Opto: Hello, everyone. Today on the show, we're joined by Ganesh Padmanabhan, founder and CEO of Autonomize AI, a leader on a mission to transform how artificial intelligence is used in healthcare. Ganesh has built AI systems that tackle the most complex administrative challenges in healthcare by organizing, contextualizing, summarizing unstructured data. So clinicians and care teams can focus on what matters most, patient outcomes. He's a seasoned entrepreneur. AI strategist and advocate for ethical, human-centric AI in regulated industries. How are doing today, Ganesh?

Ganesh: I'm fantastic at. Thanks a lot for having me.

Opto: Yeah, thanks for coming on the show. Where abouts are you calling from today?

Ganesh: I am in Austin, Texas. ⁓ It's home base. This is where I've lived the longest ever in my life.

Opto: Really, when did you move there? Like a long time ago? ⁓ No answer.

Ganesh: About 19 years ago, I came here to work for Dell, know, a while ago. We're very different life, but you know, and the city has grown so much and, the best part about it, I know you're in New York, right? So it's like 87 degrees today and it's the middle of February. you know, this is the great choice to be in a place.

Opto: I'm still struggling with the, obviously I'm from the UK, I only moved to New York recently. The conversion from centigrade to Fahrenheit, I'm just, is that hot? I'm imagining it, I'm assuming it's hot.

Ganesh: That's right. It'll take you a while, but then you'll know that the are great, ⁓ are hot. ⁓ we're having late 20s here It's amazing.

Opto: See you later. Nice, Yeah, obviously we're here to talk about your business. I thought we could start with, in one sentence, if you can, or a short summary, what is Autonomize AI?

Ganesh: Yeah. Yeah. So, you know, Autonomize AI is on a mission to transform how knowledge work or healthcare operations is done, right? And in a nutshell, what we do is this today, we provide this, I mean, think of a healthcare enterprise that's run very inefficiently across the world. I mean, it's just the reality of like, we're trying to optimize for so many things. What we are providing is an operating layer to layer in and become an agentic health enterprise. So you have a future where humans and AI agents can coexist and then provide outcome at much lower cost of delivering care, increasing access to care that places you cannot get to today with just humans, and just change the operating model of one of the most critical industries for humankind.

Opto: And I mean, there's so many inefficiencies it seems all over the place. Now AI has come to the forefront. It just seems like everywhere things are not as good as they could be. So it's an exciting time to be able to use this technology. Could you just paint a picture of some of the inefficiencies that are around today just to bring it to life for people and how specifically are you going to change that?

Ganesh: Yeah, no, think about today, for example, ⁓ healthcare is inefficient, not just because of ⁓ the inefficiencies around it. There's incentive structures that are kind of misaligned. There's multiple parties that coming in. There is a lot of friction in the system. I'll take a few examples. One simple example, somebody wants to get an MRI done. And in the United States, if you're covered by a medical insurer, you have to send a request. Your doctor has to send a request to the health insurance company. requesting a pre-approval or prior authorization to go do it. Now to support that and justify why this person needs it, they have to do the work of collecting information about their medical background, describing it, explaining it, providing medical records. So the doctor spends 15, 20 minutes just putting a request together to say, I need to get this person an MRI done. On the other side, at the health insurance company, they have people who will... You'll have data entry people who'll get this request either digitally or through faxes still exists in healthcare. And they're typing this up. They take 10 to 12 minutes typing this up, setting up the case for a nurse or an MD to review on their end. And they will look at that request that patients medical history, bump it against clinical guidelines, bump it against payment policies, coverage, determination policies, and so forth. And then they have to make an adjudication. That whole process, a simple request as like, Hey, I need an MRI done. takes 20 minutes on the doctors to make the request, 40 minutes on the health insurer side to actually go do it. And they'll be back and forth. missing this information. So you put that case to pending and then call the provider and say, hey, can you just help answer these questions? They finally get their request and stuff. It takes days, like sometimes weeks, to get this approved. The whole MRI process is probably 15 minutes. So to do that, you're actually spending three or four weeks to go do it.

Opto: Yeah.

