Bunkerhill Lands Series B to Turn Ideas Into Action

Bunkerhill Health might take its name from a short-lived medical drama, but all signs point to it sticking around for the long haul after landing $25M of Series B funding from a stacked investor roster led by Khosla Ventures.

Bunkerhill isn’t going after one workflow at a time. Its Carebricks platform lets health systems transform their ideas into AI agents that work across any clinical or operational use case – turning the data that hospitals already generate into action for the patients who need it.

  • That includes everything from automating administrative work and navigating prior auths to reviewing cardiac imaging for early signs of heart disease.
  • The value prop propelled Bunkerhill’s revenue over 20x just last year, which no doubt helped attract big names like Sequoia, Optum Ventures, and Y Combinator to the round.

The pitch flips the usual script. Instead of a vendor showing up with a fixed solution, health systems bring the clinical judgment and the problems that need solving.

  • Carebricks brings the execution, running the work as AI agents that reason across disparate data sources, take real action, and grow more capable as they work together.

The proof is in the customer pudding. Carebricks is live at over 15 health systems, including some familiar faces like Cleveland Clinic, Mayo Clinic, Intermountain Health, and the University of Texas Medical Branch.

UTMB alone has 20+ agents in production across the organization:

  • In its first month, a coronary calcium detection agent flagged a patient at imminent risk of a heart attack, leading to a life-saving triple bypass.
  • A nephrology triage agent cut specialist wait times by more than 50% by escalating urgent cases and routing others to telemedicine.
  • A lung nodule agent addressed urgent incidental findings 80% faster and doubled guideline-concordant follow-ups.

Vinod Khosla summed it up well: the bottleneck in healthcare AI was never the technology, it was getting health systems to actually run it.

  • Bunkerhill will use the new capital to expand Carebricks to new partners, while also expanding the platform’s capabilities so existing partners have even more ways to actually run it. 

The Takeaway

Healthcare has plenty of good ideas. It even has a couple that eventually end up improving care. Bunkerhill is making sure fewer good ideas get lost along the way, and it just raised $25M to fine-tune its compass.

Rock Health H1 2026: Durable Roots, Shifting Routes

Rock Health just dropped its H1 digital health funding overview, and the halftime report shows a market that knows where it’s going, even if the way it gets there keeps changing.

Here’s H1 2026 by the numbers:

  • Digital health startups raised $7.4B across 244 rounds (up $1B from H1 2025).
  • Median round size climbed from $12M to $14M, the highest since 2022.
  • 19 companies raised 20 mega-rounds, accounting for 45% of all capital invested.

That last stat is starting to look familiar. Capital concentration was the headline of the Q1 report, and the trend hasn’t let up. 

  • Just over 8% of rounds absorbed nearly half the capital, and some companies aren’t even waiting a full year between nine-figure checks.
  • Garner Health’s $100M Series E landed three months after its Series D, and Aidoc grabbed its second $150M in under a year.

Mental health is still the belle of the ball. It was the top-funded clinical indication for the seventh straight year, led by Talkiatry ($210M) and Grow Therapy ($150M).

  • Weight management took the silver on the back of the insatiable appetite for GLP-1s.
  • Both categories share a secret weapon: 64% of their H1 raisers go direct-to-consumer, versus 29% of digital health overall.

The IPO drought is creeping back. After seven exits last year, 2026 hasn’t produced a single digital health IPO, and Oura’s S-1 is the only one on file.

  • This year’s exit action is all M&A. H1 saw 115 acquisitions, including 71 in Q2 alone (the busiest quarter since 2021), with revenue cycle management consolidating fastest.

Where are all the moats? The H1 report’s big question was what actually counts as a durable advantage now that “we have AI” no longer moves the needle.

Rock Health landed on four answers and one graphic: 

  • founders with real domain expertise (sharpens product and buyer relationships)
  • platforms scaling to own more workflows (and more context to coordinate tasks)
  • hands-on delivery (forward-deployed engineers are officially a healthcare job)
  • network effects (every new partnership builds on the last so competitors can’t catch up)

The Takeaway

AI made digital health products easier to build than ever, which means the products themselves are no longer the moat. Investors are looking for the same fundamental roots – teams, trust, and traction – but the routes to find them are shifting faster than ever.

Pearl Health Closes $110M for Value-Based Medicare

The value-based care enablement segment just saw one of its biggest raises of the year after Pearl Health locked in $110M of financing, split between a $50M equity round and a $60M credit facility.

