Sponsored by: GenHealth.ai
Piedmont Medical Solutions grew paid-to-date collections 34.2% after switching to GenHealth. Their eligibility checks, prior auths, and claims follow-up now run end to end.
GenHealth's AI agents work the phone calls, the faxes, the EHR, and the payer portals. They run inside the systems the team already uses, at over 20 million actions a year across our customers.
Book a call. We will run your collections number.
Good morning —
For those in NYC, we’re hosting an end-of-summer hang with Lorikeet for healthcare founders and operators. No panels or pitches, just drinks, sushi, and good people. Join us here🗽
Now let’s get to it:
Third-party evaluators in healthcare
Health AI’s consolidation era is starting
AI is moving deeper into drug discovery
33 new tools/partnerships, 15 funding updates, new AI jobs & link-worthy content
Read time: 8 minutes
Our Picks ✨
Highlights if you’ve only got 2 minutes…
1/
Third-party evaluators in healthcare
The recent “pace the frontier” debate has put third-party evaluation in the spotlight. More than 100 AI experts just called for third-party evaluators with deep access and real independence from the labs they assess. Healthcare is an obvious place where this matters.
A widely shared FT piece this weekend highlighted the gap between statistical validation and actual clinical impact. One Michigan sepsis model generated so many alerts that clinicians eventually started tuning them out. There are plenty of encouraging examples too. AdventHealth reported an 80% reduction in time spent on certain admin tasks using ChatGPT for Healthcare, while an independent review at Penda Health found 16% fewer diagnostic errors with AI assistance.
The problem is that adoption is moving faster than the infrastructure for evaluating it. In a recent survey, nearly 3 in 4 healthcare organizations said AI tools or agents are deployed without formal IT approval at least sometimes. More than 25% already have agentic AI in production, another 44% are piloting it, and only 17% think their existing identity controls are sufficient.
Regulatory evidence is uneven too. The FDA authorized 258 AI medical devices in 2025, but only 2.4% were backed by the kind of trial evidence typically expected before a drug reaches patients.
Even the benchmarks need scrutiny. Epoch AI reviewed 50 HealthBench Professional tasks and found clear errors in half. Protege found that changing answer order or adding a single sentence to a prompt could change which model performed best. Some public benchmarks also give models fewer than 200 tokens of context, compared with roughly 8,500 tokens in the median real patient record in Protege’s data.
There may already be a useful blueprint scattered across healthcare and other safety-critical industries. Clinical labs use independent proficiency testing, where outside organizations send blinded samples to see whether a lab can produce the right result in normal operation. Medical devices are monitored after they reach the market because problems can emerge only after broader use. And in aviation, independent investigators examine serious failures and publish what went wrong so the broader system can learn from them.
Health AI probably needs some combination of all three.
Independent evaluators could test the benchmarks themselves, validate models inside real clinical workflows, monitor performance as models and patient populations change, and investigate serious failures when they happen.
A one-time leaderboard score probably isn’t enough. As AI becomes part of everyday care, the standard that matters is whether it improves safety, outcomes, access, cost, and clinician burden in the real world.
2/
Health AI’s consolidation era is starting
Health AI is starting to look a lot more like a mature software market. Scaled companies are using M&A to move into adjacent specialties, workflows, and customer bases, while some larger platforms are becoming acquisition targets themselves.
Sword Health → Headspace: Sword signed an agreement to acquire Headspace, expanding from MSK deeper into mental health. Headspace brings a consumer brand that has reached roughly 100M people, plus distribution through 20,000+ employers and a network of 15,000+ providers. (link)
Medallion → Andros: Medallion acquired Andros, bringing 1M+ providers across nearly 400 healthcare organizations and health plans onto its credentialing platform. The combination expands Medallion across credentialing, enrollment, licensing, and provider data management. (link)
Hello Patient → Converse Health: Hello Patient is moving from the front desk into the back office. Converse’s agents handle referrals, chart follow-up, authorizations, records, and other administrative work surrounding a visit. (link)
Tempus AI → Personalis: Tempus agreed to acquire cancer genomics company Personalis for $1.5B, adding molecular residual disease testing to its precision oncology platform and expanding further across diagnosis, treatment selection, recurrence, and monitoring. (link)
Waystar → ???: Waystar is reportedly exploring strategic options, including a possible sale that could take it private only two years after its IPO. The company serves 30,000 clients and 1M+ providers, processes more than 7.5B healthcare payment transactions annually, and itself acquired AI-driven RCM company Iodine Software for $1.25B last year. (link)
The pattern is getting harder to miss. Health AI companies that have established distribution are starting to buy their way into adjacent workflows and specialties rather than build every capability from scratch. Expect a lot more of this. (linkedin)
3/
AI is moving deeper into drug discovery
McKinsey estimates generative AI could create $60B to $110B in annual value across pharma and medical products, touching almost every part of the value chain.
