What did the Mercor round prove?

In its own announcement, Mercor says the round funds more investment in evaluation and post-training infrastructure and research, and more environments where AI agents can learn complex workflows. SiliconANGLE describes the focus as testing AI agents against a company’s own professional standards.

The money is flowing to human expertise.

Investors backed a clear idea: AI is only as trustworthy as the people who can tell a right answer from a plausible one.

In our view, healthcare is where this model matters most. National health spending in the US was projected at $5.7 trillion in 2025, up 7.3% from 2024, and CMS expects it to reach nearly $9 trillion by 2034 (AHA summary of the CMS analysis).

The largest opportunity for expert judgment in AI lies in healthcare, where the market is $5.7 trillion and the stakes are highest.

This isn’t just an economic argument. Healthcare is also the sector where patient safety is paramount. Beyond safety, scaling depends on clinician adoption, workflow fit, evidence generation, regulatory clearance and reimbursement, which a generalist cannot navigate. The expert layer that wins here will be the one built for it.

Why does healthcare need its own expert layer, not a generalist expert marketplace?

Because in healthcare, being wrong is not a quality problem. It is a patient safety and trust problem. In our view, a generalist expert marketplace falls short in six ways.

1. Patient safety. A fluent answer can still be an unsafe one. The AMA and the Digital Medicine Society report that models scoring 84 to 90% on medical knowledge benchmarks fall to 45 to 69% on clinical decision-making tasks and 40 to 50% on patient safety evaluations (AMA and DiMe, May 2026). Clinicians catch the missed warning sign, the wrong dose and the question that should go to a human. They read each output in the clinical context of the whole patient, not just the data in front of the model, and they define what safe means before a model reaches a patient.
2. Trust. A model that clinicians do not trust does not get adopted, and trust depends on being able to verify the output and understand the model’s limitations. A 2025 study cited by the World Economic Forum found clinician override rates of 1.7% for transparent AI and over 73% for opaque AI (WEF, December 2025). Verified, practicing clinicians involved in model development are the kind of proof that health systems and payors seek before they buy.
3. Depth. Clinical correctness lives at the specialty and subspecialty level. A cardiologist and a radiologist judge different things, and a generalist reviewer cannot tell a plausible answer from a clinically safe one.
4. Setting and clinical workflow. A tool built for an emergency department, an intensive care unit or a community clinic is judged specifically by the people who work in that setting, and each setting works completely differently. Workflow, meaning who does what, in what order and under what time pressure, decides whether a tool gets adopted or ignored. The AMA and DiMe report that roughly 60% of clinical AI pilots do not reach sustained deployment, and that the cause is rarely the model alone. Workflow changes, manual data entry and clinician trust are the usual barriers (AMA and DiMe). You cannot learn that workflow from outside the setting, so the people who work in it have to be part of the build.
5. Compliance and accountability. Paying clinicians is regulated. The HHS Office of Inspector General has said that fair market value alone is not a defense under the federal Anti-Kickback Statute, so an engagement also needs a real business need and must not reward referrals (Spencer Fane summary). Conflicts of interest have to be disclosed and managed, and patient-facing AI adds governance, liability and regulatory questions. Generalist expert marketplaces have not been built with this level of compliance in mind.
6. The path to adoption. Accurate AI can still fail if it does not fit clinical workflow or has no route to payment. A generalist can label a contract. A cardiologist can tell you whether an ECG read is clinically correct. Neither can tell you how a payor will decide. The AMA and DiMe also found no clear or consistent payment pathways for AI-supported services.

The Mercor round shows what the market will pay for expert judgment in general. Healthcare is the sector where that judgment has the highest stakes and the most specialized requirements, so it is the sector that most needs a purpose-built layer: clinicians in the loop for clinical truth, and health experts in the loop for the technical, commercial, regulatory and operational truth.

What is a clinician in the loop?

A clinician in the loop is a practicing specialty or subspecialty clinician who works inside the AI build, not after it. With Health Expert In The Loop, board-certified clinicians help with:

1. Training data definition. Clinicians decide which data is relevant, set the clinical context around it, and flag inputs that are incomplete, mislabeled or unrepresentative of the patients the tool will serve, before those gaps become the model’s habits.
2. Annotation. Clinicians label high-risk data to the standard real care demands, from findings that change a diagnosis to notes where the wording carries clinical weight. The quality of each label sets the ceiling on the quality of the model, so consistency matters as much as speed.
3. Edge cases. Clinicians find the cases a model is most likely to get wrong: rare or atypical presentations, patients with several conditions at once, conflicting or missing data, and groups the training data under-represents. They also mark where the model should stop and hand the question to a human.
4. Model evaluation. Clinicians review outputs for correctness, safety and appropriateness, and judge whether the answer fits the patient in front of it. They flag confident answers that are wrong, check that the model shows uncertainty when the evidence is thin, and write down why, so the team can retrain against it.
5. Validation and monitoring. Clinicians stress-test the model’s reasoning before launch and stay on after it, joining ongoing quality assurance as guidelines, patient populations and workflows change. Monitoring catches drift that a one-time validation cannot, so the model keeps pace with the standard of care.

The loop has to be real. The AMA and DiMe warn that having a human in the loop does not guarantee meaningful oversight. Clinicians need defined checkpoints and the authority to override the model.

