Best of LinkedIn: Health Tech CW 14/ 15

00:00:00: This episode is provided by Thomas Allgaier and Freeness, based on the most relevant LinkedIn posts of HealthTech in CW-Fourteen and Fifteen.

00:00:08: Freenes equips health tech providers with a market intelligence to identify which hospitals to target.

00:00:14: And

00:00:24: welcome back to the deep dive, everyone.

00:00:25: We have some incredible curated insights for you today directly from The Conversation's health tech professionals we're having right now.

00:00:32: Yeah!

00:00:40: pretty wild.

00:00:41: It really is.

00:00:42: we're going to cover how AI is finally shifting from.

00:00:44: you know just a cool capability, too.

00:00:46: actual workflow orchestration

00:00:48: right plus.

00:00:48: We were looking at this new wave of agentic AI that's kind of threatening both startups and consumer

00:00:53: apps.

00:00:54: Yeah And the whole regulatory battle over shadow AI which is fascinating.

00:00:57: Oh absolutely?

00:00:58: And will wrap up by looking at the digitization of the OR and well The pretty dark reality of medtech cybersecurity.

00:01:05: it's a packed agenda.

00:01:06: it Is so okay let's unpack This first theme Because watching the health tech industry adopt artificial intelligence for the last few years.

00:01:17: It's felt a bit like Watching someone buy a thousand horsepower engine,

00:01:21: right?

00:01:21: Yeah And

00:01:22: then they mounted on a wooden cart and I wonder why they aren't winning the race.

00:01:27: We've been so obsessed over The raw power of the technology

00:01:30: models algorithms yeah

00:01:32: exactly but completely ignoring how it actually connects to the wheels.

00:01:36: That is the perfect visualization of where the industry has been stuck, honestly.

00:01:41: We've had this surplus of raw intelligence but a total deficit in orchestration and looking at The LinkedIn Post from That realization has finally hit the mainstream.

00:01:53: The conversation is completely pivoted

00:01:55: away from just marveling at what AI can

00:01:57: do exactly toward figuring out how to actually integrate it into a daily grind of health care, which

00:02:02: completely flips the script on what investors are looking.

00:02:05: I mean for the longest time everyone was funding the intelligence.

00:02:07: nobody was finding the plumbing

00:02:09: right?

00:02:09: The plumbing Is What Matters Now.

00:02:11: Eric Allen posted her really sharp observation about this on LinkedIn.

00:02:15: He pointed out that the AI modder itself is no longer the bottleneck.

00:02:19: Yeah, the models are mostly commoditized.

00:02:20: now exactly

00:02:22: he says The real competitive advantage right?

00:02:24: Now isn't what he calls constraint architecture and well workflow orchestration.

00:02:30: Right, and we really need to unpack how that actually works mechanically because constraint architecture isn't about making the AI smarter...

00:02:37: It's about boxing it in!

00:02:38: Exactly-it is the rules layer The permission boundaries..the structured workflows That lock a large language model To specific safe pathways inside an electronic health record.

00:02:49: Because as Michelle Carnahan pointed out In her post And this such great quote insight without action Is just noise.

00:02:55: It's so true.

00:02:56: I mean, identifying a care gap like you're predicting a patient might develop sepsis...

00:03:00: Mathematically impressive

00:03:01: Very impressive But operationally useless.

00:03:04: if the alert just sits in a dashboard The value is entirely in the execution layer.

00:03:09: Right Like are you automating the order for blood culture?

00:03:13: Are you contextualizing that alert with real-time financial and clinical data for the attending physician?

00:03:19: Yeah, And we are finally seeing real world evidence of this working at scale.

00:03:24: Tyler Stanley from Tempus AI shared the results of their alert trial which is a great example.

00:03:29: Oh yeah.

00:03:29: The alert trial That's a big one.

00:03:31: It was a cluster randomized study across thirty five hospitals and they focused on patients with severe aortic stenosis who are historically heavily undertreated.

00:03:42: So they used AI-enabled notifications directly inside the EHR to flag these patients,

00:03:48: but they didn't just flag the data and leave that doctor to figure it out which is the crucial difference here

00:03:53: right?

00:03:53: They embedded into workflow exactly

00:03:56: by embedding those personalized alerts Right Into The Act of Workflow meaning the AI actually drafted the next logical step for the physician To Approve.

00:04:04: they saw a forty percent increase in life saving valve interventions.

00:04:08: Wow, within ninety days.

00:04:10: Within

00:04:10: ninety days forty percent.

00:04:11: That outcome completely reframes the whole conversation around alert fatigue.