Pearl was founded on a simple belief. “Healthcare should reward keeping people healthy, not just treating them when they are sick.” Pearl AI is what makes that possible.

  • The platform helps PCPs manage risk and deliver better care to Medicare patients by translating clinical data into measurable outcomes.
  • That means leveraging AI to help manage and predict risk, orchestrate workflows, and automate action before issues become costly emergencies.
  • Over 10k providers caring for 250k+ Medicare beneficiaries are already live on the platform, including health systems like University of Vermont Health and MDX Hawaii.

VBC enablement has some strong tailwinds. As CMS continues pushing reimbursement toward outcomes rather than utilization, it’s creating stronger incentives for providers to prevent avoidable illness and manage patient populations. Investors are following the dollar signs.

  • Honest Health landed $140M back in February to scale programs supporting the same Medicare population as Pearl, and Chamber Cardio picked up $60M to take the specialty-specific route with cardiologists.

Pearl plans to separate from the pack with AI and outcomes. The fresh funds are going straight to the AI roadmap, including Performance Intelligence, a natural language interface that gives population health teams real-time insights on cost, quality, and utilization.

  • The raise will also help expand Pearl’s Care Orchestration AI agents, which automate annual wellness visit scheduling, post-discharge follow-ups, and care management outreach.
  • Beyond the platform, the raise will fuel Pearl’s expansion into Medicare Advantage and new risk offerings beyond Traditional Medicare.

Preventive care shouldn’t break the bank. Pearl hit a rare milestone in both the risk-enablement arena and the wider digital health universe by reaching profitability in 2025.

  • Rarer still, it pulled it off without taking its foot off the gas. Pearl now manages $3.6B in annualized medical spend (up from $2.4B), and is on track to triple its patient base from 2024 by the end of the year.

The Takeaway

With over 70M Medicare beneficiaries and costs exceeding $1T annually, Pearl is equipping providers with the picks and shovels needed to make the VBC transition – and some smart folks (or at least some folks with $110M laying around) think these docs can strike gold.

Behavioral Health Moves From Access to Accountability

The pandemic-era land grab in digital behavioral health is behind us, and a beautiful report from 7wire Ventures makes the case that the next chapter will be written by the platforms that can prove people actually get better.

Access was just the start of the story. Hundreds of behavioral health companies launched on the premise that technology could improve access at scale, and the capital followed. Digital mental health raised $4.9B in 2021 alone.

  • Despite the investment surge, the access problem is still very real. Nearly 1 in 4 U.S. adults experienced a mental health condition in 2024, yet almost half received no treatment.
  • The problem is that utilization growth outpaced meaningful clinical outcomes, and population-level ROI remained tough to verify.

The market is adjusting accordingly. Payors and self-insured employers are now conditioning contracts on measurement-based care – PHQ-9s, functional improvement data, and documented reductions in downstream medical costs – while commercial models shift from PMPM setups toward case rates and performance guarantees.

  • First-generation platforms that scaled enrollment without outcomes infrastructure are struggling to keep pace, and behavioral health M&A jumped 42% in 2025 as a result.
  • Recent tie-ups like Spring Health and Alma plus UHS’ $835M acquisition of Talkspace suggest the fragmented point-solution era is winding down in favor of integrated platforms.

AI is the primary lever. The provider shortage isn’t budging (HRSA projects a 43k psychiatrist shortfall by 2038), so the most scalable near-term play is technology that extends existing clinicians rather than trying to bypass them.

  • The report draws a sharp line between supervised AI (where a licensed clinician stays accountable for diagnosis and escalation) and autonomous agents, with the supervised camp winning payor confidence and reimbursement traction.

One prediction worth flagging: 7wire expects outcomes data to become the primary basis for payor and employer contract decisions, turning measurement-based care from a differentiator into table stakes that determines who even makes the RFP shortlist.

The Takeaway

Behavioral health went from digital health’s most funded clinical indication to a market where enrollment numbers no longer impress anyone. Access got the industry to the starting line, but the platforms that survive the accountability era will be the ones that can show their work – to payors, to employers, and ideally to patients.

New Model Predicts 900 Diseases From Real Records

Last week brought a potentially significant step forward for early diagnosis in the form of a new AI model that can predict a patient’s next diagnosis across nearly 900 diseases using just real-world medical records.

Meet DT-Transformer. Researchers at Harvard, Brigham and Women’s, and the Broad Institute unveiled the GPT-style foundation model in a new arXiv preprint.