In discovery, AI can screen millions of compounds and potentially cut lead identification from months to weeks. In clinical development, it can help design trials, speed enrollment, clean data, and draft regulatory submissions. Commercial and medical teams can use it to create compliant content, review literature, and answer clinician questions faster.
The frontier labs are now moving directly into the science. Anthropic is testing whether Claude can control lab robots and run physical biology experiments, while Novo Nordisk just partnered with Anthropic to bring Claude Science into drug discovery and R&D, using it for scientific reasoning, biological research, and agentic software engineering.
There is still an important reality check. More than $40B has reportedly flowed into AI biopharma this decade, yet few AI-derived drugs have made it beyond Phase 2.
That makes the next chapter especially interesting. AI is getting better at generating hypotheses and finding promising molecules. Now the challenge is connecting those models to experiments, clinical development, and eventually medicines that actually work in humans. Exciting times nonetheless!
Tools & Partnerships 🔧
Latest on business, consumer, and clinical healthcare AI tools and partnerships…
TOOLS
Stanford’s Virtual Biotech deploys 37,000 AI agents for drug discovery: The multi-agent system analyzed nearly 56,000 clinical trials and found drugs targeting cell-type-specific genes were 48% more likely to reach market with 32% fewer adverse events. (link)
Latent expands pharmacy AI to 60 health systems and 70M patients: The company also launched autonomous clinical agents designed to accelerate access to GLP-1s and specialty therapies, and expects to process more than 5.8M medication journeys over the next year. (link)
Luma Health uses AI to bring missed patients back: New agentic workflows follow up by voice or SMS to close care gaps and refill schedules, with 28% of identified care gaps booked and 45% of missed visits rescheduled. (link)
Dimer brings the hospital “call button” home with AI: AiME is a personalized care navigator patients can access after discharge for real-time recovery guidance, medication questions, scheduling, and escalation to clinicians when needed. (link)
Zocdoc opens its booking infrastructure to the rest of healthcare: Care Access Network lets payers, AI platforms, search engines, and other partners embed real-time provider search and scheduling directly into their products, with Amazon Health AI, Gemini, and Yelp among launch partners. (link)
GE HealthCare predicts hospital bottlenecks 72 hours ahead: CareIntellect for Operations uses hospital data to forecast capacity constraints across beds, staffing, imaging, and patient flow before they hit, with Queen’s Health Systems and Duke Health evaluating the system. (link)
Knowtex launches frontier AI lab for healthcare: The new lab will build and rigorously evaluate clinical AI for accuracy and auditability, building on a platform already spanning 200+ specialties across 300+ healthcare organizations. (link)
Medicare explores how to pay for AI medical care: Federal officials are considering a new payment category for AI software that provides medical care or diagnostic support, potentially creating a reimbursement pathway for increasingly autonomous clinical tools. (link)
Oracle brings its clinical AI agent to inpatient nurses: Embedded directly in the EHR, the tool supports voice-enabled charting, patient search, and acute nursing summaries, extending Oracle’s clinical agent beyond physicians into nursing workflows. (link)
Anthropic is running a physical biology lab for Claude: The company has established a Bay Area wet lab where it is exploring whether Claude can direct robotic laboratory equipment and participate in real-world biological experiments. (link)
CMS expands ACCESS to more chronic conditions: Starting in April 2027, the technology-supported Medicare model will add heart failure, COPD, substance use disorder, and tobacco cessation, expanding outcome-based reimbursement for chronic care delivered with technology. (link)
Neko Health shows measurable gains after three years: Data from 80,000+ members found all seven tracked cardiometabolic biomarkers improved among 9,138 returning members, with people managing hypertension, cholesterol, or diabetes seeing larger gains than healthy members. (link)