00:04:17: You know, we always hear that physicians ignore EHR alerts because they're burned out.

00:04:22: Yeah They

00:04:22: just click past them Right.

00:04:24: But looking at the mechanics of the alert trial...they don't ignore alerts due to burnout!

00:04:28: ...They ignore them because most alerts carry zero clinical utility.

00:04:32: There's liability pop-ups

00:04:33: Exactly but when you tie the right diagnostic signal To a specific guideline recommended action At the exact moment they are reviewing The chart..you change the economics Of their time and reduce cognitive load.

00:04:46: I want to pull on that thread about cognitive load and time, actually.

00:04:50: Because i'm looking at a statistic near Berenzweig posted And it kind of contradicts the idea that AI is currently saving us Time.

00:04:57: Oh!

00:04:57: The imaging turnaround stat?

00:04:59: Yeah He noted.

00:05:00: over last decade Imaging interpretation turn around time has increased by one hundred thirteen percent.

00:05:07: It

00:05:07: more than doubled.

00:05:08: If we have all these incredible AI detection tools highlighting anomalies on stands Shouldn't radiologists be moving faster?

00:05:15: You would think so, right.

00:05:17: But it comes down to how the technology interacts with human capacity.

00:05:22: Berenswig rightly points out that this isn't an access problem and its not an accuracy problem either.

00:05:28: It is a system-wide time orchestration failure

00:05:32: Because we have four case scans now.

00:05:33: Exactly An MRI might contain thousands of slices So the AI might highlight twenty potential anomalies But the radiologist still holds the legal liability.

00:05:44: Right,

00:05:45: they have to review every single one of those AI-generated highlights...

00:05:48: Every single one!

00:05:48: They have cross reference them with patient's history and then document their findings.

00:05:53: so we increase signal but also massively increased the review burden.

00:05:58: Human capacity just cannot scale that compounding demand.

00:06:02: No

00:06:02: it can't.

00:06:03: So the AI is essentially generating more homework for the clinician.

00:06:06: If you're building health tech today Building a better detection algorithm Well, you're just contributing to the bottleneck.

00:06:12: Precisely!

00:06:13: If we connect this to The Bigger Picture AI has become an operating system of clinical time.

00:06:18: It's managed timeline end-to-end And Anthony Steele highlighted a practical example of taking root.

00:06:25: Oracle Health's Clinical AI Agent is already live in UK via NHS.

00:06:31: Right and the mechanism there is ambient voice technology right?

00:06:33: Yeah

00:06:34: it listens in.

00:06:35: The goal isn't just dictate notes faster but get the clinician away from keyboard entirely letting the AI listen, structure the documentation in the background map it to billing codes so that human can actually look at the patient and the eye.

00:06:47: Yes!

00:06:48: The AI moves from being a passive tool you query... ...to an active participant orchestrating the room which is a profound shift.

00:06:56: Oh absolutely

00:06:57: If AI can orchestrate work flow or take actions into the background.

00:07:00: its not just software anymore.

00:07:02: It's evolving into agent

00:07:03: Which brings us to massive shifts of consumer space.

00:07:06: moving onto our second theme here if an AI agent can proactively manage my health data.

00:07:12: I mean, i'm looking at these massive consumer health apps like MyFitnessPal or Calm their entire user interface becomes obsolete overnight right?

00:07:21: Oh a hundred percent.

00:07:22: Mark Sluyer's wrote a post about this egentic future that completely deconstructs the current app model.

00:07:28: It is fascinating.

00:07:29: He has described and AI assistant he calls Francesca.

00:07:31: That literally lives right inside his whatsapp.

00:07:34: And level of context Francesca Has is what makes it agent-native.

00:07:38: Right,

00:07:38: she's connected to his aura ring and this calendar...his gym schedule.

00:07:42: Yeah,

00:07:42: Slews gets a voice note at nine p.m from Francesca saying like hey your sleep score was sixty eight last night.

00:07:49: you're readiness is seventy.

00:07:50: You carrying sleep debt start dimming the lights skip the late workout And get to bed by ten point thirty.

00:07:55: It

00:07:56: so proactive!

00:07:57: If he has conflict with his gym class The AI just asks if should reschedule.

00:08:02: He says, yes.

00:08:04: And the AI goes into his calendar finds a cancellation policy and moves to

00:08:08: class.".

00:08:09: The mechanism of interaction is what's revolutionary there?

00:08:12: Sluz uses an analogy to explain this.

00:08:14: he said agent-native software like giving video game controller.

00:08:18: I love that analogy.