Ganesh: So that's one example, but then that happens everywhere. want to, know, somebody's getting discharged from a hospital and, you know, there is a lot of risk of readmissions that happen simply. And that happens because the coordination of care doesn't happen. The patient does not know exactly what they need to do because they're given a pile of 17 pages that actually is a bunch of data that says, here's your discharge summary. And it's got no useful information for them to make sense out of. What are the things they should do today? What are the things they need to do tomorrow and the day after? so they don't have to end up back in the hospital. So, and all of that is, know, we've not evolved healthcare as an industry to today's age where, you know, things are more digital, you know, you have, you know, AI as an operating layer that can be used as a leverage to go change it and so forth. So all of these inefficiencies exist because A, there is multiple parties involved in delivering care, which is the way it should be. I mean, that's, it's a multi-party system. And then two, each of these processes are built in an age that AI or even the internet didn't exist, right? So I think that's the state of operations today. If you look at an average health enterprise, they run about 7 to 8 % of operating margins, 1 to 2 % of operating profits. It's a really pathetic scenario in the United States. Like you look at the 80 % of all the premium revenue for a health insurance company, 80%, 85 % is spent on delivering that care. They're not going to make money. Everybody's margins are getting compressed. We have a structural issue with healthcare and what we haven't done in the United States or around the world even, but especially in the United States, we've been solving a lot of point problems with digital tools and health IT tools and point solutions. We haven't structurally changed healthcare, which is what we intend to

Opto: Yeah. And I was only reading this the other day. I'm not sure where on Twitter, something like this about MRI scanner. So it's interesting you brought it up. Because apparently the cost of getting an MRI at the insurance company, because often that's the case, has gone from 1000 to $3,000 over I can't remember the exact time period, but not a huge amount of time. Which is I mean, I know, you know, obviously, there's been some inflation in the system and stuff. But that seems a bit crazy. Is this the sort of thing you're going to help tackle to improvements custom.

Ganesh: Yeah, in fact, the system is so broken that if you go to an MRI center and say, hey, I need an MRI, they'll say, do you have an insurance adjudication, prior authorization, whatever? No, I haven't. They hey, if you want to just pay it directly, it's going to cost you 300 bucks. And the patients are like, wait, wait a minute. Like, I'm going to request an authorization. They're going to charge $5,000, right? And for the same service that the MRI center will do, they'll charge the insurance company a different amount than what they pay, what they ask the patient to pay if they have paying it directly. And there's a lot of these economics and the incentive networks involved. Contracting done for your provider network at the health insurance company is done at a network level, not just at a hospital by hospital level. There's a lot of group contracting. There's a lot of complexity in this thing. But imagine a world, let me just take an example of an even more impactful scenario. Let's say,

Opto: Yeah.

Ganesh: someone getting a request in a health plan, get a prior authorization request for a PET scan, which is a special kind of CT scan, usually to actually detect ⁓ cancers or any of the other ⁓ pretty critical illnesses. You know that the patient has got a confirmed diagnosis for cancer. You know they're actually getting a request for this thing. But you also know as a health insurance company or a care manager, you look at it and say, I can almost predict how the next six to eight months of that patient's journey is going to be. The radiation therapy that they have to go through, the chemotherapy they have to go through, dietician, counseling, consulting, all of that stuff. You can pretty much look at this whole thing and say, I can create an experience with this member by, you know, like making sure they have an oncology care navigator, making sure they get ⁓ clear details on how much they'll end up paying because it's a huge problem for cancer patients. They don't know. Am I going to get bankrupt? I'm going to get cured, but I'm probably going to be broke. There's all of that stuff. It's impossible to do it today because we treat each of those things as episodes. The request coming in, you're projecting, you're responding to that request. It's an episode and you forget about it. Three months later, the care manager gets a notification that there is a patient that's admitted for surgery, do care coordination. They look at it say, ⁓ cancer patient. They're going through these different things. And then they look at their claims and it's like some of the claims are coming from an in-network provider, some of them from an outer network provider. It becomes a mess. This is the big opportunity with AgentDK. I know, imagine that same experience, a request comes in, you know, a set of agents, a coordinated set of agents goes, intercepts it, goes, logs the case, gets the doctor or the nurse involved to review it, fetches the information to go do it, but also sends an event to the care manager.

