Brian Bell (00:00.692) Hey everyone, welcome back to the Ignite podcast. Today we are delighted to have Ganesh on the mic. He is the founder and CEO of Atomanize. Did I get that right? AI, whose AI agents are now running inside of three of the five largest US health plans. He left a fifteen year run through Dell and Intel to build it. And this year, that bet earned them a spot on the World Economic Forum's technology pioneers list. So thanks for coming on.
Ganesh Padmanabhan (00:24.574) thanks for having me, Brian. Great to be here.
Brian Bell (00:27.181) well I usually start the conversation with your origin story. What's your background?
Ganesh Padmanabhan (00:32.918) I grew up in India, southern India by the coast. spent I did my you know, formative education. Like I did an und I have an undergrad in mechanical engineering, although I never worked a day in mechanical engineering. found computers along the way, did a few different funky things during college, built businesses and stuff, and then got into technology, found software. This was like early 2000s, and got into software engineering. Was working for a company called Adaptec, got into Dell, I mean Intel and then Dell, did a long, you know, 11 year stint at Dell and rose to the Langs as engineer, product manager, general manager for a business. We built the youngest billion dollar business at Dell, which included a lot of data, a lot of AI, those days of AI is different kinds of AI now. and then like somewhere in you know, after that run, Dell EMC merged and I was looking at what do I do next? And got the start of Buck to start building companies again. So founded a few companies, had some good runs, worked for some really small, amazing organizations and started autonomize about four and a half years ago.
Brian Bell (01:43.103) Yeah. Really amazing background. I'd love to dig into the mechanical engineering thing. 'cause 'cause you never did that. Like how'd you decide on that degree? And are there any regrets there? Or do you feel like it was like a really good foundation for the rest of your life?
Ganesh Padmanabhan (01:59.085) I think it was a very, very good foundation. And how I got into it was like my my granddad, you know, my mom's dad, used to, I mean, he used to run a a machine shop and you know, he had this huge automobile workshop. And he would, you know, I would spend all my summers there and he'll put me to work. And at one point he actually made me, you know, take over like you know, you know, these like they had these big quarries but big compressors on the back of trucks that'll be used to break rock. He made me took apart the entire compressor, like every single piece of it, and taught me to label it and stuff to that and put it back together the same day. Took me three days, you know, eventually. But I did it. I eventually put it back. So that was the bug that caught me in terms of mechanical engineering. but you know, it gave me a lot of good foundational, fundamental things around like how things work. And physics is a a model for a lot of things in life, right? So it's it's a so it kind of helps you understand physics a lot more, math a lot more, and so forth. So build a foundation.
Brian Bell (02:54.854) Usually it's the smartest people I know is they they got physics degrees, you know. Like I did I did the hard degree, yeah.
Ganesh Padmanabhan (02:58.476) Yeah, yeah. I mean like in it th those are the those are the I mean like physics in itself as a degree is much harder than the physics that I had to take to be a good mechanical engineer, but i i it was kind of a good way.
Brian Bell (03:06.334) Right. Yeah, yeah. It's like finance to math, right? I got a finance degree and it's a lot easier than a math degree, you know.
Ganesh Padmanabhan (03:12.33) Exactly. So I think one of the things that, you know, w mechanical engineering really teaches you to work with atoms than just bits. And then that gives a really good foundation to work with bits is the way I view it. no regrets. I all I all I wanted to do back in when I was going to college was design really fast cars. And there's a little bit of like I haven't been able to do that yet. So at some point I will I don't know.
Brian Bell (03:34.676) That would be amazing. That would be amazing. And I hear I have here that you played or were training to play professional cricket. Tell us about that.
Ganesh Padmanabhan (03:41.815) Yeah, so you know, I grew up in India. Like India, one point five billion people, cricket is religion, right? Everybody plays cricket, cricket's sort of thing. It's also extremely competitive. But when I was growing up, there was no professional path to cricket. Today it's very popular. There's a lot of professional thing. I in fact I was just in India this summer with my son who plays cricket in the US right now. And then he actually played cricket there and very competitive. So I mean, in this were the days of like Big cricket heroes, and all you wanted to do as a kid was grow and pay this professionally. And playing professionally at that time is play for the country. You know, that was just the professional league. That was the only professional team that you can get paid doing that. and and yeah, so I I I lost two voice, by the way. so so yeah, so that's how I, you know, really wanted. I played at the national level. never really, I mean, I really enjoyed it, and it was a lot of hard work, extremely competitive. but it was so much fun. And then somewhere around like when I was in college, got busy with a lot of things, I kind of like fizzled out a little bit. And my my parents were also coaching me saying, Nobody makes money playing cricket. So just go, you know, sports was not a viable career option in India at the time. And I we knew I knew a lot of folks who were like really big deal in sports back day and and they were still after they retire and if they get tired, they have to go back and get a real job to get a put their life together. So it was not an option. But I'm glad I did that though. That was amazing. And I try to carry that athletics mindset cross life after that, right?
Brian Bell (05:19.56) Yeah. Yeah, I think it teaches you a lot of like video games, I think too, but like the hard work. With hard work and practice you can get better at something. Right.