Sentara pilots AI recommendations directly at the point of care: A 90-day pilot will test GW RhythmX’s precision care platform with 30 physicians and APPs, evaluating clinical recommendations, workflow integration, trust, and adoption inside primary care. (link)
UW Health reviews 100+ AI features in a single upgrade: The volume of new capabilities arriving through one enterprise platform shows how quickly AI governance is becoming a core health system function, with UW assessing tools centrally based on risk. (link)
AI scribes are changing clinical work beyond saving time: A BMJ Digital Health review of 27 studies found ambient AI can reshape conversations and documentation behavior, while patient narratives and social cues remain largely missing from existing evaluation research. (link)
25 health systems build AI agents with Epic: Projects from Epic’s Build-a-Thon span medication review, safety investigations, surgery cancellation risk, and early disease detection, showing health systems increasingly building their own agents directly inside clinical workflows. (link)
Third Way Health finds AI is improving the front-office patient experience: Among healthcare executives already using AI, 79% reported improved patient experience, while the bigger implementation challenges were human backup, vendor management, and staff adoption rather than ROI. (link)
NVIDIA builds a simulation layer for healthcare robotics: Its medical physics platform can simulate catheter navigation, surgical procedures, and synthetic surgical video, giving developers a virtual environment to train and test AI-powered robotic systems. (link)
CMS’ AI prior auth pilot hits early operational problems: Documents from the first months of the WISeR model describe delayed approvals, technology failures, miscommunication, and provider frustration as CMS tests AI-supported prior authorization across six states. (link)
AI helps sonographers catch more fetal brain defects: In a randomized trial across five hospitals, AI assistance increased detection sensitivity from 78.6% to 87.3% without increasing false positives, while clinicians correctly overrode about 60% of the AI’s mistakes. (link)
OpenAI passes Anthropic on OpenRouter spend for the first time in 2.5 years: OpenAI captured just over half of combined OpenAI-Anthropic model spending during the week of September 7, with GPT-6 Astra leading wallet share. (link)
Open models are rapidly gaining share from proprietary AI: Open models hit 78.4% of token volume on Vercel AI Gateway, while Moonshot, DeepSeek, and Z.ai combined surpassed OpenAI in spend. The shift suggests more AI economics may move from the model layer toward infrastructure and applications. (link)
TypeSafe launches an AI model built for decisions, not chat: Jev returns structured answers with calibrated confidence rather than generating open-ended text, with TypeSafe reporting response times of 70 to 500 milliseconds and speeds 40 to 200x faster than conventional LLMs. (link)
Google launches Gemini 3.8 Live with real-time reasoning: The new voice-first models can reason while maintaining live dialogue, with Extended Thinking designed for more complex tasks requiring deeper reasoning and multi-step execution. (link)
PARTNERSHIPS
Amwell + Department of Veterans Affairs: The VA issued Amwell a non-binding letter of intent to deploy its virtual health platform across the agency, supporting virtual visits, specialist connections, and care coordination for veterans. (link)
NVIDIA + Children’s Hospital of Philadelphia: CHOP built its cardiac modeling program on NVIDIA’s open-source MONAI framework, reducing 3D heart model creation from hours to seconds for congenital heart disease care. (link)
Novo Nordisk + Anthropic: Novo Nordisk partnered with Anthropic to test Claude Science across R&D workflows and use Anthropic’s models to support drug discovery and software development. (link)
Memorial Sloan Kettering + OpenEvidence: MSK partnered with OpenEvidence to bring its OncoKB precision oncology knowledge base into the platform while integrating OpenEvidence into MSK’s Epic workflows. (link)
Ultrahuman + HealthEx: Ultrahuman launched HealthEx PowerPlug for U.S. Ring users, letting members view their medical records alongside continuous biometric data to identify gaps and missed follow-ups. (link)
HonorHealth + Qventus: HonorHealth joined Qventus’ AI Solution Factory to co-develop custom AI solutions alongside Qventus, expanding an existing partnership focused on improving hospital operations and margins. (link)