00:08:19: Instead you pushing buttons opening different apps logging data make things happen.

00:08:24: the AI plays on your behalf.

00:08:27: Think about friction in current model.

00:08:29: Why would ever open app scroll past ads and manually type in a hundred fifty grams of grilled chicken breast.

00:08:38: Right, nobody wants to do that!

00:08:39: When you can just send a quick audio message to your AI agent saying what you ate?

00:08:44: And the AI does.

00:08:45: all the macro calculations and database logging in the background The

00:08:48: AI completely bypasses the app interface... ...the

00:08:50: interface dies.

00:08:52: Yeah it dies.

00:08:53: In an agent-driven ecosystem, the only consumer technologies that survive are hardware companies capturing sensory inputs like the Aura Ring or Apple Watch.

00:09:02: And foundational AI models processing data?

00:09:05: Exactly!

00:09:06: Everything in the middle, those dashboards and logging interfaces face an existential threat because they rely on human friction to generate ad revenue or subscription value.

00:09:16: But you know this isn't just a consumer app problem.

00:09:18: The enterprise side is facing a brutal squeeze too.

00:09:21: Oh definitely

00:09:22: Incbashalari pointed out that it should be a massive wake up call for anyone selling into payers.

00:09:30: They are committing a three billion dollar investment to AI.

00:09:34: Three billion, it's staggering!

00:09:37: And deploying twenty two thousand software engineers To rebuild their insurance infrastructure.

00:09:42: Yeah I mean what does this mean for the health tech professional listening?

00:09:46: Twenty two thousand engineers is not a pilot program.

00:09:50: It is a systemic rebuild.

00:09:51: now,

00:09:51: it's completely new foundation.

00:09:53: They already launched an AI companion named Avery.

00:09:55: that's live for millions of members handling benefits questions cost estimates claim status When and incumbent with that much capital decides to build their own agentic ai.

00:10:06: yeah they have a structural advantage.

00:10:08: They are the only underlying data lake

00:10:10: exactly.

00:10:11: so okay if I'm a health tech startup and my entire pitch is an AI workflow automation tool for payers.

00:10:18: Did My Entire Company just become an internal feature ticket for UnitedHealth's dev team?

00:10:22: That is the exact existential crisis Specialory is pointing out!

00:10:26: If you're a startup, you cannot compete on operational efficiency tools against an incumbent that has twenty-two thousand engineers and owns the data infrastructure.

00:10:34: You just can't...You

00:10:35: have to pivot!

00:10:36: You have to solve problems the incumbents either cannot solve or have zero financial incentive to solve.

00:10:41: like what?

00:10:41: Like building clinical decision support tools on the provider side Or patient advocacy in transparency layers That sit between The Patient And The Payor?

00:10:50: Right but there is massive irony here.

00:10:53: moving into our third theme While massive corporations like UnitedHealth are spending billions to build these perfectly locked down, regulated AI agents Patients are too impatient to wait.

00:11:04: They

00:11:04: really are, they're going rogue!

00:11:06: They're going completely rogue and it's creating this entirely unregulated shadow AI economy in healthcare.

00:11:13: Shadow AI is arguably the most complex governance challenge health systems face right now.

00:11:18: Oh for sure.

00:11:19: But Tina McMahon shared a statistic that quantifies the scale of this.

00:11:22: Two hundred thirty million people used chat GPT For Health questions recently.

00:11:27: two

00:11:27: hundred and thirty million.

00:11:28: And as McMahon points out trying to ban It Is A Fool's errand.

00:11:31: we all saw how corporate IT departments failed when they tried to ban employees from bringing their own smartphones understand complex blood test results, all completely outside the standard medical system.

00:11:51: And the regulatory framework is completely unprepared for this?

00:11:54: Yes I mean you cannot regulate a constantly evolving large language model The exact same way you regulate a static piece of medical hardware like an MRI machine.

00:12:04: Does it work?

00:12:04: No McMahon suggests we need to graduated software as a medical device or Sam D on ramp.

00:12:10: Wait how does an on-ramp actually work mechanically Like?

00:12:13: How do you regulate that?

00:12:14: So instead of a binary approved or unapproved status, it would be risk-based pathway.

00:12:19: It starts with lower risk administrative functions and gradually builds up to clinical diagnostics as the algorithm proves its reliability in real

00:12:26: world.

00:12:27: Okay that makes sense Right.

00:12:28: This will theoretically shift liability away from consumer navigating alone To shared model between developers' health systems

00:12:35: Which is desperately needed because current landscape actively harming people.