Ganesh Padmanabhan (05:31.609) Yeah, no, I think you know, I used to I just remember this quote by I think it was Jeffrey Archer, and he says, Talent and energy and you're the king. No talent and still energy, you're still the prince, and talent and no energy, you're a pauper. And by energy he means hard work and talent that's just like what are your god god given thing. And it's like stuck with me right from the beginning. Like, you don't have to be smart, you just have work hard. You know it's just
Brian Bell (05:54.44) Yeah. Yeah. Yeah, that's I'm fond of saying that. I actually just got quoted on a podcast a a VC friend of mine that that I don't consider myself an A player. I consider myself a B player that just works hard and spots A level talent. And I was like, Yeah, that's kinda that's that's kinda true. I just you just work hard and it makes up for it. so you spent fifteen years across Intel and Dell running lots of different businesses, growing them scaling billion dollars. Walk me through kind of what that experience
Ganesh Padmanabhan (06:11.062) Yeah.
Brian Bell (06:26.526) taught you about how how to sell into giant enterprises, how how do these enterprises make technology decisions and how that's informing your work now.
Ganesh Padmanabhan (06:35.896) You know, Dell is operational masterpiece as a company. I mean, like I I'm a huge fan of how they, you know, the way you you scale a business from like Dome Roman, Texas to, you know, a hundred billion dollars in revenue. It's just ridiculous, right? And then over a long period of time, very founder led, still very founder led with Michael Dell. And that whole my my I I I like to say I grew up at Dell. So a few things I learned really well. One is This whole the the the operational, how do you build a rhythm in an operational business? How do you scale the operations? And it's all about like a few basic building blocks. It's like about, you know, structuring, you know, like driving and having very good clarity on the mission, making sure you actually put the right structure and the right people in place, and then having an operational rhythm that becomes almost boring, you know, over time, but that's the rhythm that runs the business, right? It's incredible. And then even during that time, it was all this, you know, like when when you have rest of the markets at that time were optimizing for looking good or share price or things being, you know, sounding cool and stuff too, you know, Michael used to say, like boring makes billions, right? So you just like focus on the fundamentals, focus on this thing and the do the boring stuff. And that was a huge lesson for me. Like just build a lot of operational muscle in me. And you know, it's it's hard to actually automatically translate it to a smaller company. But then over time that that's the the the that's your personal personal structure you put in and layer in on so that was one thing right on on what I learned from it. Second, I think you know thinking in decades and thinking in you know multiple decades is a rare skill, especially in companies and startups. So Dell taught you that. Dell was always very, very, very long-term focus in terms of like, you know. Even like look at all the different industry transitions that happened, and Michael had figured out a way to actually come on top of that. And so that that's a very I think most really successful companies running for a long time teaches you that, like in terms of long-term focus. and then on the enterprises side, I think I got the great opportunity. I was part of the initial team that put was put together to go start the software group at Dell. I was one of the first product managers in this group at that time. We were building infrastructure software.
Ganesh Padmanabhan (08:56.034) Zell was just a hardware company. Everybody knew us for PCs and enterprise servers or data center equipment, but not really software. So we had to build an entire software business. Gave me a wide variety of experience across multiple industries, multiple kinds of organizations, and really built a muscle around, like you said, enterprise. How does enterprises operate? More than just the selling part, like the understanding motivation, understanding how they make decisions, understanding how they communicate among themselves that you can model in. And then I it was the GM for a ran a business there which was called the Converge Solutions, which was anything that was not a box or a piece of software, but an integrated white glove solution, whether it's high performance computing, whether it's you know SAP HANA in those days or Hadoop clusters for big data, you had to go orchestrate that particular a sale for for that matter when you're selling that to an enterprise across multiple layers of the stack from technology organizations to business to operations to you know, infrastructure and all of that stuff. So that was a lot of me, you you you pick up a lot of skills. And Dell is also like you know, for Dell and Intel, these large companies, like I said, the operational system is very strong. So you learn a lot from that, but you also know what won't work at a smaller scale. So that's the gauging that you have to do as you get through it. So but that was a I I would say very, very formative years for me. built a lot of relationships, very strong relationship, had some strong mentors, you know, which all resulted in like I still call back some of my old folks that I used to work with like th there was a CTO I used to work with Adele and he's retired in Phoenix and Sam I call Sam, you know, or Sam calls me once every six months or so. We talk about stuff. And, you know, I I think, you know, everybody as as Steve Jobs says, looking back, you can connect the dots. But you know, the reality is a lot of the things you learn and, you know, I was lucky to be in a place that I could maximize my rate of learning. And for some people it is like into getting into the workforce today that might be more in startups. or in large companies. But in alternate histories, but I think, you know, really have to maximize what's offered to you.
Brian Bell (11:01.81) Yeah. So you talked about how COVID forced you to stop and ask what you'll tell your grandkids about how you spent your career. walk me through that moment and like I feel like a lot of people did this kind of transition around twenty twenty.
Ganesh Padmanabhan (11:15.5) Yeah, yeah. I think look too too it was it was interesting, right? I mean, COVID it challenged everybody, like the entire world in so many different ways. And, you know, from societal, the infrastructure, healthcare, all of that stuff too. But it was more a mental tussle than everything. And like I was I had prided myself to be an AI at that time. And I was working on like so I left Dell in 2016, and then the next four or five years I was with a company. Started a company, got hired by this other company called Cognitive Scale. and I was running growth for them and we worked with several of the large enterprises to help them, you know, understand their customers, to, to target their customers, use AI to do digital marketing, better, make search better, all of that stuff, right? And then I did one more startup around data and our customers were like the NV NVIDIA was a customer of ours. They were using dig us to do digital marketing, to target people and stuff like that. The COVID moment made it really real because the people who actually stood out in co during COVID was the frontline health plan health workers, right? And because they're the ones who risked their lives for their families, well-being at that time to go out. And when you don't know a lot of these things, and you know, they they put that themselves out there and do that. And I was like, crap, I mean, look at them, and they're, you know, doing so much. And I'm here trying to make people search more and shop more with AI, right? So that was the that was a moment. But you know, that led to an exploration. And during COVID, we had I mean, like like everybody, there was several, you know, health challenges within the family, got to learn a lot of this thing, got first hand experience of how the system works and stuff. And I couldn't unsee two things. One is, I mean, my God, this infrastructure for one of the most consequential things you'll do in life, which is healthcare and taking care of people's health, is so antiquated, it needs an upgrade. The second part of that was like also I'm a capitalist. I look at it and say, as Bezo says, your margin is my opportunity. I look at the pockets of healthcare. It's so fragmented and stuff. That's an opportunity. That's an opportunity. That's an opportunity. That's an opportunity. So I'm like, well, I could spend my next two decades working on this problem of how do you make healthcare better? And I won't regret it. There's going to be enough problems to solve. And I get, you know, excited with the rate of learning and solving tough problems. So this is what better industry to go after.
