Latest / HumanifAI HealthTech / Closing the Gap Between AI and Clinical Reality w/ Dr. Sanjay Juneja, Tempus AI
Transcript
- 0:00Francesca: Most AI in healthcare fails before it ever even reaches a patient, not because the technology can't do the job, but because nobody really translated it into something a clinician could actually use.
- 0:12Francesca: Dr. Sanjay Juneja has spent his career closing exactly that gap.
- 0:17Francesca: He's an oncologist who sees patients.
- 0:20Francesca: He's also building and shipping real AI into clinical workflows, and he's an author and an influencer who has a massive global following by making all this hard stuff really easy to understand.
- 0:32Francesca: And today, he's here to do the exact same stuff for us.
- 0:43Francesca: Now, welcome back to HumanifAI Health Tech, where we explore how AI is really rewriting the rules of healthcare and putting people right back at the center where they belong.
- 0:53Francesca: I'm your host, Francesca Vereb, head of marketing at Net Health.
- 0:56Francesca: Today's guest has a gift for that genuinely rare talent in this industry.
- 1:01Francesca: He can take something that is impossibly complex, like a cancer diagnosis, an AI governance framework, or even a broken data pipeline, and make it practical to the people who need it the most, which as a marketer, frankly confuses me to some extent, but also impresses me in equal measure Dr. Sanjay Juneja is a triple board certified hematologist and medical oncologist.
- 1:25Francesca: He's the VP of Clinical AI Operations at Tempus AI and the founder of TensorBlack, an AI education platform that's really focused on oncology.
- 1:36Francesca: He's also known as The Onc Doc on social media.
- 1:40Francesca: Dr. Juneja has built a global following of more than half a million people, and his podcast, Target Cancer, has been downloaded over four million times.
- 1:50Francesca: He's written a f- on Forbes, he's been credentialized through Harvard Medical School's inaugural AI healthcare program, and he was invited to the White House for a roundtable on healthcare leaders in social media.
- 2:03Francesca: And somehow, in between all of that, he still has a couple of hours to see his patients.
- 2:09Francesca: Dr.
- 2:09Francesca: Juneja, welcome to the HumanifAI Health Tech podcast.
- 2:14Francesca: So, Dr. Juneja, you've built your career around one skill: taking topics in healthcare that are genuinely really complicated and then making them much more human.
- 2:24Francesca: And you do this with patients, and you do it with your social media followers, and you do it in one of the most complex AI deployments in oncology.
- 2:33Francesca: So what I would love to know from you is what is the one, uh, sort of moment from the bedside that just sort of shapes how you think about AI, something that happens organically in the p- in the room with a patient?
- 2:45Francesca: Um, and what does that tell us about the gap between what AI builders are thinking about and thinking that clinicians need and what clinicians actually need?
- 2:54Dr. Sanjay Juneja: Yeah, it's a great question.
- 2:55Dr. Sanjay Juneja: Um, I think really the most important thing, uh, that made me realize we have kind of a, a technology, you know, insufficiency is really just the, quote-unquote, "ball drops" that happen when it relates to just a very important thing as it relates to the patient's care.
- 3:13Dr. Sanjay Juneja: So obviously, with cancer, uh, treatments, there are all kinds of side effects that can happen, especially with immune therapy, uh, and other things that can be a little nebulous.
- 3:21Dr. Sanjay Juneja: So you have to have a really interdisciplinary or collaborative approach, right?
- 3:25Dr. Sanjay Juneja: Meaning, like, I gotta send my patient to the GI oncologist or to the endocrinologist because of a pancreatic insufficien-insufficiency, either due to the therapy or because of the, you know, Whipple's procedure or whatever they had for their, for their cancer.
- 3:38Dr. Sanjay Juneja: But the problem is, a lot of times you'll have somebody admitted or, or in clinic, and you don't have the report of the endoscopy that was done, or you don't have the findings of, of, you know, the cystoscopy for their bladder cancer and where did it originate.
- 3:52Dr. Sanjay Juneja: And these little delays add up over time, and it can be very, uh, problematic, unfortunate.
- 3:58Dr. Sanjay Juneja: It causes moral injury, you know, on, on behalf of the physician and their team.
- 4:02Dr. Sanjay Juneja: Um, and that's where I first kind of saw this, uh, excitement or hope, which was, at the very least, we've got to use natural language processing and the ability that LLMs, uh, you know, frontier models Provide us to be able to query things that otherwise we're just running around chaotically and saying, "Can you fa- you know, get ahold of so-and-so and get that faxed over?
- 4:23Dr. Sanjay Juneja: And I need that right now because they drove an hour away because I'm in a community setting." I mean, these kind of problems are just all the time, every day.
- 4:31Dr. Sanjay Juneja: And, and when they lead to medical error, that's really unfortunate just because technologically, you know, if you could snap your fingers and red tape wasn't an issue, that shouldn't be happening.
- 4:40Dr. Sanjay Juneja: So that's, that's kind of what started this whole idea
- 4:44Francesca: Right.
- 4:44Francesca: So before-- and I'm g- we're gonna come back to LLMs 'cause I love that thread.
- 4:48Francesca: Um, but let's just pause for a second because you've done a ton of things in your career, and you've also got a startup, and you're, you know, you work b- a-across a number of different spectrums.
- 4:58Francesca: But you've really watched AI go from this, like, headline at a conference, right, to, um, really seeing what gets demoed and what actually gets deployed.
- 5:06Francesca: And in our field at Net Health, the clinicians that we serve, rehab therapists, wound care nurses, the occupational health specialists, they're all asking the same question.
- 5:16Francesca: Um, and if we're really honest, many of them lived through this whole notion of, hey, an EHR is gonna deliver all these amazing things, and it didn't really.
- 5:23Francesca: So what really separates AI implementations from actually work to those that actually get abandoned, and what's a question that clinicians should be asking vendors that most of them are not doing right now?
- 5:35Dr. Sanjay Juneja: Oof, that's a good, good question.
- 5:36Dr. Sanjay Juneja: Um, you know, the Arise Collaborative, which is, uh, uh, a collaboration between Harvard and Stanford, has a great, uh, status of clinical AI, uh, report, and they mentioned exactly what you said.
- 5:49Dr. Sanjay Juneja: Uh, you know, how there's just a lot of hype on what can be done, but then what actually was done in twenty twenty-five and early twenty twenty-six.
- 5:56Dr. Sanjay Juneja: And it seems like the adoption so far has been kind of on some lower-hanging fruit things.
- 6:02Dr. Sanjay Juneja: Um, obviously, I think the biggest quote unquote "success story" so far is this ambient, uh, listening, uh, ability, which, you know, isn't just a transcription of, of a recording.
- 6:14Dr. Sanjay Juneja: It's an actual process that happens, uh, behind, uh, at least depending on the ambient company, uh, in the background that can actually take this kind of clinical context and potentially use what's called a RAG system to be able to look at guidelines and make sure the-- uh, or even parts or features of, of a chart and pull it up as it's pertinent, uh, as you're transcribing these notes.
- 6:35Dr. Sanjay Juneja: So there is that element of it, um, that, that, that is pretty widely adopted.
- 6:41Dr. Sanjay Juneja: And then some other low-hanging fruit places are, are prior authorizations.
- 6:45Dr. Sanjay Juneja: This thought that, okay, if we have diagnoses, we have CPT codes, we have, you know, something listed somewhere on what, what qualifies this person to be able to, to get this either treatment or, you know, workup or study, et cetera, then can we build like an agentic system that's able to, to put these into, uh, into something tighter so that we don't have these delays in care like I mentioned previously.
- 7:07Dr. Sanjay Juneja: So these areas are, are of interest.
- 7:10Dr. Sanjay Juneja: Um, they require natural language processing, as mentioned.
- 7:14Dr. Sanjay Juneja: They require, um, you know, understanding of clinical context.
- 7:17Dr. Sanjay Juneja: But then if you take it to the next level that I think people are, are anxious or eager to get to that isn't you know, here quite yet.