Opto: Yeah.

Ganesh: And imagine that patient getting a call that same afternoon and say, hello, I'm your health insurance company calling. I know you're going through a tough time, but let me help you navigate the next six months. Here's a list of doctors you're going to see. Here's the procedures you have to do. Here's how much it's going to cost you. And we're here for you. That's the experience that is possible with an agentic operating layer in health care.

Opto: Mm-hmm. And how did you come up with the idea for doing this? Was it something you experienced yourself or was it just researching this in the market?

Ganesh: Well, look, mean, I have my background, I've been a part of other startups before, I was in big tech for a long time. I was always close to technology, but I also had the luxury of actually working very closely with several industries, including healthcare. 2020, I just exited my last startup and company. And that was, the world was hit by coronavirus and it was, actually, participated, was invited to be part of Governor Abbott in Texas, Governor Abbott's task force for COVID-19. And the state of public health infrastructure was really disturbing and disappointing, right? We were just not ready. And then you unpack that beyond just the public health and the health infrastructure in general was disappointing. So a couple of things happened. Like I spent a couple of years studying health care, spent a lot of time with people. We were going through some illness and sickness in the family that gave me a first-hand exposure of how it is to manage being a caregiver and so forth. I lost a friend to stage four breast cancer. And when she was actually first diagnosed, ⁓ Merck was doing their trials for ketruda, which is an, it's an immunotherapy drug, first in the market, did so well. And she would have been select, she was eligible to be selected for her, their clinical trial and she would have probably survived. We don't know that, but it's a hypothesizing, but the information about her protein expression, the fact that she was eligible was locked in some page 389 of a 5,000 page report that somebody missed. Right. And that sat with me quite a bit in terms of like the first thing when we started to Autonomize was like, look, I mean, there's a larger story here, but You know, we tackle the clinical trials patient matching problem. That's where our first customers were life sciences customers. We worked with organizations like Nova Nordisk and stuff like that. But the big idea, like when you step back, you know, at what in healthcare, historically what has happened is we have solved the symptoms of the problems and the problems themselves. So you find, ⁓ patients are not getting communication. Let's create an app to communicate with the patients. Well, that app changes that communication paradigm, but that information comes back, it still needs to be reviewed by a doc, a nurse, and stuff. We didn't solve the structural foundational problems. So our big idea was like, look, and you look at all of these problems, they all use same core sets of data. They just contextualize differently by different users. So our big idea was like, if you can separate that data content layer with the context layer, the workflow layer, and dynamically pair the two, then you can solve multiple problems with the same core platform. And that was the origin for Autonomize AI. So we're like, look, we can do this. And then we started navigating and finding places where you have collection of multiple problems. That same example I gave you on PriorAuth, you saw PriorAuth, but the charts has got information about gaps in care. The patient hasn't had their flu shot. The patient hasn't had their 13-year-old ⁓ teenager has his eligible for HPV vaccine. Those information are ignored. because you're not looking at it as a collective system, you're looking at it as a siloed evidence, an episode. So now we are able to go expand that across and deliver value for enterprises across. And fast forward today, like, we power three of the five largest health enterprises in the United States. We work with several large, ⁓ complex, really unique enterprises. We also work with several other mid-market companies and organizations too.

Opto: And why did the traditional ⁓ automation software in this industry, why was it not able to solve these problems? Because people have tried to do this before, but obviously it hasn't worked very well.