Brian Bell (13:36.125) I just saw like a shocking headline to build on what you're saying in the news. It was something like, a lot of healthcare prov payers, you know, insurance payers are dropping GLPs from their plans. They're saying, Hey, we're not gonna pay for your GLPs anymore. and the reason they're dropping it is because they're paying through the nose for it. They're paying two grand a month for something that you can go online and buy for two hundred a month. So it's a it's an order of magnitude more that you know, that these pharmaceutical companies are locking in these payers to pay like enormous amounts. And there there's that margin there. And and it's just crazy.
Ganesh Padmanabhan (14:17.622) You know, it's it's it's interesting, you know, you this it to me that's a symptom of a larger problem. It's a so you you the challenges specifically in the US healthcare system, it's a multi-party system. You know, we we brought capitalism into healthcare, which is not bad because it's kind of the whole thing was to drive competition, make you know, the market actually select and self-select and stuff. There's parts of the market that it's happening. if you look at innovation that's happens in healthcare i i in the US versus anywhere else. We're like far better. I mean, like we have the the Roman, the the the Rose of the world that is actually happening here. None of the other markets are really even seeing any kind of innovation. The same old stuff from from decades ago. But then the flip side of that is like when you were trying to go do this and do this at scale for 330 million people, you have to put guardrails, you have to put, you know, regulations, you know, oversight and stuff. And over time that just evolved as like so the one of the classic s problems. when I looked deeply into healthcare, I saw was healthcare didn't have a dearth of problems to solve. But the way we were incentivizing people to solve problems for healthcare was like you if you solve just a symptom of a larger problem, you can rinse-repeat that the market's so huge that it actually becomes a profitable enterprise. So all of a sudden, 30 years later, 40 years later, what you see when you look at healthcare. Is this bunch of point solutions? Everybody optimizing and thinking they're doing the right thing by solving a one particular problem. You may may made the example of the the GLP1 drugs. So pharmacy benefit managers who actually negotiate pricing with the life sciences companies have the ears of the health plans who have the ears of the employers, which is about 50-60% of who pays for healthcare in the US. Well, it made sense when information will flow through that. It made sense when You know, you have to have the Godrails and stuff. Now information is ubiquitous. It just didn't evolve into the AI era or just the the today's era. So that one example, or a SAS era, I mean, like it's it is there's I mean, but that is just one problem. There's so many different things. Like you go to an MRI, you know, and you go do you know, hey, you're out of pocket, you haven't hit your out of pocket, you know, deductible for the year. So your MRI bill comes back and you're paying $800. And we're like,
Brian Bell (16:17.108) Or even the SAS era really. Like a SaaS marketplace era. Yeah.
Brian Bell (16:38.536) Right. I could have just went and paid cash like three hundred dollars for the same MRI cash.
Ganesh Padmanabhan (16:39.19) Okay. But then if you go and Exactly. Exactly. So it's like the things like every you the optimization happens in each of these workflows in silos rather than for the global maxima, rather than for the market, right? And that is the essence of the problem that we have in healthcare. We have a lot of really innovative stuff that is happening in narrow silos that would never compound to make the system better. Right. And I think that's one of the missions of autonomize. That we were like, look, I mean. We didn't want to when people were asking us four years ago when we started, are you a care management company? Are you a prior company? Are you a, you know, a patient access company? And we're like, no, we're an infrastructure company that's going to build the infrastructure that allows you to solve any of those problems you named in a way that doesn't become a siloed solution that I can that can compound to making a difference in the network and the in the in the in the industry itself.
Brian Bell (17:33.588) And I wanna I wanna talk about what what you guys are doing. but I wanna talk about you know, you write you wrote about losing w was it a friend of breast cancer and that that was part of what happened at this time and kind of impacted you in a big way.
Ganesh Padmanabhan (17:47.501) Yeah, no, it was it was definitely part of it. I think my I lost a friend to stage four breast cancer. And it was it was it was sad and she's fought for it fought with it like for over three or two or three years. The first time when she was actually diagnosed and she was going through the treatment, and there's all these protocols in the house. There's this there was this drug called K Truda, which is a Merck drug. It's been on the very popular radio immunotherapy drugs in the market. They were going through a phase three trials for this application for stage four breast cancer or stage three breast cancer at that time. and there were clinical trials going on. So it was experimental, right? And so my friend applied
Brian Bell (18:28.82) Well at stage three, correct me if I'm wrong. They're trying to figure out does it scale? Right? Yeah.