- 7:25Dr. Sanjay Juneja: Uh, a big one for me is referrals and the ability to look and tell when something is appropriate or necessary and help coordinate or facilitate that referral.
- 7:35Dr. Sanjay Juneja: So whether it's, you know, what's called neoadjuvant therapy, meaning like the surgeon should be cued to say, "Hey, you might wanna send this to a medical oncologist because the data showed six months ago that getting treatment before curative surgery actually improved survival." Like, there's that level of referral, uh, uh, refer- nature.
- 7:51Dr. Sanjay Juneja: But then also when it comes to wound care, as you said, or if I'm talking and, and, and dictating or at least saying out loud, "Hey, you're having trouble, you know, getting up from the bed and, and going to the bathroom, and I'm gonna talk to social work and, and to a rehab center to see if you qualify for outpatient rehab versus inpatient." If I'm saying all of this, then it should be able to assess, oh, well, that's a functional deficit because they need to be able to get up and can't without assistance to go to the bathroom.
- 8:18Dr. Sanjay Juneja: That qualifies for XYZ, so I'm gonna send it for this level of, of intensity when it comes to rehab.
- 8:23Dr. Sanjay Juneja: These are all technically possible.
- 8:25Dr. Sanjay Juneja: Uh, I don't know that they're, you know, fluidly in the field yet, part due to-- in part due to interoperability issues because sometimes these facilities work on a different system than the one that you're, you know, saying all this out loud in, in the clinic.
- 8:37Dr. Sanjay Juneja: So there's all these kind of technical barriers, but, uh, but that's where people get excited because you could just imagine how much tighter and more accurately things are happening across the board, uh, because I know what I need to do, right?
- 8:50Dr. Sanjay Juneja: The therapist knows what, what, what it qualifies for.
- 8:52Dr. Sanjay Juneja: The patient's telling you exactly what is, and you're probing that.
- 8:56Dr. Sanjay Juneja: Why can't there be an agentic workflow that can then make sure all of those things are, are coordinated rather than now me shooting off a text to this or somebody getting me out of a room, et cetera, et cetera?
- 9:07Francesca: Yeah, I-- it's such an interesting point because I think, um, a lot of what you're also saying here is so dependent on this data, right, that underpins absolutely everything.
- 9:16Francesca: And, um, one of the articles or the pieces that you had written for Forbes, this, uh, AI healthcare paradox, I believe is the name of the article, you really made the case for the fact that AI is only as trustworthy, uh, worthy as the data that's behind it, and that the challenge here is that you've got health systems, like in the examples that you just described, that are really training models, but it's only on their patient populations, uh, that they have in their particular, uh, remit, right, or their field, their scope.
- 9:43Francesca: Um, and so when we think about that, is this really sort of a data quality argument, um, in principle?
- 9:49Francesca: Um, you also mentioned interoperability, right?
- 9:52Francesca: And so what-- when you actually start to look at that, is the foundation wrong, right, in from a clinical setting?
- 9:58Francesca: And what does this full picture demand from, say, a speciality EHR in our case to fix that, uh, whole paradox?
- 10:05Dr. Sanjay Juneja: Yeah, it's challenging because, you know, the, the mind doesn't know what the eyes haven't seen.
- 10:10Dr. Sanjay Juneja: So as, as these things have been propped up to be able to catalog data, you know, it wasn't with the thought, I don't believe, that, hey, one day we need to, you know, tokenize things and use this in a, in a way to have large language models and generative AI, right?
- 10:24Dr. Sanjay Juneja: So, you know, example is the health information exchanges.
- 10:27Dr. Sanjay Juneja: So this was a great, you know, thing that started, I, I'm told that it, it related to Katrina and, and when we had a, a huge hurricane in southern Louisiana, obviously, that wiped out a lot of healthcare records.
- 10:38Dr. Sanjay Juneja: We didn't know anything about some of our patients.
- 10:40Dr. Sanjay Juneja: And, um, for different reasons that all make sense, the thought was, "How can we make sure that there is, you know, healthcare data available that's not in a, in a paper form?" And then this concept of exchanges came out.
- 10:51Dr. Sanjay Juneja: But then the problem is the exchanges are-- or health information exchanges are, are dependent on the state of what kind of quality or system they have, right?
- 10:59Dr. Sanjay Juneja: So there's, that's, there's just that one technical element.
- 11:02Dr. Sanjay Juneja: Okay, do the e- health information exchanges have a hundred percent of the information?
- 11:07Dr. Sanjay Juneja: Is it cataloging it correctly?
- 11:09Dr. Sanjay Juneja: And then do I have, you know, a vendor or t- piece of technology that can harmonize and normalize this data to then have this kind of fertile stuff that I can prop up either my point solution or p- platform?
- 11:20Dr. Sanjay Juneja: So there's that aspect.
- 11:22Dr. Sanjay Juneja: But then there's this whole other aspect that, okay, let's just say we're in the same state.
- 11:25Dr. Sanjay Juneja: Let's say the exchange system's good or, or TEFCA, you know, is a, is a, is another strategy trying to make this stuff cleaner so that we can use it properly.
- 11:33Dr. Sanjay Juneja: But even just the access to the data now is a problem because some require these kind of direct integrations.
- 11:39Dr. Sanjay Juneja: There are plenty of, uh, systems in where, where, where I live and my, my wife and I practice, Baton Rouge, uh, Louisiana, where the The systems don't communicate to each other.
- 11:50Dr. Sanjay Juneja: So you can even have Epic in two different places, and they don't talk.
- 11:54Dr. Sanjay Juneja: Like, I can't necessarily see what's happening in the other place.
- 11:56Dr. Sanjay Juneja: I could maybe opt in for Care Everywhere, but that looks a little different because they might be PDFs, and that's now unstructured data, right?
- 12:03Dr. Sanjay Juneja: So it's not tabulated or discrete.
- 12:05Dr. Sanjay Juneja: So there, there's just so many, uh, areas where this kind of normalization and harmonization of data to just then start what we're talking about.
- 12:14Dr. Sanjay Juneja: Like, the things that we're getting excited about, uh, that can only start and happen when you have all of this, you know, uh, clean, clean and, and communicating to each other.
- 12:23Dr. Sanjay Juneja: And then to your point, there's this whole other issue that is good or bad when it comes to regional stuff, and where was this, you know, uh, AI solution tr- you know, trained?
- 12:34Dr. Sanjay Juneja: What was the data, uh, that it was trained on?
- 12:36Dr. Sanjay Juneja: Is it generalizable?
- 12:38Dr. Sanjay Juneja: Is it overfit?
- 12:39Dr. Sanjay Juneja: Is it underfit?
- 12:40Dr. Sanjay Juneja: And one example, I was speaking at a, at a mental health, uh, AI and cancer summit, and I was talking about how, you know, people in the same situation or with the same angst or depression about a certain diagnosis, uh, may need to hear things differently, obviously.
- 12:55Dr. Sanjay Juneja: Like, some need to hear very positively, some need to hear it's okay to give yourself a little, you know, uh, grace.
- 13:02Dr. Sanjay Juneja: So, so they may need to hear different things.
- 13:04Dr. Sanjay Juneja: Well, that depends also culturally,'cause I can guarantee you Southern Louisiana people are probably quite different than New York people, so I don't want a New York solution necessarily, uh, in Southern Louisiana.
- 13:14Dr. Sanjay Juneja: Um, so, so some of it's a good problem, some of it's a bad problem, but the biggest problem is to figure out which one makes sense and which one doesn't, and especially when there's no script.
- 13:23Dr. Sanjay Juneja: Because again, the mind doesn't know what the eyes haven't seen, and this is kind of new for all of us
- 13:28Francesca: It totally is, and I think it brings up such an important point about when you start talking about models, right?
- 13:33Francesca: So this conversation that's happening in AI healthcare right now, um, particularly w- from the LLM s- standpoint and these generative AI tools, they're all relatively new.
