Sponsored by: Jimini Health
The frontier of AI, Mental Health, and the Brain.
Neural Compass spotlights the most cutting-edge scientists, clinicians, and executives at the forefront of AI and clinical care. Guests have included former NIMH director Tom Insel and Yale psychiatry chair John Krystal. The podcast reflects Jimini Health's experience deploying AI in behavioral healthcare and Mark's background as a former executive at an AI-driven biotech startup valued at $1B. Listen now.
Good morning, readers —
We hope you had a nice Halloween weekend! 🎃
This week, the big theme is agents moving from answering questions to actually doing things across healthcare, and what happens when that automation meets the incentives already built into the system.
Healthcare is starting to build for an agentic internet
The EHR is becoming agentic
The US health economy is a tragedy of the commons
52 tools, partnerships & deals, plus new research, policy updates, AI jobs & link-worthy content
Read time: 7 minutes
Our Picks ✨
Highlights if you’ve only got 2 minutes…
1/
Healthcare is starting to build for an agentic internet
A few announcements this week give a pretty good preview of what happens when AI agents start navigating healthcare on a patient’s behalf.
athenahealth + Claude: athenahealth launched an MCP connection that lets patients securely connect their medical records to Claude. With permission, Claude can pull labs, medications, diagnoses, vitals, visit history, allergies, procedures, and care plans directly from athenahealth. That gives consumer AI much more useful context about the person asking the question. (link)
Zocdoc + Instinct: Zocdoc is opening its infrastructure to agents so they can find providers, verify insurance, check availability, and book appointments. Agent traffic on Zocdoc is up nearly 30x since July, and the first two weeks of October surpassed the previous three months combined. (link)
UnitedHealthcare: expanding Avery, its AI health companion, from roughly 6.6M members earlier this year to 21M. UHC says Avery resolves about 90% of member interactions without a human handoff and is adding the ability to call providers and schedule appointments. (link)
General Medicine + Walmart: General Medicine is powering Walmart’s Better Care Services marketplace, bringing care options, coverage information, and pricing into one catalog. It is also the kind of infrastructure an agent could eventually search and navigate for a patient. (link)
athenahealth gives an agent access to the patient’s actual medical context. Zocdoc gives it somewhere to find and book care. Avery shows what this can look like at real scale, with tens of millions of members already interacting with an AI that can increasingly do things on their behalf.
The rest of the internet is starting to wrestle with the same shift. Reddit just announced plans to move toward a mostly logged-in experience to prevent unauthorized scraping and automated traffic. Amazon has also been pushing back on third-party shopping agents while continuing to build its own AI shopping products.
Zocdoc is taking almost the opposite approach and explicitly building for agents. If software starts doing more searching, comparing, booking, and transacting for us, healthcare companies may have to think about a new user alongside the patient: the agent representing them.
That said, the economics are still a bit messy. Elon Musk just said this weekend that the cost of serving Grok Bot had become “unsustainably high relative to revenue.” A recent NEJM AI study found AI-recommended care averaged $314 per visit versus $97 for physicians, largely because models ordered more tests and specialist referrals.
Agents could make healthcare much easier to access and navigate. They could also make it much easier to consume more healthcare. Both are probably coming.
2/
The EHR is becoming agentic
For decades, the EHR has mostly been a system of record. Now AI is starting to turn it into something that can actually help get work done.
For a Healthcare AI Guy Market Intelligence piece, Uday Suresh, PhD, a biomedical informatics researcher and product manager, and Karthik Ramesh, MD, an internal medicine resident, look at what happens when the EHR moves from documenting care to helping coordinate and execute it.
We are already seeing the first pieces. Ambient tools are expanding from notes into orders, coding, referrals, prior auth, and revenue cycle work. AI-first EHRs and specialized agents are going a step further and trying to complete workflows across the chart and the other systems around it.