Ganesh Padmanabhan (18:33.926) It's so phase three, yeah, they're trying to figure out, but it's phase three is when you're usually you're trying to look for outcomes. Stage two is usually safety that it doesn't hurt more than no do no harm. Stage one is often, you know, very limited indications. Do you see enough endpoints that you can manage? So stage three is when you're actually running a hundred people, two hundred people through that, taking through this and you're collecting data, you're inspecting the whole thing and so forth. An average, by the way, the the economics of that is insane. A pharmaceutical company takes about
Brian Bell (18:40.08) Mm-hmm. Right. Right.
Ganesh Padmanabhan (19:02.914) 10 to 12 years and about two to three billion dollars to launch one drug. Insane. That phase three trials sucks up a lion's share of that because it's a lot of orchestration. You've got to put a lot of Godrails in it and so forth. So she applied for one of these clinical trials that, you know, again, it's all would have, should have backed looking back now. Well, she can get selected. But when you look at the process in which this patient selection happens in clinical trials, you get a five to 10,000 page medical report of a patient. And that some nurses or doctors and PIs, they call it, are sitting down, shifting through. And if you have an information about the protein expression that makes you a best candidate for this therapy and this treatment, that is buried in page 497 of a 5,000-page report, it's easy to miss. It's humans, right? It's not intentional. It's in the so that experience stuck with me a lot because, you know, and then eventually she did get on a Kitruda treatment. in immunotherapy treatment, but by the time it was too late and all of that stuff. So there was a it it just but but that showed the plight of the industry, right? You know, so clinical trials are good ways for, you know, obviously pharmaceuticals to launch new drugs, but it's also lifeline therapy for folks who are looking at, hey, I've exhausted the standard standards of care. I want experimental therapy to save my life or something that is really not working for me. And But it's a lottery to get into that because it's so burdensome for the human capital that's involved in it to, you know, we have far exceeded our ability to look at and consume all the data that's being generated around us. And and then, but in healthcare, we expect somehow that, you know, the nurses and MDs are gonna be, you know, superpowers. They're gonna be able to do that. It's you know, so it it that was that was part of the the big motivation of like and it just reinconfirmed it. In fact, the first thing that we focused on at cogni at autonomize is. was actually the clinical trials patient matching as a problem. We launched our first customer was actually someone who ran a phase three trial for a oncology blood test. So there was a it was a blood test to detect pancre I think it was colorectal or pancreatic cancer early on, right? So just a blood test and they do omics AI on it and so forth. But then we were the one who helped them select the first, you know, or the the
Ganesh Padmanabhan (21:24.76) first hundred and fifty patients for their phase three trial. So we try to make a dent into that space. We eventually, you know, you know, like we got broader than just clinical trials and so forth.
Brian Bell (21:36.092) I've I've actually heard, I think I heard it on the All In podcast this past week or week before that a lot of biotech research now is moving overseas. you know, because it's just you c you can get through the the the phases faster and cheaper.
Ganesh Padmanabhan (21:51.885) Yeah, I mean, it's it's a double edged sword, right? Like so there yes, definitely. I mean, there's a lot. I mean, the amount of the biotech revolution in China, for example, is incredible. I was in China this summer and it just it's just incredible to see how how they have progressed in doing that. And part of that is actually look, you have state controlled infrastructure, you can go through a lot of these things faster than a democratic you know, process that you have to go through for doing that. The bodies are different. They're a lot looser. Like, you know, it's it's it's it has pluses and minuses. You know, also like, you know, some of these things when you're trying to do human trials and stuff, it is very, very, very you know, like it's it's it's it's risky because it's exp experimental, right? It's very early. So there's that, you know, the the the the the yin and the yang of the same problem. But that said, I think look, I don't think FDA is one of the fastest organizations in the world. But I've been impressed like the last six, eight months, like yesterday or day before yesterday, they launched that they approved the phase three trial for a moderna delivery mechanism for immunotherapy for cancer patients for pancreatic cancer. One of the and it extends, I mean, they've done enough studies now to extend lifespan at least by twelve months, because if you get pancreatic cancer, your prognosis is like you're like you're given six months to live, right? Right. So it's incredible. I mean, it's like it's a lot of
Brian Bell (23:14.492) That's one of the worst, more the fastest killing cancers out there, right? Yeah.
Ganesh Padmanabhan (23:20.108) you know, AI that's happening behind the scenes and but also using mRNA as a transport mechanism to do it. So there are there is it's the the our regulatory bodies are getting faster. They're not fast enough. but there's also like you know do you just go Yahoo Wild West and go do this? And like there are countries, you know, not naming anyone, but who will just let you experiment on their population to figure out what is the best drug that you can get out and launch in a global market. Right. You don't want to be there either. Right. So it's like the it's a balance of saying, like, you have these guardrails. By no sense that our infrastructure is modernized, right? I think the biggest thing, the biggest takeaway is like if you look at payment technology, if you look at financial technology, if you look at, you know, shopping and e-commerce and stuff, compare that and their rate of growth and evolution over the last 20 years to healthcare. We're like way behind. That said. Lot of dollars going into healthcare, a lot of investor dollars going to healthcare. I think there is, you know, the future is bright, but it's nowhere close to where it should be.
Brian Bell (24:23.944) Yeah. So autonomizes agents are now embedded inside of three of the five largest US health plans. Walk me through you know, what are those agents doing on a on a Tuesday morning? Or Thursday morning as we're recording. Yeah.