Century Health + Eye Health America: Century Health partnered with Eye Health America to use AI to transform more than a decade of clinical data into research-grade datasets for chronic retinal disease research. (link)
Cadence + ElevenLabs: Cadence built a voice AI agent using ElevenAgents to screen vital-sign alerts across 100,000+ chronic disease patients, cutting average time to patient contact from nearly two hours to 3.5 minutes. (link)
Counsel Health + Oura: Counsel Health expanded its partnership with Oura through CMS’s ACCESS Model to provide AI-native chronic care for hypertension, obesity, and prediabetes to eligible Medicare beneficiaries beginning in 2027. (link)
OpenAI Foundation + UNC Lineberger Comprehensive Cancer Center: The OpenAI Foundation awarded UNC Lineberger $40M to generate tumor and immune-cell datasets for training AI models aimed at improving personalized cancer vaccines. (link)
Deal Desk 💰
Spotlight on latest capital raises, M&A, and investments…
FUNDING
Angle Health, an SF-based health-care benefits platform, raised $600M at a $2.7B valuation. It includes $200M in Series C primary capital and a $400M tender. Vitruvian Partners led, joined by Town Hall Ventures and insiders Blumberg Capital, Portage Ventures, PruVen Capital, and YC. (link)
Thatch, a San Francisco-based employee health-benefits platform, raised $108M in Series C funding from The General Partnership, Index Ventures, General Catalyst, Andreessen Horowitz, and others. (link)
Penelope Health, a provider and payer policy intelligence platform, raised $100M from Thoreau, Bertelsmann Healthcare Investments, Twine Ventures and Seedcamp. (link)
Altis Labs, a Toronto, Ontario-based AI-powered software designed to predict drug efficacy, raised $25M in Series A funding. OrbiMed and Qiming Venture Partners USA led. (link)
Polyphron, a NYC-based biotech company using AI and lab-grown human tissue to test drugs, raised $20M in seed funding. Quiet Capital led the round and was joined by Gradient, Haystack, and Compound. (link)
Corridor, an NYC-based health benefits broker, raised $16M in seed funding. Bain Capital Ventures led, joined by BoxGroup, Definition Capital, and Constellation. (link)
Mithrl, an AI infrastructure company for biopharma R&D, raised a $20M Series A led by Obvious Ventures, with Headline and AGI House also participating. (link)
Fize Medical, an Israeli platform for predicting fluid risks, held a first close on a $20M targeted Series B round led by Asahi Kasei Medical and Rapha Capital Management. (link)
Ayble Health, an AI-enabled virtual clinic for GI and autoimmune conditions, raised a $16M oversubscribed Series A led by Neon, bringing total funding to over $27M. (link)
Ferry Health, an agentic AI platform that makes calls to find and schedule in-network care for patients, launched out of stealth with a $9M seed round backed by a16z and Index Ventures. (link)
Kairon Health, an AI-native execution layer that turns patient data into worklists for value-based care practices, raised $5M in venture funding led by Flare Capital Partners. (link)
Health Force, an AI agent platform automating hospital operations like insurance claims processing, raised €4.2M in seed funding led by LUMO Labs. (link)
M&A
Hello Patient + Converse Health: Hello Patient acquired the AI workflow automation company, expanding its agentic AI to cover the full patient journey, from first call through document and chart work to back-office billing. (link)
Medallion + Andros: Medallion acquired the NCQA-certified credentials verification organization, expanding its AI-assisted credentialing platform across health plans, health systems, and telehealth companies. (link)
Tempus AI + Personalis: Tempus AI agreed to acquire the cancer genomics company for approximately $1.5B, integrating its molecular residual disease testing into Tempus' AI precision oncology platform. (link)
as of 9/20/26
Other Relevant News 🔍
News, podcasts, blogs, tweets, resources, etc…
AI Job Opportunities 💼
Explore our AI Job Board or contact us to feature roles in our newsletter…
Visuals of the Week 📸
Funny memes, cool pics, and interesting data from around the web…
That’s it for this week friends! Back to reading — I’ll see you next week.
Stay classy,
PS. I write this newsletter for you. So if you have any suggestions or questions, feel free to reply to this email and let me know
How was this week's newsletter? Tap your choice below👇