What is a health expert in the loop?

A health expert in the loop is an expert who is not a clinician but has deep understanding of other parts of healthcare. Clinicians make sure the answer is clinically correct. Health experts make sure the product can be trained, paid for and taken through regulation. In our view, this is where most generalist expert marketplaces fall short:

• RCM specialistsKnow how claims, coding, denials and prior authorization actually work, and therefore how a product gets paid. Without them, a product can be clinically correct and still go unpaid.
• Bioinformaticians and data scientistsKnow how to train, validate and stress-test algorithms, for example for lab tests and diagnostics. Without them, a model can look accurate on its development data and fall apart on new patients.
• Regulatory strategists and FDA advisorsKnow what evidence a product needs and how to get it through the FDA. Without them, teams find the gap at submission, not at design.
• Reimbursement and market access leadersKnow how coding, coverage and reimbursement decisions work. Without them, a product can have users and no payer.

Each of these roles already shows up in how healthcare AI is built and governed:

1. Revenue cycle. An AGS Health piece in HIT Consultant describes coder corrections feeding back to improve AI coding models, with ambiguous or high-risk cases routed to humans.
2. Regulatory. The FDA’s January 2025 draft guidance on AI-enabled device software asks for data provenance, performance across subgroups, a description of the human-in-the-loop process and post-market monitoring.
3. Reimbursement. Akin Gump notes that Medicare has no standard method for covering FDA-authorized AI devices, so developers need clinical-utility evidence and a coding and payment strategy.
4. Diagnostics. The Association for Molecular Pathology describes clinical bioinformaticians as the specialists who develop, validate and implement the pipelines behind molecular tests in regulated labs.

Their judgment shapes the decisions that accuracy testing never touches: how the model is built, who pays, under which code, through which contract, with what evidence. Many healthcare AI products fail at deployment because workflow and reimbursement questions were left for later.

What does a healthcare expertise layer need to look like?

We think it needs five things:

1. Verified, practicing experts. Both clinical and non-clinical. Our network of more than 4,000 vetted experts spans practicing clinicians, health system and payor executives, lab directors, bioinformaticians, regulatory strategists, and coding and reimbursement leaders, each with deep operating experience across the healthcare continuum.
2. Specialty and subspecialty coverage. At MDisrupt, that means 35 clinical specialties and more than 65 subspecialties, with AI-powered matching on clinical depth and availability, so a question about a specific condition or device reaches someone who practices in that area.
3. Speed. Our experts are contracted and ready, so they can be available for healthcare AI projects within days, from a labeling sprint to ongoing clinical oversight.
4. Deep compliance. We handle contracting, compensation that is fair market value and commercially reasonable, and conflict of interest management, and we cover every expert with liability insurance, so the AI team can focus on the build and every engagement stays defensible.
5. Security and trust signals. Health data and model work call for controls a buyer can verify. MDisrupt is SOC 2 compliant.

That is what we built MDisrupt to be: the health expertise infrastructure for healthcare AI.

How can healthcare AI teams get started?

The best time to bring clinicians and health experts into your health product development is before the model is locked, when their input can still change the training data, the evaluation and the workflow. Put the right specialist on the point where a wrong answer would cost the most, such as labeling, edge cases or validation. With Health Expert In The Loop, that can be one clinician on a labeling sprint or a panel for ongoing quality assurance.

Sources

• Mercor, Organizing human intelligence for the entire economy. Primary source: $750M Series D at a $20B valuation, announced Nov 10, 2026, and use of funds.
• Reuters via Investing.com. Round size and valuation, and the prior valuation of about $10B in Oct 2025.
• SiliconANGLE. Product focus on testing AI agents against a company’s professional standards.
• Health Expert In The Loop. MDisrupt product page: 65+ specialties and subspecialties, services and FAQ.
• Trust starts when clinicians help build your AI, not when they start using it. MDisrupt webinar, Nov 5, 2026.
• From Benchmark to Bedside: The Current State of AI in Clinical Practice. AMA Center for Digital Health and AI with the Digital Medicine Society, May 2026.
• CMS projects national health spending grew to $5.7 trillion in 2025. AHA News, June 25, 2026, reporting the CMS analysis in Health Affairs.
• What is clinical informatics. AMIA, the professional body for the field: the clinical informaticist’s role and responsibilities.
• Clinical informatics board certification. AMIA: the physician subspecialty credential administered by ABPM and ABPath.
• Who will lead hospitals into the age of AI? Becker’s Hospital Review, September 30, 2025.
• Transforming medical coding operations with hybrid intelligence design. HIT Consultant, May 11, 2026 (AGS Health): human corrections in AI-assisted coding.
• FDA issues AI-enabled device software functions guidance. DLA Piper, January 2025: summary of the FDA draft guidance of Jan 7, 2025.
• AI in clinical decision-making: regulatory roadmap and reimbursement strategies. Akin Gump, December 11, 2025.
• Preparing for the bioinformatics age. The Pathologist, November 4, 2025: interview with the AMP Clinical Bioinformatician Body of Knowledge committee chair.
• The trust gap: why AI in healthcare must feel safe, not just be built safe. World Economic Forum, December 5, 2025.
• OIG reiterates the need for full safe harbor compliance. Spencer Fane, law firm summary of OIG guidance on fair market value and the Anti-Kickback Statute.