A lot of the most useful applications are pretty unglamorous. Think chasing a referral that never got scheduled, coordinating a SNF discharge, reconciling medications, or following up on a prior auth. Today, that work often goes to case managers, staff, or a patient’s family. For patients without someone helping them navigate the system, getting more of those loose ends handled could make a real difference.
There are still some basic questions nobody has fully answered. If an agent places the wrong order, acts on bad information, or misses something important, who is responsible? The technology is moving into these workflows faster than the rules around oversight and liability.
The AMA is starting to grapple with this too. Its new clinical AI liability framework argues that responsibility should follow who actually has the authority and ability to prevent harm.
The piece also makes a good point about where this could go wrong. Faster billing matched by faster denials does not make healthcare much better. Some of these workflows should probably disappear rather than become more efficient.
The point isn’t to make every existing workflow move faster. Some of these workflows probably shouldn’t exist in the first place. The bigger opportunity is figuring out which work AI should take over, which decisions still need humans, and which parts of the current system we should stop doing altogether. (link)
3/
The US health economy is a tragedy of the commons
Trilliant Health’s latest report essentially argues that every player in US healthcare is making rational decisions for themselves, but the system keeps getting worse as a result. A few numbers make the case:
Employee premiums and deductibles are up 151% since 2000, versus 73% growth in wages and 55% inflation.
By 2038, physician supply is projected to meet just 88% of demand, with adult psychiatry at 50% and primary care at 80%.
CMS tracks 800+ quality measures, but only 27% are tied directly to outcomes.
In Los Angeles, the hospital charging the highest negotiated rate for a heart attack also had the fifth-highest mortality rate among 41 hospitals studied.
GLP-1 patient volume rose 720% from 2019 to 2025 while bariatric surgery fell 32%, showing how quickly new treatments can disrupt old revenue pools.
The report’s bigger point is that healthcare still rewards activity more reliably than health. Providers get paid to intervene, payers try to contain spend, and patients carry more of the cost.
Agents can make healthcare cheaper, faster, and easier to navigate. But they will still operate inside the incentives we give them. If the system rewards more tests, more claims, more denials, and more procedures, AI can scale those too.
That may be the bigger lesson from all three stories. Better technology can help a lot, but at some point we also have to fix what the healthcare system is paying everyone to do. (link)(twitter)
Healthcare AI Rundown 🔍
Latest on tools, partnerships, research and policy…
Tools & Product Updates 🔧
Counsel Health launches physician-led AI weight management: Members receive on-demand physician support and AI check-ins during GLP-1 treatment, with the program also available through LillyDirect. (link)
eClinicalWorks launches AI Workbench for healthcare workflow agents: Organizations can build and manage agents for revenue cycle, referrals and prior authorization as the company expands into hospital care. (link)
Ro launches Robin, an AI health agent: Robin uses clinical notes, prescriptions, patient feedback and Apple Health data to personalize suggestions and draft messages to a patient’s doctor. (link)
Doximity adds a safety check for AI-generated drug doses: Free DoseCheck verifies recommendations against structured drug information and professional guidelines, adding rule-based verification inside Doximity Ask. (link)
Abridge adds in-visit orders and chart-informed inpatient notes: Its October update stages orders in Epic for clinician review and signature, incorporates chart data into inpatient documentation, and adds an evaluation dashboard. (link)
Corti expands controls for teams of healthcare AI agents: Its Agentic Framework adds orchestration, tracing and governance tools for developers building complex, multi-agent healthcare systems. (link)
OpenAI launches GPT-6.1 Terra for faster computer-based work: The model combines visual reasoning, computer use and writing, with standard API prices 50% below GPT-5.6 Terra. (link)