  • DT-Transformer reads a patient’s medical history as a sequence and predicts which disease will show up next, and when it might come knocking.

The real headline is the training set. DT-Transformer was trained on 57.1M structured EHR entries from 1.7M patients across MGB’s 11 hospitals and 200 clinics (2000 to 2024) – the messy reality of U.S. clinical care rather than a polished research data set.

The numbers were impressive. A few standouts:

  • DT-Transformer achieved a median AUC of 0.871 across 896 disease categories, with AUC over 0.5 for every condition.
  • The model crushed an age- and sex-based baseline by +0.214 AUC (0.871 vs. 0.657), beating it on 96% of diseases.
  • All of that was accomplished with a featherweight 2.2M parameter model that’s small enough to run just about anywhere (by comparison, Claude Fable 5 has about 6 trillion parameters).

The real test told a humbler story. When the team ran a true prospective test forecasting new diagnoses DT-Transformer had never seen, median AUC slipped to 0.713.

  • That still beats the baseline on 80% of diseases, but that gap between the retrospective flex and the prospective reality is fairly significant.  
  • A 0.871 headline number and a 0.713 crystal ball aren’t the same product, and the second one is the one that patients would actually have to deal with.

One other highlight worth mentioning: including every repeated diagnosis worsened model performance, “drowning out the signal” rather than improving predictions – another reminder that more data isn’t always better. Better data is.

The Takeaway

Population-scale risk forecasting that runs on a model smaller than most phone apps is a real milestone, and training on routine records instead of a spotless biobank is exactly the kind of thing that could get models like DT-Transformer in front of actual patients – assuming the data holds up to peer-review.

UpDoc Lands First FDA Clearance for Patient-Facing AI 

UpDoc just landed the first FDA clearance for a patient-facing AI model, which acts as a “concierge doctor” to support patients between visits. Good news for the human docs reading this – it isn’t going after your job just yet.

What’s UpDoc? It’s a clinical AI platform that unifies clinical guidelines, longitudinal patient context, and physician governance to safely execute real-world care workflows.

What isn’t UpDoc? An AI doctor.

  • The 510(k) clearance had a narrow scope. It allows the AI to call or message patients between visits and adjust their insulin doses within parameters set by human clinicians.
  • UpDoc says its AI will ease doctors’ workloads and help patients better manage illnesses like Type 2 diabetes. 

The data backs that up. A study in JAMA Network Open saw 32 patients with T2D randomized to receive support from UpDoc (daily voice AI check-ins to record blood glucose and adjust insulin) or standard care (AKA log their own data until they see their doctor in person).

  • The AI group hit their target blood glucose in 15 days, compared to the standard care group where less than half got there at all within the 8 week study period.
  • That trial provided the clinical foundation for the now-cleared solution, which is set to be piloted at Cleveland Clinic, UCSF Health, and Allegheny Health Network.

UpDoc is taking the road less traveled. It’s not the only AI startup in this wheelhouse, but so far it’s one of the only ones that doesn’t seem to be actively avoiding FDA regulation. 

  • The most notable example is Doctronic, which has been testing its AI prescription tech through a state-run program in Utah rather than seeking a full-fledged authorization.
  • That’s an easier path to market than vaulting over FDA hurdles, but it doesn’t get you the “world first” feather in your cap that now belongs to UpDoc.

So, now what? The FDA has long debated how to regulate AI, and UpDoc could be the first sign that they’re getting comfortable enough with the tech to give the green light to more models.

  • With the first clearance out of the way, other AI developers also have an established precedent and a blueprint to follow suit.

Don’t forget about the docs. Besides the regulatory shakeout, it’ll be equally interesting to see how this new breed of AI ends up in the hands of clinicians.

  • We saw OpenEvidence fold a new biomarker for heart disease into its platform just last week, and the most direct path to a wide distribution for many soon-to-be-cleared AI tools could be similar licensing partnerships.
  • Plenty of companies already have a massive user base and are actively expanding the clinical scope of their platforms – Abridge, OE, Doximity, the list goes on for a while. It feels like licensing models from the UpDocs of the world is a natural next step after all the journal partnerships we’ve been seeing now that FDA clearance is part of the picture.

The Takeaway

The FDA finally cleared its first patient-facing clinical AI model, and UpDoc might have been the first domino, but it definitely won’t be the last.

Assort Closes $120M to Scale Voice AI Across Healthcare

If you needed any more proof that communication friction is one of the biggest pain points for patients and providers, look no further than Assort Health’s just-closed $120M Series C – its third funding round in 18 months.

