Leah Houston, MD

The through-line

Everything Leah Houston argues starts from one conviction: the physician, not the institution, should hold and control their professional identity. She earned her credentials once, and she thinks the system should treat them that way. "I went to medical school once. I went to residency once. It's a permanent credential. It's permanently verifiable, but we have to keep going back and verifying it, verifying it, verifying it because of the way uh the fact that there's no real standards around what verification means, what primary source verification means" [1]. That position has been consistent for years and has travelled across technologies. The earlier expression of it was HPEC, a physician-owned digital identity and credential wallet built on blockchain and self-sovereign identity principles, framed as a way for doctors to organise digitally while staying independent of centralised hierarchy and control [8][9][10][11][12], and described as a decentralized autonomous organization of practicing physicians [3]. The current expression is Evercred, an agentic AI credentialing platform [1][2]. The underlying claim has not moved: control of credential data is control of professional autonomy [4][6][7].

What has shifted is the mechanism and the buyer. The autonomy argument was once made largely to physicians about reclaiming power [7][9]; it is now also made to hospitals and CMOs as an operations and cost argument, with the physician-held wallet sitting inside a full credentialing suite that medical staff offices and CVOs use [1][2].

On credential misuse and what it revealed

The origin is a specific injury. After Houston left a hospital, it kept her on the roster as a supervising physician, and continued billing under her name after she had left the state and after her license there lapsed, "committing essentially fraud against my medical professional credentials" [1][2]. The federal Centers for Medicare and Medicaid were notified, and it "made it appear as if I was practicing without a a license when in fact it was really just a paperwork snafu" [1]. She briefly lost her right to bill Medicare and Medicaid, reversed once the situation was understood [1][2]. The detail she keeps returning to is that someone else had been signing her charts all along without her knowledge: "I thought I was the only one that was allowed to sign off on those charts. But apparently they had been doing that on my behalf the whole time and I didn't know that" [1].

Her point is that this is not a freak case. Talking to colleagues, "I kept hearing stories, similar stories," including physicians at telemedicine companies with durable medical equipment ordered under their names and "thousands of patients getting continuous monitoring under their name and they never knew about it, nor did they even know who these patients were" [1][2]. She treats this as structural rather than anecdotal, "a pervasive problem" and "one of the things that's leading to, you know, our health care system waste" [1]. The enabling condition is the blanket privileging agreement, where a physician signs over the right to use their credentials for billing and "really that's it. It's just kind of like a blanket. Here you go" [1]. Losing control of professional identity this way is what drove her out of clinical practice and into building a solution [3][5][6], and it is what she calls "the last straw" [1][2].

On why credentialing is still broken

Houston's account of the current state is that digitisation has changed the interface without changing the process. Some systems have a portal, but "most systems, even large hospitals are they're still mailing them packets where they have to manually fill things out," and where things do move by email it is "still like a download and then re-upload situation" [1][2]. Every credentialing specialist she speaks to, including those at the largest US systems on the most advanced available software, describes the same bottleneck: getting the physician to respond and update their information [1]. Deadlines compound it, since missing a five or ten day window can force the whole process to restart, which she counts as pure administrative waste [1][2]. She also agrees that much of the delay comes from chasing apparent red flags that turn out to be nothing more than a mistake on a piece of paper [1][2]. In her own experience, spanning roughly a decade of emergency medicine across multiple states and systems, this inefficiency produces two harms at once: barriers to care and barriers to professional mobility [1][2][6].

On physician-held credentials and portability

The design answer is a wallet the physician keeps. Doctors store credentials digitally, share them directly through the system, and receive update notifications in the application, with text message alerts added specifically to attack the response-time problem [1][2]. When a physician leaves, "they still have those credentials bundled in their wallet that they can take with them," which is what makes onboarding to the next place faster [1]. She accepts the framing of it as a portable, physician-managed, authenticated credential set, comparable to carrying an ID in an Apple wallet, while stressing that there is also a hospital and practice facing portal, because "it's a full credentiing suite" [1][2]. This is the same principle behind the earlier HPEC wallet, which was built to make the physician the primary source of ownership and control of their professional credentials and data [10][12].

On revocation she is precise rather than absolutist. An organisation that credentialed the doctor keeps the data it generated: "they have that data that's theirs. Nothing's going to change there" [1]. What the physician can sever on leaving is the live link. "Essentially you have everything that you've ever done as an organization, but the connection between the doctor gets lost" [1][2]. She also notes the obvious practical limit, that a physician currently working somewhere is not going to revoke their credentials from that employer [1].

On what AI agents actually replace

Houston is specific about which work she is automating: primary source verification, the requirement to go to external sources and prove a credential is real [1][2]. Today that means visiting a website, typing in a name and number, screenshotting, and uploading, or in many cases sending and receiving faxes to confirm education and employment history [1]. Her description of agentic AI is deliberately literal: "you can spin up an agent that can actually go to the website, click the buttons, fill out the form, take the screenshot, upload it for you, so you don't have to" [1][2]. OCR handles data extraction so nothing has to be typed in manually [1]. She frames the timing as capability catching up with an old idea: "I've always kind of known that at some point AI would reach maturity to do this and now it has" [1].