Kipu brings AI agents into behavioral health workflows: The company says its Utilization Review Agent has prepared 23,000 reviews, while its AI tools have supported 3M+ clinical notes. (link)
Evidently brings patient-history insights into EHRs and AI tools: Evidently Anywhere uses background agents to review notes, scans and outside records, with APIs supporting integrations including Nabla. (link)
Tecsys adds AI agents for hospital pharmacy risks: New TecsysIQ capabilities flag 340B compliance issues and expiring inventory, helping pharmacy teams make purchasing and stock-management decisions. (link)
Sohar Health connects insurance checks into automated access workflows: Flows chains eligibility verification and coverage discovery through one integration, with configurable rules for follow-up actions. (link)
Rhapsody launches Guardian to control AI access to healthcare data: The platform verifies agent identities and enforces permissions governing access to health information, APIs and services. (link)
UT Austin forms an academic health system around AI: UT Dell Medicine will unite its medical school, clinics and planned hospital, with AI built into care and operations. (link)
Superscript opens real-time healthcare pricing API nationwide: Practices and EHR vendors can integrate upfront patient pricing across 3,000+ services and 25+ specialties, according to the company. (link)
Google launches practical AI training for healthcare professionals: The self-paced certificate takes under 5 hours and offers up to 5 CME credits, covering administrative work, communication and responsible use. (link)
RevSpring adds phone-based payment plans to its AI agent: Let’s Talk expands patient self-service, with early company results reporting 36% higher payment rates and 27% fewer human handoffs. (link)
symplr combines AI scheduling and timekeeping for hospital staff: The company says its predictive workforce platform can cut staffing and timecard work approximately 50%, with additional AI capabilities planned for 2027. (link)
Partnerships & Deployments 🤝
NYU Langone + OpenEvidence: The health system is embedding patient-aware clinical AI into Epic across its hospitals and outpatient network, supporting more than 6,000 physicians and faculty. (link)
Microsoft + Concurrence: Concurrence’s Health World Model and AI agents are available through Microsoft Marketplace, allowing healthcare organizations to deploy them within existing Azure environments and governance controls. (link)
OpenEvidence + American Society of Addiction Medicine: The partnership will bring ASAM’s clinical guidelines and implementation resources for substance use disorder treatment into OpenEvidence’s AI platform. (link)
Tremont AI + AstraZeneca: The partners will develop and test AI for preclinical toxicologic pathology, aiming to speed drug safety screening before treatments reach human trials. (link)
NVIDIA + C-BRAIN: The 18-institution consortium is building AI research pipelines that draw on unpublished findings, unsuccessful experiments and proprietary Alzheimer’s data using NVIDIA’s tools. (link)
Edison Scientific + Pfizer: Pfizer will deploy Edison’s AI platform and Kosmos research agent across R&D, using proprietary data to support drug discovery and development decisions. (link)
egnite + Pathway Labs + Stamford Health: A pilot combines FDA-cleared ECG-based AI with CardioCare to help clinicians identify and manage patients with structural heart disease. (link)
Anthropic + Enveda: Claude reduced human review time 89% in manufacturing-document tests and produced a first regulatory draft in 4 hours versus approximately 3 months manually, the companies report. (link)
Research & Evidence 🔬
AI detects missed specialist referrals across 7 primary care clinics: MR-AI scanned approximately 25,000 notes weekly, with a quasi-experimental study estimating sevenfold higher missed-referral detection and requiring clinician verification. (link)
Noah Labs begins pivotal trial of voice-based heart-failure monitoring: Approximately 450 patients will test whether daily voice recordings detect deterioration, with results intended to support FDA authorization of the investigational device. (link)
Hudson Labs and Citrini map 353 AI drug-development companies: The interactive report organizes companies across the value chain and includes 705 sourced statements about AI’s role in drug development. (link)
AI models repeatedly recommend the same drug product names: Across 5 models, leading products averaged 97.5% of first-choice recommendations for each disease–ingredient pair, often favoring originator brands over generics. (link)