Ganesh Padmanabhan (24:34.262) Yeah. Yeah. So yeah. So our our goal, you know, we we work with health enterprises and we're one very singular focus and what we want these agents to be deployed on. So a a couple of things. One, we believe every organization is going to be a hybrid enterprise. It's going to have, you know, agentic workers that'll actually do the mundane, the not critical, the not low risky in some cases, but also things that doesn't require the human ingenuity, judgment, taste. you know, touch points, all of that stuff. And then the humans will do the orchestration work, will do the reviews, they'll sign off on things and all of that stuff. So that's a fundamental belief that we have.
Brian Bell (25:12.04) Yeah. The the phrase I use is like everybody becomes an executive with with a team under them.
Ganesh Padmanabhan (25:17.314) Yeah, and no, and and in the team may not be just all humans, right? It could be a lot of bits running around. Yeah. And I I I I think the the the truth is somewhere in the in the middle, right? Like for example, it's it's not gonna be all the way like, you know, we talk about an open claw experience for hospitals. I don't think it's gonna go there because healthcare is one of those industries. Healthcare is more about care than health. So and care actually has you know human
Brian Bell (25:18.908) Yeah. Right, no, it'll be agents and AI, yeah.
Ganesh Padmanabhan (25:41.903) You know, it's how you feel when you meet a doctor, not just what the doctor says, right? And that is very hard to be replaced with just agents and things. So on a Tuesday morning, what we do is actually look at the entire we mapped out workflows across the enterprise for healthcare, be it front office, middle office, back office, and we identified pockets that can just be, you know, you can identify it instead of a human's doing it, like a nurse spending 30 minutes reviewing charts to approve a prioth, right? You know, they still have to approve the prior auth, but if it's an approval, reviewing that a 30 minutes of review, can you compress that to two? Right. And if it says an approval, why do they even have to touch it? Can you auto-approve? All of that stuff is like one example of agents. So what we focus on is like the administrative layer that this people who are not working at the top of their licenses are spending and wasting their time on. And we identify that. Now in that process, we find pockets where you can reimagine the workflow. For example, Like I took the example of you know, like care gaps identification is a problem that is basically you're looking at did our population, it's a population health problem where you're trying to see that you're getting your hemoglobin A1C tested, you're making sure you're having your wealth checks happening, you're doing your mammograms annually if you're over 40, all of that stuff. You that whole process, it sounds so mundane, but it starts with somebody pulling a list of all the codes in a of that particular members. Someone looks at it, someone identifies, this patient has not had their mammogram. Now, should they have their mammogram? They should actually look at their age, look at their demographics, look at the medical plan, what plan they're in, and then you go and let them let the hospitals know, let them know. They figure out appointment and stuff like that. That particular problem is a fully automatable problem. Or, you know, you're getting prior authority requests for a health plan that has got You know, charts in it, you identify a care gap, you can activate a particular agent to go and update and check those different parameters, identify it as a cap, confirm it, automate the appointment booking for them, send them an update. They show up to the doctor, they finish their mammogram, they get their results, and everybody's happy. No human really got involved in it, right? Other than for sign-off and you know, when you start monitoring it and stuff like that. So those are problems that you can just completely do autonomous workflows that is agentified.
Ganesh Padmanabhan (27:59.715) The others, it's a little bit more human in the lead, human in the loop. So you have the human being the executive. This come up with this thing, hey, I have collected all this information. They don't meet this criteria, but you might want to override it and look at this information. But instead of you spending two hours digging through this, you do it in five minutes, right? So we do all of these varieties of things. So we power now four of the five largest health enterprises of the US. And, you know, and and we do everything from Utilization management, care management, claims and payment integrity and revenue cycle management. Mostly that bucket of you know capabilities that we actually offer is how do you orchestrate the business of care? Right. It's less about the clinical decision support, even though because you don't want to offset and you know, it's not about telling the humans. There's a lot of there's a Jama article that just came out how it was a little tasteless, which is it basically said human doctors will be obliterated by AI doctors. I don't think that is actually true because it's not very contextual. And LLM actually doing, you know, better in tests is not an indicator it's going to perform well in real life, right? So that was a problem there. But we see that the big opportunity is look at that entire admin layer of work that gets performed and take it off of the human's hands and put it in agents. So that's what we do. So on a Tuesday morning, we have some of those agents just, you know, making calls, making sure that they're actually reviewing things, processing information. And making humans better at what they do.
Brian Bell (29:28.052) I what sounds like it's it's more the administration burden of delivering healthcare. That's that you're covering here. Right.
Ganesh Padmanabhan (29:33.783) That's our primary focus, yeah. And we're seeing the impact of that. Like there's a first order benefit, is you get the admin savings and the other, you know, the fact that you're not spending this amount of time. The second order is you drive consistency in the process. So instead of you depending on a tired, you know, clinician who's been on a end of our 18-hour thing, you reduce that or so you make it more consistent. And the third order is that allows you and you generate so much data to look at it and say, should this workflow be designed this way, or is there an opportunity to go redesign this?
Brian Bell (29:49.86) Errors and yeah. Yeah.
Ganesh Padmanabhan (30:03.182) To my point, like, you know, the fact that, you know, you have this process of humans going left to right is, you know, because you assume that all the work happens on a screen on a SaaS software. But if the work happens behind the scenes, you know, how should software look? Or should the software even be there? Right? All of that stuff allows you to re-architect and reimagine that particular workflows.
Brian Bell (30:25.404) That's what you guys just launched with Genie AI, right? Which lets lets the actual users, nurses, case managers build their own workflows. Maybe you could tell us about that.