- 13:43Francesca: So naturally, you know, their track record is much thinner and, and folks like yourselves, you know, you're dealing with, um, sophisticated oncology systems that have run forms of AI for years, right?
- 13:54Francesca: You've had imaging models that have been reading scans.
- 13:56Francesca: You've had pathology models that have been grading slides, um, risk stratification, treatment planning optimization.
- 14:03Francesca: I mean, you name it.
- 14:04Francesca: Um, and these tools have all got FDA clearance, peer-reviewed outcome data, and years of clinical validation behind them.
- 14:11Francesca: But enter in the ChatGPT era, right, with Claude and Grok and all these other things, and you've got AI that's actually now coming in and having an impact on oncology today.
- 14:21Francesca: So what do you wish that more oncologists actually understood about that dynamic relative to, um, the data and just anything else that you're sort of experiencing in your current, uh, remit?
- 14:32Dr. Sanjay Juneja: Yeah, I think you nailed it.
- 14:33Dr. Sanjay Juneja: The first thing that still s- you know, somehow surprises, uh, people is that we've been using AI for a very long time, as you said.
- 14:41Dr. Sanjay Juneja: Uh, when it comes to machine learning and a lot of the, you know, some of the algorithms that we use for, uh, end-stage renal failure risk and, and, and several things.
- 14:49Dr. Sanjay Juneja: We've been using data, uh, and then it's been validated and, and able to, uh, be acted upon.
- 14:56Dr. Sanjay Juneja: Uh, that's number one.
- 14:57Dr. Sanjay Juneja: Number two, you know, it seems that there is an awareness of, you know- Confabulation or hallucinations, et cetera.
- 15:05Dr. Sanjay Juneja: But I worry that it's sometimes too polar or, or, or binary rather, uh, for some clinicians.
- 15:12Dr. Sanjay Juneja: Like it can happen and therefore, you know, until it doesn't happen, there's nothing to really consider right now.
- 15:19Dr. Sanjay Juneja: And I think that level of thinking may-- I, I worry that it may delay what could still be better than the status quo.
- 15:30Dr. Sanjay Juneja: So like that's the, that's the biggest thing at hand is what is the statistical, you know, percentage of time that we're doing right by patients in a certain circumstance?
- 15:40Dr. Sanjay Juneja: One of those examples is, you know, uh, screening for lung cancer.
- 15:43Dr. Sanjay Juneja: So it's like a lot of, uh, screening modalities when it comes to colonoscopies and mammograms and, and arguably, uh, Pap smears are, are well, you know, seventy percent, eighty percent-ish.
- 15:55Dr. Sanjay Juneja: But then lung cancer screening is below twenty percent.
- 15:57Dr. Sanjay Juneja: And we're talking something that saves, saves-- like has overall survival data, and it's been around for years and years and years and years and years.
- 16:04Dr. Sanjay Juneja: So if you're using generative AI to be able to flag, hey, this person is, is, you know, just by the book, the one that benefits from having an annual low dose CT scan.
- 16:14Dr. Sanjay Juneja: But say, you know, it's only catching that eighty-five percent of the time, it's still an error fifteen percent of the time, be it a false negative or false positive.
- 16:22Dr. Sanjay Juneja: But the status quo or the baseline right now is that we're, we're, we're only, you know, referring fifteen percent of the population.
- 16:29Dr. Sanjay Juneja: So, so that's where you kind of-- it takes this judgment call of like, is, is You know, perfection, the enemy of good, as s- as some say.
- 16:37Dr. Sanjay Juneja: So I think that's just really important is think about the status quo, number one.
- 16:40Dr. Sanjay Juneja: And number two is just to recognize that when it comes to AI tools, a lot of this is an iterative process.
- 16:47Dr. Sanjay Juneja: So it truly takes conditioning and then, uh, reiterating and reiterating to reach higher levels of success.
- 16:55Dr. Sanjay Juneja: Not least l- you know, at least the op- open models or at least things that, that have any kind of, uh, learning or feedback loops outside of something that's just, you know, purely closed.
- 17:06Dr. Sanjay Juneja: A lot of these require patience and feedback and monitoring, and you kind of develop them over time.
- 17:13Dr. Sanjay Juneja: They're not out of the box.
- 17:14Dr. Sanjay Juneja: They're not, you know, software in the sense that, you know, it's the CD and it is what it is, and it's gonna work how it works.
- 17:21Dr. Sanjay Juneja: And I think that's, you know, an important thing to understand, uh, because if we just delay that whole process of reiterating or considering or playing in the sandbox, as some say, then, then I worry we're delaying poten- delaying potential benefit to patients.
- 17:36Francesca: It, it's so, it's such a key point.
- 17:37Francesca: I mean, 'cause, you know, what you're basically saying there is this whole notion of, um, the promise of AI is absolutely real, but the liability questions are also a key point, right?
- 17:46Francesca: And, um, I'm really interested in your perspective on agentic AI, 'cause we feel that that's moving at least sort of in the spheres that we're in faster than anything we're seeing in health tech.
- 17:56Francesca: But in your view, where does agentic AI actually belong both in that clinical and that administrative workflow today?
- 18:02Francesca: And how do you think the human, um, needs to stay in the loop?
- 18:06Francesca: Will that always be the case?
- 18:07Francesca: What's your perspective?
- 18:09Dr. Sanjay Juneja: Oh, that's a tough question.
- 18:11Dr. Sanjay Juneja: Um, I mean, I think, number one, again, you know, tech and medicine are both humbling, humbling spaces, especially when it comes to, to what we've been able to do in AI today.
- 18:23Dr. Sanjay Juneja: I, I think everyone can agree that it seems that error is reduced when you have an agentic workflow, meaning you have specific detailed tasks that this agentic workspace, uh, puts together, right?
- 18:37Dr. Sanjay Juneja: So instead of having these like OpenC, you know, high level of data, uh, uh, systems where you just hope that it gets it right and retrieves what it needs to retrieve, you do this handoff of tasks.
- 18:49Dr. Sanjay Juneja: So you might have one that is able to look at the guidelines, another that looks at, you know, specifically the molecular data in the patient's chart, et cetera.
- 18:55Dr. Sanjay Juneja: So we know or, or believe that that is the way to reduce error and have improved accuracy.
- 19:02Dr. Sanjay Juneja: So I don't think agents are going away, for sure.
- 19:05Dr. Sanjay Juneja: I think they are showing promise, especially on the administrative side, as you were, uh, stating when it comes to-- And, you know, insurance companies beat, you know, clinicians on this.
- 19:16Dr. Sanjay Juneja: Like, they're-- they've been using these, you know, agentic workflows for prior authorizations and appropriateness on that same exact, you know, uh, concept.
- 19:24Dr. Sanjay Juneja: Like, what am I being told?
- 19:25Dr. Sanjay Juneja: What does the note say?
- 19:26Dr. Sanjay Juneja: Let me extract it.
- 19:27Dr. Sanjay Juneja: Let me compare it to what I have here, and let me, you know, shoot off a, a rejection or not.
- 19:32Dr. Sanjay Juneja: And those do not have human-in-the-loops, I'm told, uh, depending on the insurance company.
- 19:36Dr. Sanjay Juneja: Uh, because for them, you know, the consequence would be us then calling and then kind of working it out, and hopefully the, the right, uh, thing comes, comes to be a-a-at the end of the discussion.
- 19:48Dr. Sanjay Juneja: I think when it comes to clinical decision-making gaps or flagging someone or identifying someone for potentially a certain test or treatment, I do think we need human in the loops, not just because of, of, of the consequence.
- 20:02Dr. Sanjay Juneja: Like, you know, somebody might say, "Oh, wow, I have this mutation that, that can really give me a better survival if I-- compared to if I didn't have that mutation with this cancer." The problem is too, though, the psy- the psychology of it, because now you might have caused fa- caused false excitement for a patient, false excitement for, for their family members, and then kind of pulled it back and said, "Never mind, we have to do standard chemo.