She does not claim autonomy for the agents. "We always will need a human in the loop at least now. Maybe 10 years from now we won't need a human in the loop" [1][2]. On the standard objection about hallucination, her answer is comparative rather than defensive: in workflows that are highly structured and where you know exactly what is supposed to happen, "there's a lot less errors already and it's going to continue to get better and better" [1]. She counts error reduction as a cost saving in its own right, alongside time and headcount [1][2].

On the economics of credentialing

She makes the business case in hard dollars, using projections for a hospital she has an LOI with that has 700 practitioners. That hospital spends over $250,000 on credentialing software alone plus roughly $2.5 million on staff whose entire job is managing credentials [1][2]. Her model brings software to around $140,000, reduces credentialing staff to roughly a one-to-ten ratio of current headcount, and puts agent costs at around $600,000, for a net saving near $2 million [1][2]. The agents scale in a way people cannot: "You can deploy them in parallel. They can be deployed while you sleep. You can have dozens or hundreds or thousands deployed at the same time" [1].

The larger number is time to work. If physicians could start "in a couple of weeks instead of 4 to 6 months," she calculates 63,000 hours recovered, totalling roughly $560 million, and while these cycles recur only about every ten years, "half a billion dollars is a lot of money even over 10 years" [1][2]. She is aware this is the part finance leaders discount as soft dollars, and pairs it with the direct cost reduction so the case works on both sides of the ledger [1][2].

On selling to small practices first

Houston is unusually blunt about where her product does and does not fit. "Our product is really designed for small to mediumsiz practices. It's not really designed for large health systems" [1]. CMOs have approached her over LinkedIn wanting to use it, and her answer is that she does not know if it is "ready for prime time at large health systems yet" [1][2]. The near-term market is the one still running on spreadsheets and Dropboxes, where both the practices and the doctors working in them respond enthusiastically [1]. She is nonetheless open to design partnerships with large systems, and has signed an agreement with a 700-bed hospital targeted to launch in the fall, by which point she expects the product to be mature enough [1][2].

On decentralized identity and finding the right doctor

The longer-term ambition goes past credentialing into identity infrastructure. Her starting complaint is that "right now physicians and their patients don't have access to their data," and that a physician-controlled credential layer could be used to add identity to records and improve continuity of care [1][2]. This is continuous with her earlier work on medical privacy and the argument that doctors need to manage their own data [11][12].

The application she describes is matching. She imagines hospital systems and private practices alike being able to find the right physician for a specific need, using verified experience rather than an insurance roster. Her example is a parent of a blind child who would benefit from a pediatrician who has treated a larger volume of blind children, because that clinician brings more experience and more resources [1][2]. Her verdict on the status quo is direct: "just spray and pray and pick any random pediatrician on the insurance roster list isn't really going to work for people in the future when we have more access to this kind of information" [1]. She makes the same point about rare disease, noting that only a handful of physicians in the entire US are highly specialised in neurofibromatosis [1][2].

On becoming a founder as a physician

She did not plan any of this. Recommended computer science courses in high school on the strength of her math scores, she refused: "I'm going to be a doctor. That has nothing to do with practicing medicine... I'm going to go save lives for a living. I'm not going to sit in front of a computer" [1][2]. Now she spends all her time behind a computer and took AI training at MIT, an outcome she says her earlier self would find bewildering [1][2]. The entrepreneurial instinct was always there, from flipping cars in high school to real estate investing [1][2]. What sustains her is not the founder identity but the problem: "I enjoy it because I'm solving a problem. It's a real problem. It's a real need. Everybody who sees the demos gets really excited. And it's way better than spreadsheets and Dropboxes" [1][2].

Takeaways

• A hospital kept Houston on its roster and billed under her name after she left the state and her license there lapsed, briefly costing her the right to bill Medicare and Medicaid; colleagues reported similar cases, including DME and remote monitoring ordered under their names without their knowledge [1][2][3]
• Credentials are earned once and permanently verifiable, but are re-verified endlessly because there are no real standards for what primary source verification means [1][2]
• Most credentialing is still paper packets or download-and-re-upload email, with tightening five and ten day response deadlines that can force a restart of the entire process [1][2]
• Physicians should hold credentials in a portable wallet they take between employers; the hiring organisation keeps the data it generated, but the live connection to the physician can be severed on departure [1][2][10][12]
• AI agents can perform primary source verification end to end, visiting sites, filling forms, capturing screenshots and uploading, with OCR extracting data, though a human stays in the loop for now [1][2]
• For a 700-practitioner hospital, her model cuts software from over $250,000 to about $140,000, reduces credentialing staff to a one-to-ten ratio, and saves roughly $2 million a year, separate from an estimated 63,000 hours and $560 million tied to faster time to work [1][2]
• Evercred is built for small to medium practices still on spreadsheets and Dropboxes, not yet for large health systems, despite inbound interest from CMOs [1][2]
• A physician-controlled identity layer could let patients find clinicians with real volume in their specific condition rather than picking at random from an insurance roster [1][2]