Ganesh Padmanabhan (30:32.108) Yeah. You know, so so Genie AI. It's actually I'm super excited about this launch. So we over the last four or five years, we launched several things, right? Like one is we have a core platform, which includes how you normalize data, organize data for a health enterprise. We have a context layer, we're gonna launch a a a new product on that in the coming weeks, which allows organizations to store and maintain and own their own knowledge instead of offloading it to an LLM. And then we build this library of agents and applications that use these agents to solve different problems, right? Now, one thing we learned during this process was look, we can build all of these solutions for customers, and we do a lot of work, you know, we research it, we work with multiple customers, and we do that. But there is nuances that are very unique to organizations. And so, and then in within organizations, within groups and individuals. So the big idea was like, how do you give the power of creation? These powerful agents and these powerful applications to the people who are closest to the problem. That's your clinician, your care manager, your case manager who know the nuance of how this actually gets done. So they don't have to traditionally that process is they come up with requirements that go to the technology team, they do an RFP, get a bunch of vendors, they look at it, they do it. And by the time they get to see it and they say, this is not what I asked for, right? It's a common thing that happens. So we said, why don't we push that in, but maintain that instead of them having so what they're doing today without a Gini AI is they go to wipecode this in cloud or wipe code this in Chat GPD. And that's not going to be enterprise scale, enterprise ready. So this gives the safe space for them to design and automate and orchestrate their own unique workflows on top of the, you know, the the infrastructure that we laid it for them. So it's been hugely well received by the organizations and it's a You know, our teams have like so one of the big transitions we did is like we have a team of clinicians within our organization, people who are more business subject matter experts and clinicians like nurses and MDs. They use it internally. Like so earlier it used to be they will be there and a solution engineer or forward deployed engineer will be there with them to build these solutions. Now they sit with a customer and say, You mean something like this, and they can model that entire workflow. And then the technology in the IT team is not unhappy because it uses, it doesn't build.
Ganesh Padmanabhan (32:49.462) random stuff from, you know, secure s you know, vulnerability oriented, you know, GitHub repos and stuff like that. So it's more governed and controlled and stuff like that, but you're getting the value and it's closest to the product, people who understand.
Brian Bell (33:03.092) Awesome. And then your research team just published a paper about benchmarking it marking it against other AI systems. Maybe you could tell us about that.
Ganesh Padmanabhan (33:10.946) Yeah, no, it's awesome. So the the the thing about AI in general is like the market's evolving really fast. Research is a very expensive proposition, as you see the balance sheets of all the big frontier labs. but super, we have a small but very, very mighty research team. And one of the things they actually found was a f of the following. One is when you're using LLMs, age for for agents, LLM agents and using that to do reasoning. The pro one of the largest problems, it's not so much there's a token problem, which is like it takes a long time in doing that. But more than just the tokens, it is about like you're trying to reason for every single time. And then because of the way LLMs work, you may get slight variances in the answers because it's meant to be creative. It's generative in nature, right? How do you solve that? So we have a lot of these techniques around it. We use like if you prove a fact once, turn it into a deterministic code. If you will, instead of actually having to reason again. So that kind of stuff. It's basics, right? So and it's so but then this new paper that we published was something called RLM opt, which is basically RLM is recursive language models, which is the idea that you know when you start solving complex problems, your your tokens and your your prompts become so complex and large. So we use AI to automate that particular prompt to to to optimize that using.
Brian Bell (34:07.482) If if this then that every time, right? Yeah. Yeah.
Ganesh Padmanabhan (34:35.242) recursive pattern. You tie you test it against the validated data set and you get it closer and closer to get to the 95, 96%, whatever you want it to be, but it's done by AI instead of humans having to do it. The results of doing that is one, you're making you're getting better performance for your prompts. second, you're using up to 90% less tokens to do that. Because usually prompts are like just software code, it's just lights, you know, you just start writing and you never get take it off. Right. The so there was a state of the art for this technique called JEPA, G E P A. I don't know what it stands for, but that was a state of the art published by all the frontier models and stuff. We beat their met benchmark with this RLM opt, you know, that's now actually being cited. We already I think we published the paper last year, last week, and it's already gotten a lot of citations and stuff too. So it's it is now the state of the art from this little company in Austin, Texas, you know. that's yeah, it's so that very excited about it.
Brian Bell (35:25.428) Nice.
Brian Bell (35:29.458) That's cool. Yeah. I can relate. I mean, we run we run Team Ignite on AI and I I used to I l I led AI at AWS. So I've built that system that you're describing to predict startup success. You know, where I have this huge data set and we recursively improve it. probably once a week. I I rerun it. sometimes, you know, once a month, but because it's pretty expensive now to to run it across tens of thousands of companies. But we've recursively made the prompt better and tighter and more token efficient to predict, hey, if I take a startup's like, you know, website and LinkedIn profiles and pitch tech and whatever else I have, you know, how likely is this to become a unicorn? So I, you know, not not as good as like solving healthcare, but
Ganesh Padmanabhan (36:14.062) No, but we know it but the technique is very, you know, applicable beyond healthcare. So my ch if you if you're doing a recursive thing, the the state of the art in recursive LMs is actually, you know, JEPA. So you should check out the paper. It's called RLM opt OPT. and it's it's gonna give you Yeah, it's gonna give you and it's got the technique. It's a detailed paper, it's open source, if you will, right? So you will get the technique on what to do and and you know, our customers just use it. For healthcare context, we have that pre built for a lot of the different use cases we support so forth. Yeah.
Brian Bell (36:26.792) Yeah. I'll go I'll go check it out, yeah.
Brian Bell (36:42.514) Right. What what model are you fine tuning? Is it like a Kimmy three or whatever or?