- 20:23Dr. Sanjay Juneja: Targeted therapy isn't an option." So there's consequences even outside of medical error that are psychological in cancer that I think we have to be very sensitive to.
- 20:32Dr. Sanjay Juneja: You know, am I, am I optimistic that, that maybe HIT, uh, humans in the loops aren't necessary, uh, as much in the future?
- 20:41Dr. Sanjay Juneja: Yes, I hope there is a day when automation, uh, is that good because there is a very clear deficiency in cancer that, that no one has an answer to.
- 20:50Dr. Sanjay Juneja: You can't make, you know, a new oncologist without fourteen years of training, all the staff members and people involved.
- 20:56Dr. Sanjay Juneja: There is a growing gap that we know because in twenty-thirty we're gonna have this like doubling of cancer patients, uh, without the same in-increase in staffing.
- 21:05Dr. Sanjay Juneja: So there has to be something.
- 21:07Dr. Sanjay Juneja: So the hope is, you know, that, that there's automation in certain points or places because otherwise there is no solution
- 21:13Francesca: Completely.
- 21:14Francesca: And I love the fact that you said, um, you know, it's both healthcare and IT or healthc- um, you know, health tech are very humbling professions.
- 21:23Francesca: Um, one thing I wanted to ask you before we dive into, like, the staffing workload and, and that sort of burden is, um, talking about this pilot sort of purgatory, right?
- 21:32Francesca: We've got a lot of folks, especially in the healthcare space, lots of startups, lots of folks that are good at running a bunch of, of pilots, but very few are actually good at getting to that next step.
- 21:41Francesca: Now, you've actually navigated this, right?
- 21:43Francesca: And you've done it successfully.
- 21:44Francesca: I think you've actually went ahead and sold your company, so terrific.
- 21:47Francesca: I think that was-- that's all great.
- 21:49Francesca: Um, but as a founder and as a person running also the clinical AI operations, what really has to be true to make that transition from this, hey, promise, to the actual operational infrastructure?
- 22:01Francesca: Um, what's the most difficult part of that transition that never really gets discussed?
- 22:05Dr. Sanjay Juneja: Honestly, you know, and this is just my opinion, I think, I think one of the biggest bottlenecks is just having the bandwidth on both, uh, parts, the, like, the vendor or the service or the tech company, as well as the end user.
- 22:22Dr. Sanjay Juneja: Having the bandwidth to be able to discuss enough and frequently enough, uh, the different steps and processes to make this a, a, a more scalable solution.
- 22:32Dr. Sanjay Juneja: So the, the thing that hinges for a pilot, I think, is, like, having your champion, like, at, at the, at the clinical, uh, uh, setting, if it's a clinical tool, or at the institution.
- 22:43Dr. Sanjay Juneja: Having that champion that can ensure that all of the different factors, again, interdisciplinary, but this time in a tech way, your IT person, your informaticist, your or innovation officer, the informatics team on the vendor or health tech side, uh, having a, a, a steadfast engineering team to be able to adapt and, and to fix problems and find solutions in, uh, a quick amount of time so that clinicians don't get frustrated.
- 23:09Dr. Sanjay Juneja: You have to have all of this in place, and that's the purpose of a pilot, is how much can I do and how much tinkering and tightening and loosening and, you know, waiting and all of this other stuff to be able to make this as semi-autonomous as possible and as reliable as possible.
- 23:26Dr. Sanjay Juneja: And unfortunately, at least as it relates to healthcare and, and cancer care, there just isn't a lot of bandwidth, like, that, like, especially when you talk about now community settings, uh, or places that are feeding, uh, you know, rural populations, et cetera.
- 23:39Dr. Sanjay Juneja: So I think the success lives and dies by the amount of involvement in that pilot, and I think pilots fail when there isn't enough communication back and forth, and it's more just a tryout and a feel, and I'm like, "Eh, dissatisfied customer, I'm gonna return, you know, return the thing," like, as if it's a pair of shoes.
- 23:57Dr. Sanjay Juneja: So, like, that-- it's, it's, it's very different in that manner It takes also resources in the way of, of finances because you might need a, a, you know, an FTE, a full-time, uh, employee for, for this purpose.
- 24:09Dr. Sanjay Juneja: Uh, and again, just a lot of beta testing.
- 24:12Dr. Sanjay Juneja: And I think that's why, I think it's a resource deficiency to be able to then make it, uh, scalable and have everything done right, uh, a- at a robust enough level to be able to scale it just because of the resources
- 24:25Francesca: So what should a healthcare system be asking before they turn agents loose in this particular, in, in, in sort of your perspective there?
- 24:32Francesca: So you go from pilot, then you start to scale.
- 24:35Francesca: You were talking about the agentic thing.
- 24:37Francesca: What should they be thinking about that?
- 24:39Francesca: And I'm also curious around governance, right?
- 24:42Francesca: Because there should be questions that are in place ahead of the pilot.
- 24:46Francesca: How do you think about that, and how do you think about that driving what agents they do deploy in, um, in practicality?
- 24:53Dr. Sanjay Juneja: Oh, boy.
- 24:54Dr. Sanjay Juneja: Yeah.
- 24:54Dr. Sanjay Juneja: I mean, this one is-- this one's huge, and, and I've noticed, you know, it actually even varies kind of quite considerably just depending on not just the region you're at or the size of the institution you're at or the, or the, or the number of systems, but even just, like, personality or just the flavor of the, the person you're working with.
- 25:12Dr. Sanjay Juneja: Um, but you really wanna have a very, uh, steadfast idea of what does the governance and monitoring plan look like before you prop up the agentic workflow.
- 25:26Dr. Sanjay Juneja: If you have this mentality that I'm gonna prop up these tools, models or agents, and then we're gonna, like, we're gonna, you know, be iterative and back and forth and figure it out as we go, that is a lot of times the experts I've talked to, especially at larger, uh, health centers, that's where things fail.
- 25:41Dr. Sanjay Juneja: You need to have all of that identified beforehand.
- 25:44Dr. Sanjay Juneja: Who is the person that this first goes into the hands of?
- 25:47Dr. Sanjay Juneja: Who is the person on your side that's actually auditing and taking X percentage of these things to ensure accuracy?
- 25:54Dr. Sanjay Juneja: Who's the person on the, on the health tech side or the, or the vendor side that's gonna be auditing?
- 25:58Dr. Sanjay Juneja: What is the frequency of audit?
- 25:59Dr. Sanjay Juneja: And then that first, you know, point of contact, what are the different pathways they escalate to?
- 26:04Dr. Sanjay Juneja: Who's the point person of the IT, uh, uh, team that's gonna be able to, uh, to be flagged?
- 26:09Dr. Sanjay Juneja: Are they gonna be available?
- 26:10Dr. Sanjay Juneja: Who's the person that tells in the clinician or the clinic team, "Hey, we need to pause all of this because it's starting to either drift or, or be erroneous because we found that it's mistaking this one term that we use here that n- apparently no one else has used-- uses with their vernacular elsewhere." Whatever those triggers are, you have to have that entire process mapped out beforehand.
- 26:29Dr. Sanjay Juneja: And the, uh-- I'm told that is especially the only way that something succeeds.
- 26:34Dr. Sanjay Juneja: That doesn't mean, doesn't mean everything succeeds.
- 26:36Dr. Sanjay Juneja: But these are the questions that has to be, uh, uh, addressed, uh, that have to be addressed by the team at the institution that the health tech team cannot give you.
- 26:45Dr. Sanjay Juneja: And now, once you've identified those things… And you can build those things even with the health tech team, where you can say, "How often do I have to audit?
- 26:52Dr. Sanjay Juneja: What do I need to audit?" Um, you know, "Where is my data coming from?" The second piece that I worry about is dependability, because now a lot of these, these solutions sometimes have multiple Uh, chains in the link, so to speak, or links in the chain.
- 27:07Dr. Sanjay Juneja: So you have, you know, potentially if it's an HIE solution, you have a vendor that's just cleansing the, the health information exchange data, and then you have someone, or if it's a FHIR connection that's delivering it, you know, based on, you know, a certain cloud, uh, of storage that it, is the cloud the same at the institution as it is at the, at the vendor?