Ganesh: Yeah. Yeah, look, I think there's a few different things. One is automation is only going to help so much. If you just step back, think about it. Automation is automating what humans already do. So it is just like if you are clicking on a button going from left to right, you're just trying to automate that particular process. Well, what if the whole process that you're actually following is wrong? Automation doesn't know to fix that. historically with robotic process automation, automation software, digital health, all of that stuff, it was all to me, those are all band-aids that you put in because you defined processes in an age where humans had to log into a screen, click from left to right, read information on the screen and take action. And automation, all it did was actually whatever you humans do, I'm going to turn it into an automated software so you don't have to do it. drove a lot of efficiency, no doubt it made a huge difference to the industry, but that's not enough. Because we live in the United States where we have 330 million people and there is less than 1 million MDs delivering care for 330 million people. We have less than 4 million nurse practitioners, physician assistants and folks who are part of the healthcare system taking care of patients. There's no way you can scale. And people are... You know, generally I think there's a trend towards, you know, longer life, but the health span lifespan is extended, health span is actually compressed. Right. So people are not as healthy anymore. There's a lot of chronic disease problems and stuff. what, where automation stops is like, you just do more of the same with automation. That's why it's, it's what this generation will be, or now have the right tools and technology with AI and stuff with agents is to really autonomize those processes and autonomizing doesn't mean. You know, it is making that process autonomous. That again doesn't mean sometimes full automation. There may be a process that has got a regulatory oversight that is needed, a human judgment that is needed. But how do you look at these processes end to end and how do I optimize it in a way that what part of that process should be done by an agent? What process part of the process should not even exist? Can you reimagine that workflow? So like that exam, I still come back to the original example. It is not a serial process of request coming in and you're reviewing it. Well, there is an event that happens. What hundred other events should you trigger based on that event? What judgments that you can make based on data? What judgements do you route to a human to make a decision? So we call that autonomizing. So how do you autonomize that particular workflow? And that's the, you know, that's how we think we can shift the bend the cost curve in health.

Opto: And we're talking about costs. So do you think this is, will it actually get to the end user? Like we talked about this other problem with, which we just described in it about the MRI scan or the insurance companies and how they're charged and these layers of incentives maybe wrongly aligned along this path. So does it actually disrupt that as well so we can lower health costs for people?

Ganesh: Yeah, ⁓ yes, I think yes and no, right? mean, there is longer term, yes. I mean, there's two things. One, there's multiple macro things that are happening. Patients are getting more access tools to educate themselves better. So they are more informed than they were before. They also are, you know, the hard problem is how do you distill between noise and signals? So that's still a problem, but patients are getting more consumerized. They want the experience they see in other industries. They want to see it. And then so there will be some disintermediation of middlemen and incentives and stuff like that. There will be some of that that will happen. But if you look at 300 plus years of history, we've entrenched the system in the way we deliver care, how people get recruited. It takes 10 years to produce a doctor, right, in the United States, right? How do people get into the workflows? How do they work? Who do they get paid from? all of that incentive structure employers like you know private employers are ⁓ one of the largest sources of in like that the government pays for half of the healthcare medicare medicaid and then the other half is actually private employers so it is going to be hard to pull all of that out and just say i'm just going to go direct you to patient right so i think there is a there's a way you think about it what is also great about the US there's a lot of things wrong about the US health care system but one of the things that's great is It's actually the entire US system is very capitalist in nature. So if you really want to make a difference to the healthcare industry, as they say, follow the money or optimize the flow of that dollars, right? So our focus at Autonomize, we like to say like, we want to autonomize the business of care. So it becomes, you know, the simple mantra is like, make healthcare more profitable for healthcare enterprises because it's not today. Make it... really seamless for patients because, you know, end of the day, they want to make sure they deliver care, they get care where they want to get it from the person they get it and so forth. And more importantly, the most forgotten people of the whole thing is the actual caregivers, the doctors, the providers, make it joyful for them to deliver care. Today, it's a nightmare.

Opto: Yeah. So scratching the surface, really, of how big this problem is. But we've got to start there. We've talked about agents quite a few times. It's possible. This is like a buzzword at the moment, but it's buzzy because it actually has real value. Could you explain to the audience who might not know so much about what an agent is, maybe more specifically about how you use it at Autonomize, but what is the agent and, you know, Why is it good businesses to good powers?