00:12:41: Gilles Friedman posted a deeply analytical critique of consumer chatbots that highlights the systemic failure in their design.

00:12:48: Yeah,

00:12:48: this was powerful post!

00:12:50: He pointed out fundamental difference between human clinician and an algorithm.

00:12:55: Clinicians have duty-of care... Chatbots only have terms service.

00:12:59: And when you rely on terms of service instead clinical judgment You create what Fredman calls reassurance

00:13:05: loop.

00:13:06: Yes The mechanics are fascinating Really dark.

00:13:11: Friedman points out that AI chatbots lack the therapeutic use of uncertainty,

00:13:16: which is such an important concept in medicine.

00:13:18: it Is when a human goes to a doctor and distress?

00:13:21: A good clinician will name The doubt set limits and actively resist the patient's impulse To endlessly obsess over a terrifying diagnosis.

00:13:29: right they'll tell the patients stop googling And just wait for the biopsy.

00:13:32: exactly that resistance as a form of medical care.

00:13:35: But a generalized AI doesn't resist.

00:13:37: Its core programming is designed to comply with user prompts and generate engagement.

00:13:42: Exactly, Freidman shared this anecdote of patient who spiraled into one hundred hour chat loop over ten days during cancer scare.

00:13:51: A hundred hours?

00:13:53: It's terrifying!

00:13:54: The AI never stopped him.

00:13:55: it never set up boundary.

00:13:56: It just endlessly fed his compulsion, validating his fears by providing more and more obscure possibilities.

00:14:03: Because it wants to answer the prompt?

00:14:05: Yes!

00:14:05: It completely dissolved the therapeutic boundary.

00:14:07: because this system is designed to provide answers not care.

00:14:12: It's a perfect example of how the language of patient empowerment can completely collapse.

00:14:17: A system that just mirrors a patients panic and never challenges distorted thinking isn't empowering them, it is trapping them And this connects directly to macro-level policy bottleneck we are seeing globally.

00:14:28: Governments simply do not know how to classify these tools.

00:14:34: The primary barrier isn't the underlying technology, it's fragmented governance.

00:14:47: Only eighteen percent of OECD countries even have a national oversight body for AI and health

00:14:52: Which leaves developers in really dangerous limbo.

00:14:56: Rudolf Wagner pointed out that Even Google Health AI lacks FDA and EU-MDR regulatory approval For its AI as software or medical device

00:15:06: Even

00:15:07: google.

00:15:08: Yet these underline tools are operating on market And Kiran Kumar Prabhu warned developers about the incredibly costly line between a simple wellness app and true Sam & D.

00:15:20: So what happens if you cross that line?

00:15:22: If guess your classification wrong, bypass proper trials get an FDA warning letter product launch dies!

00:15:29: So the algorithm is really just a blueprint.

00:15:31: Regulatory clearance Is The Building Permit.

00:15:33: You can't build a skyscraper without both, so regulatory clarity is no longer Just A Compliance Checkbox For Health Tech Startups.

00:15:40: It's The Ultimate Competitive Advantage.

00:15:42: Yes!

00:15:42: The companies that figure out how to safely navigate the SAMD classification While Proving Real World Clinical Benefit Those Are The Ones That Will Actually Scale.

00:15:51: The Code itself is becoming an easy part.

00:15:53: Navigating the Governance Is The True Differentiator.

00:15:56: All right, well we spent a lot of time talking about software algorithms and data layers.

00:16:01: For our final theme let's look at what happens when that software meets the physical

00:16:05: world?

00:16:05: The operating room!

00:16:06: Yes...the OR is undergoing a radical transformation.

00:16:11: Procedure rooms are turning hyper-digital robotic and fully integrated.

00:16:22: advanced imaging, and robotics fuse into single unified platforms.

00:16:27: Vinay Shetty and Jason Wright shared some updates regarding Medtronic's newly cleared stealth XIS system.

00:16:34: This is a wild piece of tech.

00:16:35: It really is it the first platform that unites AI based surgical planning real-time navigation And robotic assisted execution for spine and cranial surgery.

00:16:44: The integration mechanism here Is what matters because in the past these were separate steps with separate machines.

00:16:49: right now The AI helps automate the design of surgical plan calculating exactly where a screw should go based on patient's unique anatomy.

00:16:57: It maps neural pathways, then that data feeds directly into robotic arm that executes exact plan with sub-millimeter precision constantly adjusting real time navigation feedback during procedure.

00:17:11: And this isn't just theoretical future tech either.