- 27:24Dr. Sanjay Juneja: There's just so many different places that these things could go awry.
- 27:28Dr. Sanjay Juneja: Uh, having all of those identified and ensuring that there's a solution should this have to be paused is also very important because otherwise it could be catastrophic.
- 27:37Francesca: So really your advice here is, you know, ask those difficult questions upfront.
- 27:42Francesca: Um, really focus on things like the accountability, the escalation, it- failure, risk, any of those things.
- 27:48Francesca: And, uh, I guess that's sort of the advice we heard, which is, you know, m- measure twice, cut once kind of thing, right?
- 27:54Francesca: So- Yes,
- 27:54Dr. Sanjay Juneja: exactly.
- 27:55Francesca: Yeah, yeah.
- 27:56Francesca: But underneath it all, I think what you were also sort of, um, alluding to is this notion of sort of a staffing crisis that is sort of happening under every conversation, um, that we speak about w- with respect to AI.
- 28:07Francesca: And just in specialty care in particular, we're seeing things like, um, our wound care nurses, right, that are just covering caseloads that they were never trained for, as an example.
- 28:17Francesca: We've got physical therapists that are spending every lunch hour or pajama time taking, you know, documentation home.
- 28:23Francesca: Um, these are not marginal roles, right?
- 28:26Francesca: Their interventions actually prevent hospitalizations, but they're burning out and they're being overlooked.
- 28:31Francesca: And you were talking earlier about AI as sort of a workforce solution.
- 28:35Francesca: Um, when you think about it that way, do you actually think it's something that keeps skilled cl- clinicians in the profession?
- 28:42Francesca: And what is the opportunity that isn't getting enough attention when you're advising, um, other folks on what, you know, sort of use case is going to deliver the first relief for these workers?
- 28:53Dr. Sanjay Juneja: Yeah, I mean, you know, at least for the short to mid, you know, mid, uh, maybe even long-term future, but at least let's just say at least short, there is just so much stuff that is not at the top of license.
- 29:06Dr. Sanjay Juneja: That's, that's the biggest problem that I think is most immediately solvable.
- 29:12Dr. Sanjay Juneja: When it comes to this, you know, wound care, uh, uh, team or it comes to the clinician or whoever these, you know, these different, uh, uh, parts of the team that you mentioned, uh, whoever they are, there is so much time going into coordination and just communication of handoff and retrieval of certain pieces of information or scheduling or cancellations or making sure the data is there or the past history, which is always relevant for wound care or the antibiotic history.
- 29:38Dr. Sanjay Juneja: So much of this that is below the level of specialization for that specific person you're talking about, that is immediately and abundantly available to be addressed.
- 29:50Dr. Sanjay Juneja: So that, that to me is, is really where we need to be honing on.
- 29:54Dr. Sanjay Juneja: Because maybe in some future great problem to have, things are so good that we don't need that top of license specialization.
- 30:00Dr. Sanjay Juneja: I just don't see that, like, anytime soon because there's just so much to be had before then.
- 30:06Dr. Sanjay Juneja: So, you know, I think the, the argument on the other side is a little bit more concerning, which is, you know, are clinicians or are any specialists of the healthcare provider team as, as the labor of a lot of these administrative and more rote things, uh, as they're taken care of, is that-- does that mean we're just gonna have to see more patients?
- 30:26Dr. Sanjay Juneja: And I think that's, I think that's an ethical question, which is just like, you know, there are more patients that need to be seen.
- 30:31Dr. Sanjay Juneja: You can't make an oncologist except fourteen years, so we're just gonna have to figure that out as, as humanity.
- 30:36Dr. Sanjay Juneja: Uh, I think that, that is maybe a more of an immediate-ish concern, uh, that, that is realistic, uh, as far as an oncologist, just the numbers that you're able to now have, uh, which again, in a weird way is kind of a good problem to have rather than not needing us at all.
- 30:53Dr. Sanjay Juneja: And I, you know, I think just ASCO this past, you know, past month really showed that.
- 30:58Dr. Sanjay Juneja: I was participating and moderating a bunch of specialty, uh, uh, panels for lung cancer and, and hemoc, uh, malignant heme cancer and all this stuff.
- 31:07Dr. Sanjay Juneja: And the complexity of this data and debates between the leading specialists in the country, I think just really underscored.
- 31:13Dr. Sanjay Juneja: I made a post about it, like how that's not going anywhere.
- 31:16Dr. Sanjay Juneja: I mean, you can't make an agentic workflow when even the top specialist of understanding that understand things today, uh, to these complicated diseases, uh, have different experiences and, and observations and opinions.
- 31:28Dr. Sanjay Juneja: So for that reason, I think, I think we'll be able to practice that top of license much better and more purely, which I do hope will help, uh, fulfillment and reduce burnout.
- 31:37Francesca: Yeah, it's, uh, you know, again, it's, it's a very human, um, dynamic, right?
- 31:42Francesca: It's interesting that we are using technology to help us get more human.
- 31:46Francesca: Um, and sort of staying on that thread, this whole notion specifically in your field, you've spent years really building trust, right?
- 31:53Francesca: Whether it's with the case- the cancer patients, as you mentioned earlier, who are just scared, or families who don't really fully understand or absorb, you know, what they're being told, um, or clinicians that are just being burned out by the technology, right?
- 32:06Francesca: Uh, that whole confidence y- factor that you were talking about earlier, um, has really dropped, right?
- 32:12Francesca: It was sort of 70% in, in 2020.
- 32:15Francesca: Now it's down to 40% confidence in about 2025.
- 32:18Francesca: I mean, that's big.
- 32:19Francesca: Those are big drops, right?
- 32:20Francesca: And with our clients are telling us that clinicians are really skeptical, uh, excuse me, skeptical about another tech solution.
- 32:28Francesca: Um, and, and we say that specifically in our space 'cause we play a lot with sort of the EHRs, and we know the EHR really has had some impact on the confidence within the healthcare space.
- 32:39Francesca: So what's the whole difference that sort of has to happen now for clinicians to actually start to trust this technology and trust the governance, um, so that it actually starts to become a solution to contribute, like as you mentioned earlier?
- 32:53Dr. Sanjay Juneja: Yeah.
- 32:53Dr. Sanjay Juneja: I think, and this is purely just my opinion, but I, I worry that The, there's just been so many point solutions, you know.
- 33:02Dr. Sanjay Juneja: And by point solutions, I mean something that just has a, a, a very specific, uh, honed-in solution.
- 33:09Dr. Sanjay Juneja: And then you have, like, a dozen, uh, you know, different variations of, of this point solution, and some are better than others, and some are not better than others.
- 33:18Dr. Sanjay Juneja: And then having an experience with just one, say, was, you know, too many points of potential failure and, and it was breached and there was no plan, now it's gonna make you really ha- have kind of guarded excitement about that specific point solution in general.
- 33:34Dr. Sanjay Juneja: If the prior auth thing was terrible, you know, and nothing was actually getting authorized, or the, the bed scheduling, you know, was, was canceling appointments when it shouldn't.
- 33:42Dr. Sanjay Juneja: I mean, pick your example.
- 33:43Dr. Sanjay Juneja: I think my opinion part of, that I disclaimed was I think this whole concept of platforms and the integrity of a platform is gonna be, be really key and essential.
- 33:55Dr. Sanjay Juneja: And there are platforms that can take care of or hedge a lot of those problems that are amenable to multiple vendors and point solutions.
- 34:04Dr. Sanjay Juneja: Um, and, and I think a focus on that platform aspect is the hedge against potential investment disappointment and failure because the integrity of that can, can help coordinate and be the, be the interstitial, you know, uh, uh, process or network by which the success of these more focal points of solution can, can live and be viable.
- 34:30Francesca: Yeah.
- 34:30Francesca: It's a, it's a fascinating topic, and I, I think it's one that we have a lot of work to still do to earn that, that trust, right?