Ganesh: Yeah, no, so an agent is, you know, it's a very, it's not a, it's actually, it's gotten some buzz here, but it's not a new word. It's not a new thing. It's basically a self-contained autonomous piece of software that you can optimize around, hey, here's the goal for this particular, the function that you need to perform. Here's the tools you have access to, to go perform this, you know, these functions. And, you know, here is what you're trying to optimize for, right? So you give them some guardrails, you give these agents. a goal or a task, and then it goes and performs this and operates as an independent unit. What's what is important to understand is like traditional software versus AI. The big difference is determinism versus probabilistic nature, right? So traditional software, would write the rules and say, here's what if this happens, do this. If this happens, do this in AI, the nature and especially with language models, world models, diffusion models. Recently, all of these different architectures allow you to give ⁓ an algorithm a goal and then some limitations and guardrails and figure out the best path to doing that. And that kind of goal-oriented optimization is how you scale, because traditional software cannot scale. You cannot produce enough, write enough lines of code to go do that. So agents are, think of them as autonomous units of work that can be performed and you instruct it with a goal. with ⁓ some access to certain tools to perform its goals and some guardrails on either safety or what they can and cannot do. How should the output be? How the input should be taken? How do you respond to things? And so we put some instructions, but you largely let those ⁓ agents or agents are basically a software instantiation of a large of a language model or kind of model and they can just perform work and deliver that. I'll give you the same example. Like if you're looking at a

Opto: Hmm.

Ganesh: If you're a doctor and you're spending time talking to patients, but you also have to write down what happened in that particular thing. There's a listening agent that'll just listen to the doctor patient communication, package it, writes the notes in the way that the doctors will A, get reimbursed on if that goes to an insurance company for any claims or approvals. B, it has all the relevant clinical information that is not missed out in that conversation. And then, you know, after that, there is a missing information. The doctor says, okay, I'm going to call your nearest pharmacy and put this prescription on. It automatically takes the cue and delivers that. So those might be one or multiple agents that can work in concert. That can be a great augmenter for that doc. You're having that conversation with the patient, right? As an example, like that's, those are the kinds of things that we talk.

Opto: And it's obvious now that agents will automate a lot of processes moving forward. And the question now is, at what point is it important to have the human in the loop? And how do you help make those decisions?

Ganesh: Look, think, no, you look, there's a, the way you think about this, the intuition here is like, look, most of the agents today are built on language models. Language itself has limitations. Language is very powerful, right? Humans don't make decisions based on just what you read. It's how you feel, you draw back on your experiences. You reference very quickly the last 10 times you've had the similar conversations. You know, if piece of information doesn't have what you're looking for to go look elsewhere, agents don't have all of that, you know, unless you define that this is what you need to do. Right. So it's a very interesting problem where these are powerful pieces of software. That'll be. If you can direct them well, if you can actually, you know, put the scaffolding and go drive it, it'll be extremely powerful. Now there are certain domains like. coding, math, places where the boundary conditions of solving the problem is very finite and very, there is only so many variations into that thing where you can go fully autonomous very, very quickly with these agents. And we are seeing that, like you can go to cloud code and write up instructions and it produces a website. And if it encounters an error, it'll go rework it and doing it because everything is contained within that. Now, does that work in a production environment in Fortune 100 enterprises? Probably not. You have to then look at code review and make sure it's actually more enterprise production grade code. So there's a lot that limitations. But the line you had to think about is like, if your workflow or the task you're performing, the work you're doing, involves a lot of human judgment, that could be based on the fact that you have to draw on human experience. Healthcare is a classic example. ⁓ regulatory thing, I'll come to that. But then healthcare is a classic example. If you have a doctor's note and that's summarized by an agent, the agent is going to summarize it based on what's in the note. But a doctor, when you look at the note and he sees that, this note does not have the BMI information that I need to go make a decision, they know it's not in there. And you can train all the models you want, but the fact, because agents are optimization engine, that's going to optimize for what's in there. But the doctor will know that I have to go outside of that, right? So it's a hard domain like that. In a place like healthcare, knowledge is not written down. These language models require the knowledge to be written down so you can train models on it, or you can fine tune the models on it, or have access to those things. Most of those things, if you're in a health insurance company and there is a request for a wheelchair approval for a prior auth or a claim, they'll call Sally who's done that for 30 years and say, Sally, how do you handle this? And Sally will tell you how. An agent is not going to know that unless it resides inside Sally, right? So we'll get to a place where it's going to be hard. So human in the loop for, and that's one part of it, like problems that require human judgment bring the best out of humans. Use agents for the mundane stuff. The other part is regulation, right? For the same prior to the example, you should not, the government, the CMS will not allow you to approve or deny claims or prior authorizations. completely with AI without an MD oversight, without a knock oversight. So there's regulatory reasons that you want to make sure that there is a human in the loop in there. The big unlock, Ed, is not to look at an absolute saying it's agent, autonomous versus non-autonomous versus human in the loop versus fully autonomous. mean, those are honestly very academic, those discussions. The real problem is like, depending on the problem, how do you layer an infrastructure? where you can go from, if this is a problem that I can do fully autonomously without human oversight, or with just a slight oversight or off-cycle oversight, then I can just automate, I can autonomize the whole thing. Or if this is something that requires the humans to be brought in at the right points and stuff like that. So you need an operating system to go make this happen, and which is our big focus, right? We believe, know, if you break down any of these processes, it's not one giant process. There's multiple things happening across. And how do you go determine what process to be done? And then also in the same process, in the same way, not just completely believe that the way it's being done is the way it should be done. How do you optimize for that? So the big unlock is going to be having an infrastructure layer that'll understand the nuances of the industry and the domain, and then optimize whether I'm going to go fully autonomous or deliver autonomy for a group or a group of individuals with the right tools around.