Ganesh Padmanabhan (36:48.236) No, it so it could be anyone. So for what we're do using for that is probably a, you know, the the so we we we have a library of things. Like talking about models, we have a library of capabilities or models, if you will, right? The thing with LLMs is like it's a hammer looking for a nail. And most problems that enterprises use, you don't need that large model that is trained on trillions of parameters and all the you know world's data. It helps in a lot of different cases, but it it's an overkill in most. So we have a library of fifty-five odd models. In addition to the frontier models, which are we we being in US healthcare, we avoid using a lot of the Chinese models. So even Kimmy K3s, we test and we benchmark along that. We use a lot of Gemma, which is from Google. We use a lot of the you know, the the the the European models, but open source open weight models that we have further fine-tuned for you know, for for healthcare use cases and a specific thing. So we have a library of that. And we also have an auto selector. So when we actually f encounter a new problem, it actually looks at what is the best base model to go solve this and it automatically creates a benchmark and tells you like yeah, you should probably use this on Quen three billion instead of trying to do it on, you know, Opus, you know, six dot Right. So Yeah.
Brian Bell (38:05.108) Yeah, it's more the routing and the harness. And you've actually made this case that everybody's racing to deploy more agents, but nobody's building the thing that manages them. kind of how did you come up with that insight?
Ganesh Padmanabhan (38:13.454) Exactly. You know, you know, it's it's if you if you things if you look at how organizations evolve, if you believe in the fact that this, you know, for us to do all the work, humans have far exceeded our ability to deal with all the information and make decisions and be everywhere and we're just becoming more and more ambitious to do more than less. that will play out in a way that's saying, Look, you need AI to actually be doing, be an exoskeleton doing things for you. When you do that, like you we talked about in the beginning, you're gonna have AI workers in addition to human workers, right? And for organizations and enterprises, you have an ERP system, you have a CRM system, you have a, you know, HR system, you have all of these different things that you had to manage human populations. But we have nothing like that for agents, right? So the the the the the the insight was more about saying, look, this is something the industry will want to do this. So for when we build these solutions for each of these customers and each of those you know, applications and products that we built, we always had the view, like like I said earlier, which is you want to make sure that you're not just solving one problem, but solving the infrastructure that'll solve any problem. Right. So that made us actually look at this from a perspective saying, what do we need if I have to go take this prior, you know, pattern and apply it to payment integrity? So what kind of underneath, how do you manage it? How do you evaluate it? How do you actually harness it? How do you you know put the governance layer structure on it? So, but you know, I would say you extrapolate another five years or so, maybe not even five years, it'll look like the you know, the SaaS software for human resources, it'll look like the recruiting software, it'll look like your training and LMS software. That's what all this harnesses and the infrastructure is gonna look like to manage agents in the future.
Brian Bell (40:02.494) Yeah. I think you said this before. you compared today's AI rollouts to early cloud migrations, where companies just picked up broken workflows and kind of threw them in the to the cloud. Unexpected magical results, but you still have to work with them and it it takes time to to get it right.
Ganesh Padmanabhan (40:21.132) Yeah, no, I you know, look one of one of the very understated role of like what's, you know, you're seeing this, right? Like a lot of token spend that's happening. Enterprises like, why am not really seeing the results? And the result it's one thing to just put a model and expect it to work, which is what 90% of people are doing. But then you don't you forget the fact that you have to now work with it, make sure the model works, make sure you can evaluate it. You're you know, simple things like, you know, I just start doing thirty, forty percent more code using you know AI now. Well, now that just means your PR volume increased that you now have to have more people review PRs, or you're to now become, you know, use AI to review those PRs and approve it. Well, that what does that lead to? You get a production issue, you get into a war room, you sit and review that, and all of a sudden you're looking and saying, I don't know how we came up with that. I mean, how do we change it? All of that is the work that you need to do beforehand. Nobody will realize it early on. So And then on top of that, you have to start thinking about like like you said, the cloud migration is like, yeah, this is the I mean, I ha I remember back in I think it was 15 years ago, we were I there was a VMware software called Lab Manager, which allows people self-service access to build their own compute to do it. I gave access to my team overseas, came back Monday morning, and all my systems were red because the infrastructure were completely utilized. They just got it, but they don't know what to do with it. They don't know how to shut it down. They don't have the governance layer on it and stuff like that. Just use it. That's the state we are in with AI today, right? Everybody has it, everybody's using it, but we haven't really figured out what it takes to make it work. Part of that is software. Part of that is operational changes. Like one of the big fund foundational questions here is like if AI does 60% of what you would do, you know, in a in your workflow, in your work. How should you be behaving differently? And we never address that, right? and then that happens in healthcare, that happens in every industry and so forth. But I think I do think that this is there is there is precedence on what we're how we got here and you just need to follow the follow the history.
Brian Bell (42:26.068) Well, what what are you excited about over the next year? Like, what's what's on the the the one year horizon for you guys?
Ganesh Padmanabhan (42:32.566) Yeah. So a few things. One, we're like, you know, coming off of like we've been we're we're in this phase of, you know, we've grew from like you know thirty-eight, forty employees last year to about a hundred and forty, hundred and fifty employees right now, plus the partners and folks that are use and contractors and stuff like that. we've tripled business since, you know, mid last year and It's it's a it's been growing really fast. So a lot of the stuff that we're focused on is like, okay, with scale comes a lot more desire for structure and you know, so a lot of the work that are doing is a lot of the company building that's happening.
Brian Bell (43:13.096) Yeah. Turns out you were at Dell and you know how to do this though, right? Not only did you take something zero to one, but you can you can go ten to a hundred. Like that's that that's Dell boring billions saying, yeah, you know, boring makes billions. Yeah.