- 34:37Francesca: So, um, but super exciting times and, uh, definitely, definitely interesting to see what happens.
- 34:43Francesca: And actually that brings me to a good challenge question.
- 34:46Francesca: Do you think 2026 is, um, really a bubble or, or a correction?
- 34:52Francesca: Um, we've read a lot of stuff in the news, right?
- 34:54Francesca: We're hearing analysts talking about 2026 as sort of this mass extinction event for AI healthcare startups.
- 35:01Francesca: Um, essentially this moment when that promise to prove sort of really exposes lots of fragile economics underneath.
- 35:07Francesca: And it started also debate a- about, a lot around what actually gets funded.
- 35:11Francesca: Um, and we're seeing a lot of AI healthcare companies that are going public on a promise, but sometimes the proof is still being assembled, right?
- 35:19Francesca: So, and some say, "Hey, this is gonna be more of a correction because until the outcomes data is really real, it's, it's not really something that we can bank on." Now, you've watched this from inside a company that's had to make that exact transition.
- 35:31Francesca: So in your view, what do you see actually separates the companies from what those that survive and will continue to survive past 2026 and those that are just, you know, "Hey, it's a bad bet," and may not survive the next 18 months?
- 35:44Francesca: And the point I really want to understand is for vendors that are considering, um, AI s- you know, specialists or, or people that… You know, technology companies, what should they be demanding from those vendors that may not survive the next 18 months?
- 35:59Dr. Sanjay Juneja: Oh, yeah.
- 36:00Dr. Sanjay Juneja: I mean, I, if I may, I'm gonna say it's, it's kind of both.
- 36:03Dr. Sanjay Juneja: Like, there is an exting- extinction that is happening, sadly.
- 36:08Dr. Sanjay Juneja: And I say sadly because I've met many good intention people, and CEOs, and founders of amazing, uh, you know, solutions that just have put everything in, and I mean resource-wise and, and, and heart-wise.
- 36:21Dr. Sanjay Juneja: Um, I think the reason they're failing is there is a resource exhaustion, because I don't know that overnight there's gonna be this disappearance of where you don't need upfront resources that eventually, you know, may make the ROI net positive, um, uh, at least for the vendor, uh, and hopefully definitely for the institution, if nothing else, just in, in the quality of patients' lives.
- 36:45Dr. Sanjay Juneja: My point in saying all that is, and part of what, what the n- my newer startup, uh, TensorBlack is trying to do is, how do we bring people, all of us, you, this podcast, how do we bring those good meaning people together and finding opportunities of collaboration and, and, you know, I don't wanna call it M&A, but, but, but really, uh, not trying to invent everything on their own?
- 37:08Dr. Sanjay Juneja: There is a lot of people-- There are a lot of people that are doing amazing things that are all different pieces and parts of, of this entire, you know, transformation that could be, and you're starting to see that now.
- 37:20Dr. Sanjay Juneja: So you've seen the companies that have failed, but you're starting to see companies come together a little quicker.
- 37:25Dr. Sanjay Juneja: And, and on the flip side, I do see, uh, companies saying, "Oh, well, we can build that," or, "We have a fantastic, you know, group of engineers, and we can do that." But then the time lost in building it and doing that entire validation process and auditing process and stuff could be the Achilles heel that kind of prevented this, this solution from being viable and scalable in another aspect.
- 37:45Dr. Sanjay Juneja: If somebody's already done that, I think there needs to be an openness for the chance at viability compared to any other, you know, place or point in the history of, of, of either, uh, health tech or, or just even capitalistic kind of pursuit, right?
- 38:01Dr. Sanjay Juneja: Like SaaS, a SaaS system is very different and may not make sense or need to have people come together and acquire because a SaaS is a SaaS.
- 38:10Dr. Sanjay Juneja: But, but there's so many different elements to these solutions, these platforms, and to, uh, enriching and harmonizing the data, uh, complex data at that when it comes to medicine, that collaborative efforts need to be, um, uh, almost like a, a section of the company is what I would say.
- 38:28Dr. Sanjay Juneja: Like, just, just have somebody looking to see who's doing similar stuff and where does it make sense to do this together to kind of aggregate the resources, uh, delay the potential exhaustion of those resources, and that at a point where the solution can be demonstrated before that happens
- 38:47Francesca: Yeah, it's a, it's a fascinating journey, and I'm really excited about what's gonna happen next.
- 38:52Francesca: I am gonna ask you a question later today what your vision for 20-- uh, 30 is.
- 38:56Francesca: So I'll park that so that you have a little bit of time to think about it.
- 38:59Francesca: But in the meantime, one thing that is really fascinating me, and we're seeing a lot of headlines in the news, right, about, like, the, the next generation, how are they handling AI, no future jobs.
- 39:09Francesca: And especially as a mother of two young adults, I think a lot about what this means.
- 39:14Francesca: Um, you know, you've got healthcare systems, and you've got your clinicians that have been trained before Ambien scribes, you know, existed, and they really built, um, their diagnostic instincts and skills under pressure.
- 39:27Francesca: But this next generation is gonna train with AI right from day one, and they may not have the full practical experience and know when to be able to push back on AI and, you know, trust it or when to say, "Hey, this doesn't feel right.
- 39:40Francesca: I've seen more of this in my own experience." So what does de-skilling risk actually look like in practice, and is it as serious as some of your podcasts have actually suggested?
- 39:49Francesca: And what do you recommend for the next generations of, of clinicians as they're thinking about, you know, leveraging AI under these types of circumstances?
- 39:58Dr. Sanjay Juneja: You know, you bring up a great question that I think, you know, I should disclaim that I, I studied philosophy in undergrad, so I, I might be a little more radical than other, than others on, on this, uh, opinion.
- 40:10Dr. Sanjay Juneja: But I, I, I was-- I have a 10-year-old son, and I was, uh, telling him that he needs to do a project to us like, you know, it wasn't really punishment, but like he's gonna present to our, to our family about, about why, uh, vegetables are healthy.
- 40:24Dr. Sanjay Juneja: Like why do we need vegetables, right?
- 40:25Dr. Sanjay Juneja: And, um, you know, I-- he's like,"Okay, well give me your laptop." And I gave him the laptop and, and this is the first time he's, he's putting together a, a PowerPoint.
- 40:35Dr. Sanjay Juneja: And I just immediately had the debate because I was gonna try to explain to him, you know, Google and how you have to kind of authenticate or vet the source that you're looking at.
- 40:44Dr. Sanjay Juneja: Like is it like actual literature or is it just an article on literature?
- 40:48Dr. Sanjay Juneja: You know, trying to give my 10-year-old these concepts.
- 40:50Dr. Sanjay Juneja: And then immediately I was like- Or a TikTok
- 40:51Francesca: video, right?
- 40:52Francesca: Exactly.
- 40:52Francesca: Or a
- 40:52Dr. Sanjay Juneja: TikTok
- 40:53Francesca: video.
- 40:54Dr. Sanjay Juneja: Exactly.
- 40:54Dr. Sanjay Juneja: It could be anything.
- 40:55Dr. Sanjay Juneja: And then I was like, but then, you know, there's this whole aspect of experience that like you just can't be taught.
- 41:01Dr. Sanjay Juneja: So like how is my 10-year-old gonna know what that vetting process looks like when he's actually like exploring the internet for the first time other than YouTube?
- 41:07Dr. Sanjay Juneja: And then I was like, or we could just ask ChatGPT, and I'll tell you, we ended up going with the ChatGPT route, right?
- 41:13Dr. Sanjay Juneja: And then, you know, he got good at, you know, verbal pilling, which is like just start, you know, asking out loud because he can't type well and, and he had this great discussion.
- 41:20Dr. Sanjay Juneja: So in one aspect he did really sharpen a skill that a lot of adults, you know, don't have, which is, uh, verbal pilling, uh, or voice pilling, uh, AI where, where you're n- doing less typing and actually having this conversational process to be able to come to a conclusion.