00:17:14: Vicky Baker noted that Nuffield Health Exeter just completed their three-thousandth Mako robotic assisted joint replacement.

00:17:21: Three

00:17:21: thousand?

00:17:23: That's impressive!

00:17:24: They're proving at scale, the level of robotic precision leads to faster recovery times and less pain for patients.

00:17:31: It is a remarkable achievement in precision medicine.

00:17:33: Yeah, but you know as we unify these environments As we connect the imaging equipment to The AI planner and the planner To the robotic arm And all of it to the hospital's main network We expand the attack surface exponentially

00:17:47: Which brings us to the dark reality Of this digital transformation.

00:17:51: As our ORs become these brilliantly integrated environments, they become highly vulnerable targets.

00:17:56: Very mono!

00:17:57: They render them to.

00:17:57: the chowdery posted an analysis of a catastrophic cyber attack on Striker one of the biggest medical device manufacturers in the world.

00:18:05: And this wasn't normal attack.

00:18:07: No This wasn't a standard ransomware attack where someone steals patient data to sell on the dark web.

00:18:14: An Iran linked hacktivist group used wiper malware software designed purely to destroy data in brick systems,

00:18:21: just raw destruction.

00:18:22: they wiped over two hundred thousand devices across seventy nine countries.

00:18:26: it forced global office shutdowns and took entirely automated manufacturing systems offline.

00:18:32: It is a sobering reality check for the entire industry.

00:18:35: But wait, how does an attack spread to two hundred thousand discrete devices that quickly?

00:18:39: I mean hackers can't manually breach those many machines one by one.

00:18:43: They didn't have to.

00:18:44: According the analysis Choudhury shared, the vulnerability was an over-reliance on a unified environment.

00:18:50: without proper segmentation The attackers infiltrated the Microsoft and Mobile Device Management or MDM environment.

00:18:56: And MDM is basically a central dispatcher.

00:18:59: It's designed to push authorized software updates instantly To every enrolled phone tablet in terminal globally.

00:19:04: Oh

00:19:04: wow So the hackers just compromised the Central Trust of Dispatcher.

00:19:08: Once inside they weaponized that trust pushing the wiper malware out to the network as if it were a routine security update.

00:19:16: Exactly, because everything was deeply integrated to make the workflow seamless.

00:19:20: there was insufficient network segmentation to stop the spread

00:19:24: so that we're no firewalls between departments right

00:19:27: when the attackers triggered them where through the MDM it wiped every thing enrolled in this system and real time.

00:19:34: It highlights a critical flaw in how we're designing health tech infrastructure.

00:19:39: We build these perfectly integrated systems to make operations smoother, but without rigorous segmentation... ...we create single point of catastrophic failure.

00:19:47: So cyber security is no longer just an IT issue.

00:19:51: it's fundamental clinical safety requirement.

00:19:53: Absolutely

00:19:54: A compromised network doesn't mean lost emails anymore.

00:19:57: it means robotic surgical arm loses its navigation data mid-procedure.

00:20:02: HealthcareTech is now actively a battlefield in geopolitical cyberwarfare.

00:20:06: Which is really the underlying truth connecting all of the insights we've explored from these LinkedIn sources today.

00:20:12: Whether were talking about constraint architecture, bounding and AI in EHR?

00:20:17: The regulatory on-ramps needed for consumer algorithms or network segmentation in a robotic OR... Right!

00:20:23: ...the technology itself was incredibly powerful but the architecture around it security regulation clinical boundaries ultimately determines if heals or harms.

00:20:34: Well said We have covered some serious ground today, looking at how the industry is moving from theoretical capability to actual implementation.

00:20:43: We really have.

00:20:43: and as we wrap up this deep dive I want to leave you the listener with a final thought to ponder... Who actually owns the digital twin of your body that these systems are building?

00:21:05: That's

00:21:05: a great question.

00:21:06: If an agent native AI knows you're baseline health better than you do, but that agent is owned and operated by your insurer or massive tech conglomerate how to we ensure algorithm is fundamentally designed optimize for lifespan rather than their profit margin.

00:21:22: Man, that is definitely something to think about!

00:21:24: If you enjoyed this episode new episodes drop every two weeks.

00:21:27: also check out our other editions on ICT and tech insights defense tech cloud digital products & services artificial intelligence and sustainability in green ict.

00:21:37: Thanks so much for joining us everyone And a huge thank-you To all the professionals on LinkedIn whose insights fueled this conversation.

00:21:43: Yes Thank You.

00:21:44: Keep questioning keep building and don't forget to subscribe.

00:21:47: We will catch you On The Next Deep Dives.