Ganesh Padmanabhan (43:15.542) Yeah, it turns out, yeah. At least I have a little bit of exact
Ganesh Padmanabhan (43:22.498) Yeah. Yeah. So, so thankfully. So I have a little bit of intuition. We're still hiring some amazing people to help us fit that. So that's that's one part of the company building thing. It's a lot of interesting things, right? What is the culture? What is the ideal culture in the age of AI? Right. It's not just move fast and break things. It's like how do you teach people discernment? How do you teach them to actually challenge and be more confrontational in an age where AI is gonna be psychophantic and tell you exactly what you want to do?
Brian Bell (43:40.809) Mm.
Ganesh Padmanabhan (43:50.606) So there's all of these things that come in in terms of yeah.
Brian Bell (43:52.253) Or or it's smarter than you too, right? So how do you how do you deal with like this this raw intelligence that knows more than you and is actually smarter than you? Yeah. All right.
Ganesh Padmanabhan (43:57.603) Yeah. And and so how do you harness it to become even more, you know, not just productive, but even just smarter? And how do you keep up your curiosity? How do you follow your gut and curiosity? You know, that otherwise that you you know, like the classic example, you give someone a problem, they'll take it, copy into clot code, take the answer and give it back to you. And I'm like, no, you got to challenge it, you know, have the dialogue. So it's a it's a very imp important skill that you need to plug in into your culture. So a lot of company building, culture building. On the customers, like we know power four of the five largest health enterprises. We're continuously adding more customers. We're doing more work with them. We see a future that we want to really, I mean, with some of our customers, we've been able to compress like hundreds of millions of dollars from their GNA, you know, over the 12 to 18 month period because it's it's it's they love us for that, right? But we do think that there is an opportunity to build an AI native health enterprise, even some of the big companies that we work with. So we want to help them with that transition. It could be it could look like, hey, they launched a new.
Brian Bell (44:44.136) I love you. Yeah.
Ganesh Padmanabhan (44:59.35) you know, business unit within that particular thing that is run differently than their core engine powered by autonomy. So re and we're very close to actually doing things like that that wasn't possible even a year ago, two years ago, right? So really excited about what that few healthcare native or AI native health enterprise will look like and bringing that vision to fruition in practical reality. I can do it in my lab. We have a synthetic, you know, health enterprise running in our environment. That's where we do all our validation. Like Five million patients and all that things. We do all the simulations almost like a digital twin, but it's very different than the real world, right? So we want to go get that vision into some of these organizations to really go beyond use cases and workflows to transforming the enterprise. So that's a lot of focus on that front. We're really excited about this. You know, we we we're building out a transformation team that is helping customers, not just like, hey, you put this in production and they get like 20, 30% efficiency gains. That's great. Now what? Yes, I've compressed 100 million, 200 million off your GNA, but how do I do more? And it's not the answer is not do more of the same. Now you're to re-look at it and saying, well, I've solved that problem. Now let's reimagine the the way work is done and how do you redesign it? So there's a lot of that for focus that's actually happening too. So super excited about it. you know, I think the the talent that we're assembling is just, you know, state of the art and really awesome, as you can see with the outcomes from. from papers or research or you know customer outcomes. We just published today a case study with Cigna Healthcare, where we actually were able to, you know, the the impact of what we do on it's called an intelligent benefits management, benefits routing kind of you know, workflow, patients are seeing their access to care 50 times faster from what that was being done, right? I mean you can argue like, man, how was is that bad, right? But the the the thing is like it's complex and it's complex for a commercial health insurance company and so forth. So we we you know it's really proud of the team there as well as our team to actually come by and solve a really meaningful problem. So we're really look I'm really excited about all those things around like, hey, we there is so much more that in I want to really see a world in the next year, two years that we'll stop talking about healthcare as it is today, which is sick care. Like, how does you fix that problem versus this saying?
Ganesh Padmanabhan (47:23.33) Turn this whole thing to talk more about longevity, prevention, you know, healthier living, health span instead of just lifespan, right? So there's a whole transition that we're gonna see in the market that's gonna happen and you wanna play a very strong part in actually making the transition.
Brian Bell (47:40.232) Yeah. Well, I wish we had another hour to chat. really enjoy the conversation. Where can people find you online?
Ganesh Padmanabhan (47:46.522) I'm very active on LinkedIn. So you can find me on LinkedIn, Ganesh Padlanabhan, my or you can check out what we do at autonomize.ai. I used to run a podcast, Stories in AI. So you can go to Stories in AI too or in YouTube and stuff too, but those are lot lot of you know we gotten too busy to actually do new new episodes, but there's yeah. So but yeah.
Brian Bell (48:06.882) yeah, you can't do that as a CEO, yeah. Yeah. I could barely do it as a as a GP of a venture fund. Yeah. Yeah.
Ganesh Padmanabhan (48:13.676) I know it's hard, but no big great job in in pulling this together too. It's it's it's fun because you know, it's it's it's it's mutual learning too in this kind of formats and so forth. So thank you for yeah.
Brian Bell (48:17.512) Yeah. thank you. Thank you.
Brian Bell (48:22.43) Right. Bingo. Yeah. The podcast is great for that. It it builds the the culture and the relationships and yeah, really appreciate it. Everybody go check them out. And if you're definitely if you're working in healthcare, go check out autonomize. thanks for coming on.
Ganesh Padmanabhan (48:37.455) Thank you. Thanks for having me, Brian. Ciao.