- 41:36Dr. Sanjay Juneja: It's very effective I'm, I'm told even for coders and engineers.
- 41:39Dr. Sanjay Juneja: But, uh, but what he didn't do was that, that grind that, you know, cholinergic dopaminergic process that behaviorally has been shown to teach especially someone like, like a physician where you have to be very fastidious and go really deep just to be able to get a minimal amount of, uh, reward.
- 41:58Dr. Sanjay Juneja: Sometimes just one answer like you'll have to study for an hour or two.
- 42:01Dr. Sanjay Juneja: If that goes away, my bigger concern is really just the hardiness and the durability of individuals in the future to be able to have, you know, the patience, the wherewithal, the, the The length of frustration and failure that it requires to be able to come to a conclusion.
- 42:25Dr. Sanjay Juneja: Now, the counterargument could be, well, we're not gonna live in a world where that's even an issue.
- 42:31Dr. Sanjay Juneja: But, but, you know, I need to remind people, you know, even with all of this stuff that's been happening, you know, uh, uh, as it relates to worldwide, you know, tensions and, and wars, there are a lot of bottlenecks that can occur to make this stuff go away, including access to minerals, uh, and key elements that, that are required by these AI systems and these GPUs and everything.
- 42:52Dr. Sanjay Juneja: I had somebody on my podcast where they were a legal AI person, and they just outlined so many bottlenecks of why something could get on pause, like even a, you know, potential frontier m- uh, model if it related to geopolitical tension.
- 43:06Dr. Sanjay Juneja: And so what happens then if something-- that retrieval process was to pause?
- 43:12Dr. Sanjay Juneja: Do we have the durability, fastid-fastidiousness, and that kind of learned training of difficulty, uh, of not getting A to B, but rather A to Y to get the same answer?
- 43:25Dr. Sanjay Juneja: That's my big, big concern, and I do not have an answer to that other than I still try and make my kids, unlike the example I just gave because that, that calls me out, to try to just h-have them used to this level of, of failure to reward ratio.
- 43:39Dr. Sanjay Juneja: Um, but it's a serious concern that I-- that does truly worry me.
- 43:44Francesca: Yeah, it's a fascinating social experiment, and I almost to some extent wish I could parachute, you know, 50 years into the future to sort of see what's gonna happen and then come back.
- 43:53Francesca: But, um, yeah, we'll just have to continue to wait and see.
- 43:56Francesca: But you brought up something fascinating when you, you referenced the ChatGPT, so I can't resist asking you this because as the one doc, the, sorry, the onc doc, you really talk and es- establish yourself as a fascinating and credible voice for patients, right?
- 44:10Francesca: Navigating this whole landscape of potential bad information.
- 44:14Francesca: And you mentioned ChatGPT.
- 44:16Francesca: Now, I know I've done this myself with my daughter.
- 44:18Francesca: She had a healthcare, a scare, and what I did is also used all the AI models, right, ChatGPT, Claude, et cetera, to try to run clini- clinical notes through that.
- 44:27Francesca: And I think your medical doctors are doing the same to some ex- some extent.
- 44:31Francesca: But what does that do when you start to think about your role fighting health misinformation?
- 44:37Francesca: What worries you most about sort of AI spreading some of that, right?
- 44:41Francesca: And what does this whole notion of clinical-grade AI need to include to reduce that?
- 44:46Dr. Sanjay Juneja: Yeah.
- 44:46Dr. Sanjay Juneja: It's, it's an important question, and, and, you know, just on, on the spot, I, I hope that there's almost these profiles of these frontier LLMs that are similar to what I've prompted mine to be.
- 44:58Dr. Sanjay Juneja: Like, you know, in my default setting and, you know, I just say you need to, you need to only go to, you know, literature sources, cite every single one where you make any statement on a dosage, on an indication, on a guideline, reference the guideline.
- 45:10Dr. Sanjay Juneja: You need to double-check your answers, and then, and then I'll basically take all that, and I have my own agentic workflow that then just cross, cross-challenges everything that was said through a different frontier LLM.
- 45:20Dr. Sanjay Juneja: So that thing that I built, you know, I know most people haven't, but I don't think we're far away at all to be able to just opting in, "I want the physician version of this," that's kind of doing a similar thing.
- 45:30Dr. Sanjay Juneja: But to your point, until that day comes, there is a, a plethora of, of information.
- 45:35Dr. Sanjay Juneja: I mean, we know how much Reddit is used by these frontier LLMs and these other sources that could just, like, very matter of fact, you know, put it in front of you and perpetuate these kind of, uh, misunderstandings.
- 45:46Dr. Sanjay Juneja: And I think, unfortunately, you just need to have this, like, you know, discernment that only comes with experience, like I said my 10-year-old didn't have.
- 45:54Dr. Sanjay Juneja: Like, how could he be discerning if he doesn't have, uh, the experience to be able to know what does or doesn't sound right?
- 46:01Dr. Sanjay Juneja: So that's why I also think specialists, uh, and this human type of, type of license thing is not going anywhere.
- 46:07Dr. Sanjay Juneja: Like, it just, it just can't be.
- 46:09Dr. Sanjay Juneja: I do think rather than maybe a human in the loop, there could be this almost, like, you know, specialty guild that one day sits in front of the stuff that's coming out, uh, to be able to vet it from a medical aspect because of those 14 years of training or whatever, and potentially in mass volume.
- 46:26Dr. Sanjay Juneja: I don't know.
- 46:26Dr. Sanjay Juneja: Or you have this, like, these agents that are reflective of that individual, and then now you're pooling them together so that you can actually scale them.
- 46:33Dr. Sanjay Juneja: But I, I don't think human in the loop is going anywhere, and I think it's necessary even- Because of the unknown and the unknowing on how we're gonna receive these things.
- 46:43Dr. Sanjay Juneja: Because one of the biggest paradoxes that I kind of live in when it comes to tech and AI and medical solutions is how m- how much of a lack there is of medical and health professionals with these tech companies.
- 46:56Dr. Sanjay Juneja: Like, I will ask them like, "Oh, you have an oncology solution.
- 46:58Dr. Sanjay Juneja: Like, you know, who's your CMO?
- 46:59Dr. Sanjay Juneja: How many oncologists are you on your team with?" Like, "We don't have any." I'm like, "Well then, how in the world," you know, respectfully.
- 47:05Dr. Sanjay Juneja: Like, and I-- and it's not, it's not, you know, bad meaning.
- 47:08Dr. Sanjay Juneja: It's just like the mind doesn't know what the eyes haven't seen.
- 47:10Dr. Sanjay Juneja: They're not aware of all of the, oh, but this, this, this, this, this, this, all of these different, uh, uh, nuances that can, can, uh, be in place.
- 47:20Dr. Sanjay Juneja: That sometimes I've seen like anxiety over the founder's face.
- 47:23Dr. Sanjay Juneja: Uh, I, it has to take collaboration.
- 47:25Dr. Sanjay Juneja: I think all of this keeps going back to, to, to collaboration with a capital C in every single aspect, right?
- 47:31Dr. Sanjay Juneja: When you're propping up a solution, when you're building a solution, when you're vetting a solution, when you're using a third-party one, I think this collaborative process is gonna be so key, and I think we're gonna mess up, unfortunately, you know, m- before we actually figure out all the places that this is necessary in.
- 47:49Francesca: Yeah.
- 47:49Francesca: Thank you.
- 47:50Francesca: I, I love your solution of, of the governing boards, you know, the humans in front of the agents, so to speak, or agents or the governing agents of agents with humans.
- 47:59Francesca: There's, I mean, there's so much that can come out of that.
- 48:02Francesca: Uh, clearly it's an area of focus that we definitely need to, uh, continue to watch.
- 48:06Francesca: Um, and it sort of brings me to the, the, the challenging question for the podcast today.
- 48:10Francesca: We've spent a lot of time talking about, um, sort of ripping things apart, right?
- 48:14Francesca: Like where AI actually fits, where it doesn't, where we're talking about it getting oversold, where it's paying the price, where it's, it's really sort of helping and supporting.