Only 1 in 4 Americans comfortable with providers using AI: In a nationally weighted survey of 500 adults, 88.4% expected clinicians to explain care plans better than AI chatbots. (link)
Novellia finds major gaps in Ehlers-Danlos patient histories: Across 494 patients, health systems typically missed a third of their medical history, and median time to diagnosis was 8 years. (link)
58% of life sciences leaders say AI standards remain unclear: ACTO’s survey of 135 leaders found 64% have governance in place or under development, while 54% worry about internal AI use. (link)
17% of surveyed employers have live AI benefits tools: An Aon and Define Ventures survey found another 30% prioritize AI-enabled benefits but have not yet chosen a vendor. (link)
AI helps trial staff detect immunotherapy side effects faster: A Nature Cancer crossover study found AI assistance reduced annotation time by 2.71 minutes while improving accuracy and agreement. (link)
OpenAI reports 3 AI-assisted advances across scientific fields: Results from unreleased models cover harmonic analysis, complexity theory and quantum field theory, with academic review and interpretation underway. (link)
Policy & Regulation 🏛️
HHS cuts 14 certification criteria with final HTI-5 rule: The rule revises 7 more, with $1.5B+ in projected regulatory savings; proposed information-blocking changes remain under review. (link)
Johns Hopkins Bayview notifies 3,182 patients after AI upload: An employee sent patient information to an unapproved public AI tool; the hospital requested deletion and offered 12 months of credit monitoring. (link)
Cleveland Clinic Cancer Institute forms an AI task force: The group will vet tools, guide adoption and continuously monitor performance and safety across clinical and operational workflows. (link)
AAMC sets AI competencies as hospitals guard against deskilling: The framework emphasizes independent reasoning, oversight and recognizing errors, while CHOP develops proficiency requirements before trainees access certain AI tools. (link)
OpenAI faces wrongful-death suit over seizure-medication advice: The complaint alleges ChatGPT designed an antiseizure medication taper and told a patient to conceal it from doctors before her death. (link)
Hackers used Claude Code to breach healthcare organizations: CrowdStrike reported compromises at 15+ Argentine healthcare organizations, with other AI-assisted campaigns targeting healthcare and biotech companies elsewhere. (link)
Deal Desk 💰
Spotlight on latest capital raises, M&A, and investments…
FUNDING
BioGeometry, an AI protein design company, raised new Series A+ funding led by China Life Capital, bringing total Series A financing above $100M to expand its AI protein design platform. (link)
nPhase, an Atlantic Beach, Fla.-based clinical data platform, raised $30M in Series D funding led by Viking Global Investors. (link)
Vera Health, an AI-powered clinical search platform, raised $28.5M in Series A funding to expand its technology and accelerate growth. (link)
Midstream Health, a San Francisco-based AI financial operations platform for health systems, raised $20M in Series A funding. Valtruis led, joined by Andreessen Horowitz and Foundry. (link)
ClarityCare, a healthcare admin AI company, raised $16.5M in Series A funding led by Headline, bringing total funding to nearly $20M to expand its AI agents for payer workflows. (link)
adsilico, a University of Leeds spinout, raised £10M in Series A funding led by PXN Ventures to expand its AI-powered virtual patient simulations for medical device testing. (link)
SpendRule, an AI contract enforcement platform for health systems, raised $4M in seed funding led by Mark Cuban’s Radical Investments, bringing total funding to $6M. (link)
M&A
Anthropic + Mithril Medicine: Anthropic acquired the one-person company developing treatments for two rare forms of muscular dystrophy, adding them to its own drug discovery programs. (link)
Chapter + Hively: Chapter acquired the retirement navigation platform to expand beyond Medicare and support older adults through post-career life. (link)
Net Health + Tissue Health Plus: Net Health acquired key clinical technologies from Tissue Health Plus to strengthen its AI-native specialty care platform. (link)
Equiti + Lexi: Equiti acquired the AI language interpretation company to pair instant AI language support with human medical interpreters across the patient journey. (link)
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👇