Opto: And you mentioned you're in some of the biggest medical practices in the US today. In practice, where do you believe you're driving the most value today? Just to simplify it for people, is there a couple of areas that you really transform?

Ganesh: Yeah. Look, I think, you know, we took a very different approach when we started the company. And this is like, you know, if you ask, you know, healthcare last, I like to say this, like the last 10 to 15 years of healthcare in the U S or around the world, digital health as an investor category did well, health IT did well, health services did well. But we still, if you ask a normal patient saying, has healthcare gotten better for you? The answer is no, it hasn't. It's gone worse probably. Right. So we took a very different approach saying, look, we used to go and tell a health plan or health insurance company saying, hey, we can help you ⁓ orchestrate better cost savings on care management, your cost of care, your claims, and your prior auth And they will ask us, like, are you a prior auth company? And we'll say, no. We can do prior auth and all the other prior auth companies out there, but our real value is when you compound across workflows. And so the value that we have delivered in certain areas, like, of course, prior auth is a huge part. It's the huge, the largest source of friction for patients and providers and health plans across the United States. Right. So that obviously is a huge part that we enable that for several organizations, both on the provider side to make sure they collect all the required information that the doc doesn't ever spend 30 minutes or 40 minutes collecting and collecting all that information to submit an auth to going to the health plan and helping them navigate this, compress it, compress the time that they take to go. and solve that problem. So that is one area. But the large source of value in ROI or the, you know, that we provide for our customers fall in three buckets. Number one is shorten the cycle time, right? And the cycle time shortening will unlock, you know, better member experience, better, you know, lesser provider abrasion, all of that stuff, right? So how do you compress that? You know, we just launched a case study yesterday with a company called, with a healthcare provider group called Altace. We compress their, case decision time, 63%. It was the last six months of being live with them. And so it's an incredible value. I imagine the benefit for the patient. I pull out a question, I'm getting hearing back 60 % faster. The second big part is cost savings. And this is hard and soft cost savings. And usually it's things like, if a clinician is spending 40 minutes doing paperwork, compress that to four minutes, do all the work for them.

Opto: Yeah.

Ganesh: presented to them so they can make ⁓ any adjustments and go forward. We've been able to deliver things like 36,000 hours of clinical FTE hours a month kind of savings in one of the largest health plans. And then the third part is actually, this is how do you bend the cost curve for healthcare, right? And the cost curve is, it is basically better coordination of care. It is making sure that, look, if I'm catching a patient who is in the hospital,

Opto: Yeah.