- 48:22Francesca: Um, and I promised you I was gonna ask you this, so when you look out to 2030, what does success look like, and what would make you say, "Hey, in healthcare, we got AI right"?
- 48:34Dr. Sanjay Juneja: If I may answer for oncology specifically, the thing that I was hit the hardest with in practice is when failures happen.
- 48:46Dr. Sanjay Juneja: And I mean failures despite doing everything right by the guidelines, right?
- 48:50Dr. Sanjay Juneja: So the-- you know, there's this thing called EGFR.
- 48:53Dr. Sanjay Juneja: It's a mutation in lung cancer, and it's very favorable in the sense of if you have stage four lung cancers, this qualifies you to get a pill to where you don't need chemotherapy, and it can have, I mean, years of, of control where you don't even see any disease.
- 49:09Dr. Sanjay Juneja: And I would just-- like I've had, I, you know, the first three or four that I had, uh, the patients, even if they were rural or completely medically non-educated, would have tears of joy just because they could see how excited I was, like, about like this is so great, you know, da, da, da.
- 49:23Dr. Sanjay Juneja: Um, um, I'm so happy.
- 49:25Dr. Sanjay Juneja: This changes everything.
- 49:26Dr. Sanjay Juneja: And then I had my first time that I had to walk in there and after six weeks this patient blew through and progressed, and I was so embarrassed, right?
- 49:35Dr. Sanjay Juneja: And, and, and, and discouraged and even like felt somehow it was my fault because I was excited.
- 49:41Dr. Sanjay Juneja: Well, I can tell you I've never had the excitement again even though it happens infrequently, one to two times out of ten, out of ten patients, right?
- 49:48Dr. Sanjay Juneja: But I, I've, I've lost the enthusiasm forever because I don't ever wanna, you know, experience-- and I got to experience one time what it's like to get excited about something and be wrong.
- 49:56Dr. Sanjay Juneja: So when you look at studies You see these, like you see where they've, you know, fall on the cer- on the, on these plot graphs.
- 50:03Dr. Sanjay Juneja: Okay, it's better than survival.
- 50:04Dr. Sanjay Juneja: It's great.
- 50:04Dr. Sanjay Juneja: Wow, look, it's all the way to the right.
- 50:06Dr. Sanjay Juneja: But you see these tails that can be very variable, unlike the people that actually fell below that and above that.
- 50:12Dr. Sanjay Juneja: I think in 2030, all those tails of all these studies are gonna be so tight, and we're-- they're gonna be tight because we're finally able to consider so many different variables we know are part of the cancer treatment process that we just cannot consider.
- 50:28Dr. Sanjay Juneja: Like your, your polymorphisms or your, or your germline inherited stuff that you were given, like the way you metabolize drugs.
- 50:35Dr. Sanjay Juneja: I'm Indian, I'm gonna be different than, than, you know, a white male because a lot of the stuff, you know, for anesthesia may have been studied on white males.
- 50:42Dr. Sanjay Juneja: All of these things that we know matter, medication influence, and even, you know, apparently how much you can taste bitterness has to do with your innate immunity.
- 50:50Dr. Sanjay Juneja: And people that can taste bitterness really sensitively have much more robust immu- innate immune, uh, immune health, so they don't get, uh, uh, viremic from viruses like COVID and flu nearly as much.
- 51:00Dr. Sanjay Juneja: This is validated in NNGM.
- 51:02Dr. Sanjay Juneja: So you have all of these factors.
- 51:04Dr. Sanjay Juneja: In 2030, you're gonna have… I wish we didn't use the term precision medicine already because th-this kind of precision and personalized medicine is so different than anything we've ever, like, conceptualized.
- 51:15Dr. Sanjay Juneja: Uh, if you wanna get techy about it, I think we're gonna have like, you know, stuff that's very tokenized about us.
- 51:21Dr. Sanjay Juneja: I think they're gonna be, you know, these almost like, not an actual thumbnail you plug in, but just tokenized information that's fluid and everyone can, can analyze, and you can put in your different AI solutions and point solutions and platforms and immediately extract things with, with high, uh, fidelity and, and applicability to where you don't have this record deficiency or this data deficiency to even be able to make your, your decisions on time.
- 51:43Dr. Sanjay Juneja: I think that will go away.
- 51:45Dr. Sanjay Juneja: And I worry, let me put it this way-- Well, you said everything that's right.
- 51:48Dr. Sanjay Juneja: I think those are all the things that will be right.
- 51:50Dr. Sanjay Juneja: Uh, and, and And I think there's gonna be probably two to three elements that we can't even conceive today, that, like, we couldn't have even conceived.
- 52:00Dr. Sanjay Juneja: And I think the way to make sense of that is look at what people have done with these videos of AI, right?
- 52:08Dr. Sanjay Juneja: When they showed it two or three years ago, you know, people eating spaghetti, this and that, it was, like, so obviously AI.
- 52:13Dr. Sanjay Juneja: And now it is unbelievable what, you know, was, was classically called deep fake, and now it's just, like, AI stuff.
- 52:20Dr. Sanjay Juneja: If you could do that in-- and it-- I don't think anyone two years ago would have thought we could have done it this well today, uh, that's, you know, nano banana, all this stuff that, that, that they're able to achieve.
- 52:30Dr. Sanjay Juneja: Imagine what that looks like when it comes to potential molecular targets and protein targets and protein, you know, confirmations for, for, you know, antibody drug conjugate delivery.
- 52:42Dr. Sanjay Juneja: I'm saying all this complicated, you know, terminology, but my point is a lot of it has taken individuals studying for weeks at ten hours a day in a lab.
- 52:51Dr. Sanjay Juneja: The computational power you have and the mass amount of, of, of what can be is gonna be accelerated in a way that we've never been able to not only see, but can't even conceive.
- 53:02Francesca: Dr. Juneja, what a conversation.
- 53:04Francesca: I know I came in with a ton of questions, and I know our audience did too, and yet we're leaving with all the answers that we didn't have an hour ago.
- 53:13Francesca: Thank you so much for your time.
- 53:14Francesca: And on that note, I know you're really a busy man, so I'm gonna let you jump while we pull in our top takeaways.
- 53:20Francesca: Thanks again.
- 53:21Francesca: Hope to chat to you soon All right.
- 53:27Francesca: Recap time.
- 53:28Francesca: Here are the things that really stuck out for me from this conversation that we just had with Dr. Joneja today.
- 53:35Francesca: The first point is the hardest problem in healthcare AI is that translation between what the technology can do and what a clinician can actually use at the bedside.
- 53:45Francesca: Most implementations really stall in that gap, and closing it takes builders who understand clinical workflow as deeply as the model.
- 53:53Francesca: Dr. Joneja's whole career is proof that that translation work is the actual work.
- 54:00Francesca: It was also really interesting to hear about data quality as being the foundation for everything else that AI sits on.
- 54:06Francesca: Models that are trained too narrowly on institution-specific patient populations produce AI that might perform well for those patients that are already in the systems, but it quietly fails for everybody else.
- 54:20Francesca: Specialty providers and the vendors who serve them have got to integrate that foundation into anything before they build anything else on top of it.
- 54:29Francesca: And the honest measure of healthcare AI is really whether it does give clinicians back the capacity to be present with patients.
- 54:35Francesca: Done well, it returns time, attention, and judgment to the people who trained to provide that care.
- 54:41Francesca: But done poorly, the consece- consequences are that it adds another layer between the clinician and the person that's sitting in front of them, and that is the version that we have to collectively refuse to ship.
- 54:53Francesca: And that is a wrap of what has been yet another great episode of HumanifAI HealthTech.
- 54:59Francesca: As always, I'd like to thank you, our listeners, for tuning in and if you're joining us on Spotify or YouTube, we'd love to hear your thoughts in the comments, and don't forget to leave us a rating while you're there.
- 55:11Francesca: Once again, I'm Francesca Vereb, head of marketing at NetHealth, and I'll see you next time