Assort started with a simple thesis. Unlock the front door of healthcare, and the rest will follow. Assort originally aimed its voice AI agents at scheduling because it meant solving for two key ingredients needed to solve everything else downstream: 

  • The care protocols required to handle that first interaction.
  • The patient communication data that flows into the rest of the journey.

The first call is an important moment. Mistakes here mean the patient never comes back, and Assort’s edge in preventing that is its Synapse agentic model.

  • Synapse learns specialty workflows across every deployment, then simulates the edge cases to stress test them before any agents go live.
  • That allows even non-technical teams to safely implement Assort’s agents at scale, which fuels an AI development flywheel that’s already learning from 190M patient interactions, 62M care protocols, and 1.6M decision pathways.

Assort covers the entire patient journey. What began as the first voice AI agent to schedule a specialty appointment has grown into a full-fledged voice AI platform that includes:

  • Concierge – handles inbound calls, triage, lab requests, med refills, scheduling, eligibility checks, and intake.
  • Activate – reaches patients proactively to close referral loops and act on care gaps, recover no-shows, and resolve payments.
  • Orchestrate – runs the operational work behind each visit and writes every detail back to the EHR.
  • Empower – equips staff with an AI copilot to manage complex patient access needs in real time.

Patient Journey Memory ties it all together. The capability is built on three pillars.

  • Each patient gets a personal AI agent that knows their context and preferences so they don’t have to keep repeating the same story every time they interact with their provider.
  • The agents share the same data and talk to each other, so care gaps surface wherever the patient happens to engage.
  • Having a continuous journey across every interactions allows the platform to activate patients when they’re high intent.

Next stop: everywhere. Every new tech generation sees a flood of new solutions, then only a few survive. Voice AI is about to hit that same shakeout, and Assort plans on sticking around.

  • The funding was earmarked for bringing on veteran C-levels to make that happen, and expanding into health systems ranging from community-based organizations to the biggest academic medical centers in the country.
  • Major systems like John Muir Health are already signed on as demand grows for platforms that can support increasingly complex ambulatory operations – the exact kind Assort is uniquely tuned to solve.

The Takeaway

Assort is looking to become the voice AI transformation partner for every healthcare provider in the country, and if its funding tempo is any indication, it’s moving with enough urgency to actually pull it off.

Cadence Lands $100M to Scale Chronic Care

One of the hottest names in chronic care just got nine-figures hotter after Cadence hauled in $100M of Series C funding to scale up with AI agents.

Welcome to the unicorn club. The round vaulted Cadence’s valuation to $1.2B as it looks to extend the reach of its Clinical Intelligence platform to more patients managing chronic conditions like hypertension, diabetes, and heart failure.

  • The platform is purpose-built to deliver continuous support between visits, with regular vital monitoring and AI agents that translate the data into timely medication adjustments and personalized lifestyle coaching. 
  • Cadence integrates directly with its partners’ EHRs and clinical workflows, then equips every patient with connected devices that allow its system – supervised by physicians – to flag high-risk patients and predict adverse events before they happen.

Cadence lets its numbers do the talking. It’s easy to see why.

  • 70% relative improvement in blood pressure control (JACC).
  • 27% fewer hospital admissions (Mayo Clinic Proceedings).
  • 230% increase in heart failure patients on GDMT (JCF).
  • $1,300 annual cost reduction per patient (Circulation).

Now it’s time to scale. Cadence is already treating 100k patients, and this investment will support the infrastructure “to treat millions.”

  • The current partner roster includes over 20 major health systems, including new additions like Texas Health Resources and Duke Health, which liked working with Cadence so much that it participated in the round.
  • The word on the street is annualized revenue is pacing $140M for FY 2026, more than double the $62M it brought in last year.

Last but not least. Scaling from thousands of patients to millions of patients is no small task, and Cadence is working on deploying new AI agents to help carry the load.

  • A new prescription hypertension management agent is reportedly en route, which “pre-authorizes” medication recommendations to steer patients toward their blood pressure goals, with new prescriptions and dose changes made autonomously by the AI.
  • That falls squarely in medical device territory, and Cadence is actively working with regulators to bring it to patients – including an application to the FDA’s TEMPO pilot that will let certain devices for chronic care on the market before they’re officially approved.

The Takeaway

Chronic disease is the single largest driver of healthcare spending in America, and a significant amount of that cost is avoidable. Cadence is building the tools to avoid it, and another $100M should only help the cause.