Ganesh: before they get released. So they know what they need to do after they get discharged. Where do they go? What is their next step of continuation of care? I cannot just say money here, but I can lower the cost of care. That example I gave you on the oncology request, right? On a cancer request. If you know that the group of in-network providers is what this patient should use and you help them navigate that, a cancer treatment is not $340,000 for the insurance company, it's $120,000. So there is those kind of, again, the business of care can be optimized around it. So those are the three vectors that we see consistently across. And workflows wise, we do utilization management, we do care management, population health, claims, revenue cycle management. So everything that touches the different business aspects of delivering care and managing care.

Opto: And something that people, you know, they hear AI and immediately a lot of people get scared. How do you protect against the agents making the wrong decisions and things like this?

Ganesh: No, it's a huge problem. I'm sure you probably had a guest talk about the OpenClaw experience that happened a few weeks ago. So OpenClaw, so I mean, here's my perspective on it. So OpenClaw was, from a developer perspective, the fastest growing GitHub project ever in the history of GitHub. It's incredible.

Opto: I haven't I didn't know that.

Ganesh: What it really did, the insight there was like, you know, the ability to actually take, you know, create a personal agent with WhatsApp or any communication mechanism as the way to interact with the agent that has got access to all your information, that's where they went wrong, and then be able to make autonomous decisions on your behalf, right? Now, lots of things wrong with it. The architecture was not perfect. The code quality was terrible. There was like so many attack vectors opened up and stuff. But the core insight in that project, we studied it. in fact, there's an element of the same idea. What they did was they separated the model, the intelligence layer, like whether it's an anthropic or an open AI, to access to your information, your unique context, your unique memory of things. They separated the interplay of the two. But today, what do people do? They get to open AI, use an API, shove all the data in there, ask it to make a decision. You have no idea. But you have more control when you do this. And that was the idea. Of course, they implemented it not really well. So I think what that, know, the scary part of this agency, what they did was actually they didn't have the right guardrails on it. They didn't have the right layers of, you know, segmentation of what kind of data should be shared versus not and so forth. But the framing of the idea was actually a very unique innovation, if you think about it. So when you. Think about deploying AI and AI agents and enterprises, which is where we are focused on. You want to put in, can do guardrails in multiple different ways. One, you can ensure that everything goes through a validation cycle that spans not just the engineers looking at code, which was a problem that in there, the regulatory folks, we just recruited a chief regulatory officer to actually come and help us put up a playbook for that to help our customers navigate that. ⁓ the clinical teams are actually involved in actually that process. It's a team sport, right? So there's a lot of that that you have to do on the foundational layer. Then there is things like, look, everything an agent does, how do you build in traceability, source it back to saying, if I'm giving you a summary, what is the source data that I use to go get that summary? Defining what decisions should be made by the agent and should not be made by the agent. And baking that into the agent instructions to make sure that if you are faced with that choice, They're going to reach out to a human, get them engaged. They changed the whole, the open clause, big insight for us personally was that they changed the whole system to an event driven system and they separated the intelligence layer. So a lot of our customers ask us, for example, can I now just use Claude to do this? Or, you know, ⁓ well, why do I need to automize? Right. And the answer is actually very simple. Look, I mean, if you can just prompt engineer a way through solving all of these problems, we'll probably get there at some point. That's great. but you lose complete control, right? Because you're going to take your data, throw it to Anthropic. And by the way, Anthropic has been amazing at actually starting new businesses to take down SaaS companies that are in that particular thing. So if your healthcare enterprise, you give your data to Anthropic and that's doing it, next thing you know, they're going to build an AI powered health enterprise after your business, right? It's like the Amazon marketplace story that we had going on for the last 10 years or so.

Opto: Yeah.

Ganesh: So I think there is a unique context and healthcare is very unique. It's very special as an industry. There's a lot of nuance in each organization. There's a lot of nuance in the way you treat different conditions. There's a lot of nuance in how you manage different kinds of populations. There's regionality, it's not a one size fit all at all. So what you need to be able to do is know when do I use intelligence and how do I preserve and store context that is relevant to using that intelligence. So technical capabilities exist. to solve that, you and we built a platform, for example, to help our health enterprise customers are building agents on our platform, still may use one of those foundation models and premier models, but then that gives them the enterprise scaffolding, the regulatory scaffolding to make sure this is not going to go rogue and it's going to actually be, you know, within the constraints or the expectation of a regulated industry and so forth. So it's hard stuff, but you know, it's, it's, I, think it's well worth it.

