahara brings the engineering side - a physics background and experience building production AI systems across healthcare and enterprise software, with prior stints at LinkedIn and Conversion.ai. The pairing is the familiar founder shape for this kind of company: one who has lived the customer's day, one who can ship the system that fixes it. The company is backed by Y Combinator, with group partner Diana Hu.

05Where it sits in the market

The health-tech field is crowded with automation: ambient AI scribes that draft the note, revenue-cycle tools that scrub claims, coding vendors that surface missed diagnoses. Beacon's distinction is the layer it operates on. Rather than integrating alongside the EHR, its agents operate the EHR itself, which lets it reach legacy systems that never opened an API and to run whole multi-step workflows instead of a single step.

There is risk in that approach - operating a live clinical interface is unforgiving, and reported funding figures for the young company vary across public trackers from roughly a hundred thousand into the mid six figures. But the early signal is hard to argue with. Going live across a 40,000-patient physician association before most startups finish their first hire is the kind of distribution that, in healthcare, tends to matter more than a polished demo.

The expertise on display is less about a single technical breakthrough and more about a point of view. Plenty of teams can build an AI that reads a chart. Fewer have sat in the back office long enough to know which tasks are worth automating first, which ones a practice will actually trust a machine to run, and where a human has to stay in the loop. Beacon's early customer list reads like a set of practices willing to test that judgment - names such as MWA, Kaaya Health, Boroughs, Town Center, and Nully. In a market where trust is the scarcest resource, letting a small vendor operate inside your live EHR is itself a strong vote.

Beacon's stated mission is smaller and more human than the technology behind it: bring joy back to primary care. Strip away the agents and the dashboards and the pitch is simple - give the doctor back the hours the software took, and let the practice keep the money it already earned.