New Studies Show AI Outperforms Physicians, Just Not at Medicine

In case last week’s AI drama wasn’t hot enough, a pair of new studies in Nature cranked up the heat by finding that AI agents beat physicians on ER and care management tasks – just not real ones.

“Towards autonomous medical artificial intelligence agents.” The first study took a look at MIRA, an AI agent developed in Germany that operates inside a sandboxed EHR environment.

  • Using 574 real emergency department cases, researchers had MIRA chat with another patient agent and execute entire care workflows, such as investigating diagnoses, ordering labs, and triaging for hospital admission. 

The headline: MIRA significantly outperformed four board-certified physicians. The agent had higher overall diagnostic accuracy (87.8% vs. 78.1%), was better at ordering correct procedures like laparoscopic appendectomy (53.5% vs 38.3%), and had 35% better guideline alignment.

The reality: ER doc Graham Walker, MD, put it perfectly on LinkedIn: “There is no way in hell that humans mismanaged almost 30% of appendicitis cases, the most common ‘surgical emergency’ that we’ve all seen hundreds of in our career.”

  • It turns out the EHR sandbox needed 21 keystrokes to get this right, and the physicians failed unless they explicitly searched and entered a “laparoscopic appendectomy.” AI is built for that, humans not so much.

“Towards conversational AI for disease management.” The second study explored whether Google’s AMIE agent could expand from pure diagnostics to longitudinal care management.

  • The blinded study pitted AMIE against 21 primary care physicians on 100 multi-visit cases, with the agent pulling live guidelines and drug references to produce structured management plans.

The headline: AMIE’s care plans were better than PCPs across the board. The agent notched higher marks on management reasoning, precision of investigations, and guideline alignment.

The reality: AMIE operated in a world without prior auths, without formulary restrictions, and without social needs that patients didn’t want to bring up. The authors didn’t pretend otherwise.

The Takeaway

This might sound familiar, but these studies show that MIRA and AMIE performed well in ideal scenarios, not in the messy trenches of real-world medicine. That said, the results aren’t important because AI beat a benchmark, they’re important because AI took another big step toward “delivering actions” instead of just “delivering answers.”

Easing the Data Burden in Diabetes Care

By Mark Clements, M.D., Ph.D. and Trisha Martinez , RN, BSN, MBA, Glooko

Health systems are not short on diabetes data. They are short on the time, workflows, and signal clarity needed to act on it.

That distinction matters. Continuous glucose monitoring (CGM), connected devices, remote uploads, insulin delivery data, and inpatient glucose trends are generating more information than ever before.

In Glooko’s latest Annual Diabetes Report, more than 60 billion CGM readings flowed through the platform in 2025, supported by more than 1 million active patients, 30,500 clinicians, 9,000 clinics, and a global footprint across 1,082 geographic locations.

  • The scale is no longer the story, but what health systems do with that scale is.

For leaders focused on digital transformation, the opportunity is to shift diabetes management from retrospective review to prioritized action.

  • Traditional measures such as Time in Range (TIR), Glucose Management Indicator (GMI), Time Below Range (TBR), and Time Above Range (TAR) remain essential, but they can obscure when risk occurs, how severe it is, and which patients need attention first.
  • Two patients may look similar by familiar metrics, yet one may face recurring overnight lows while another carries persistent daytime hyperglycemia.

The report’s overnight hypoglycemia analysis illustrates why this matters. Glooko’s model surfaced patients who appeared near target by common measures but had substantially higher overnight hypoglycemia exposure.

  • In validation across 586,549 patient weeks, the highest-risk group showed a 2.79x lift in identifying observed overnight hypoglycemia compared with baseline selection.
  • This is where digital transformation becomes clinical transformation: turning connected data into prioritized lists, cohort-level visibility, and workflows that help teams intervene between visits.

The same safety lens extends into the hospital. EndoTool provides an inpatient view of glycemic management, supporting individualized insulin dosing during complex episodes such as Diabetic Ketoacidosis (DKA) and Hyperosmolar Hyperglycemic State (HHS), renal impairment, steroid exposure, and changing nutrition status.

The Takeaway

The next chapter of diabetes care will not be defined by more dashboards. It will be defined by connected intelligence that helps health systems identify risk earlier, focus clinical attention, reduce cognitive burden, and support safer decisions across the hospital, clinic, and home.

Learn how Glooko is partnering with organizations like yours to transform diabetes care.