Opto: And obviously today you've mentioned regulation. You're building within the confines of how regulation is today. Do you think regulation is going to change in the near future? Is it fit for purpose in this new era of AI? I assume not because it's based on how old world worked. What do you think is going to change? Is there something in particular you is going to change in the near future or not?

Ganesh: Yeah. Yeah, no, think I'm hoping it will. And I think we're seeing evidence that is going to change. It's interesting. Like I said, 50 to 60 % of health care cost in the economy is borne by the US government in the United States. Medicare, Medicaid, Medicare Advantage, and so forth. And so they have a huge incentive to actually do it. And we've been really pleased with this current administration's effort. They're thinking beyond just I'm just going to go have a lobby or a body to go do this. Like too big, like the fraud, waste and abuse initiative that we just heard from administrator Oz yesterday. We were part of that meeting and then ⁓ we listened to ⁓ Dr. Oz and he was like, look, mean, this fraud, waste and abuse is a huge problem. How do you go solve it? There's a lot of these wrong coding, billing and stuff like that. We're helping our customers solve that problem at the enterprise level to say, make sure like before I pay out a claim. make sure you verify the medical records meet those standards, making sure that the right code is actually maintained. All of that happens would be called prepay before you pay the claim. So you're not just paying for it and then going and trying to collect the money back, which is a huge problem in health care. So we're seeing that. We're seeing several of this ARPA-H projects that are actually being launched into the ⁓ Advanced Center, the Government Center for Health Care. And ARPA-H projects are on different things. And there's one program that recently got I think it was called the access program. The economics of that program will only work if you embrace AI to build and deliver care. So to me, it's a huge opportunity. And then the administration is actually doing all they can to muscle through it and making sure that they're going to force the industry to change. And I think it'll be a huge and very relevant, very critical lever, which is also why we invested in having a chief ⁓ regulatory officer in our company at this stage. I think the regulation will change. We're seeing some, you there's two parts, right? There's also, you know, health insurance companies have, they're the money holders of the industry. So they are ⁓ incentivized to change the behavior of the market too, because their margins have compressed, right? So what they're doing is there's a group called AHIP, for example. So the ⁓ American Health Plan, something. So ⁓ the AHIP organization, is there's a group of 40 plus health plans that made commitments and how they're to improve things like prior to for patients, the experience for patients and providers. The administration is doing its part. I the private industry is doing its part. And more importantly, I think the best part about healthcare today is the average patient has got a lot of tools to educate themselves, to ask intelligent questions. ⁓ like Dr. Chad GPD is always there for them, right? It may not be all accurate, but it... raises the level of fairness for everybody in the system. So I think we're in this period. I call it the ⁓ of healthcare, right? This is the Kairos moment in healthcare.

Opto: Yeah. And yeah, I definitely agree with you. It's a golden period, seems across a lot of industries, apart from the ones that are probably getting disrupted quite heavily by AI. But when you're selling into companies, I'm assuming, and maybe I'm wrong, there is, you know, a lot of people have quite a lot of reservations about AI coming into their systems and deciding things. How do you get over those barriers? Is it becoming easier?

Ganesh: Yeah, no, there's a lot of noise in the market, right? I there is like, I like to say there's two kinds of folks starting healthcare AI companies. There is folks who been in technology and they think they can actually come and, ⁓ I can just do this. I've done it in payments. I can do it in healthcare. And they all come in and get a root shock of how healthcare operates. great tech. mean, tech is commoditized. So it's about trying to make this happen. And then there's a lot of other folks who are like coming in, new companies started from healthcare executives and healthcare folks who think they can actually just throw AI at a problem they very well know, and they can solve this thing. The reality is both those camps are not, you know, there's like one camp promises a lot of ROI and never delivers because, you know, it's, that's not the way it works. The other camp don't know what, what they're getting into and they get hit in the face and the moment and go like, what do mean? It's going to take me 18 months to have ⁓ a, a contract with you. Well, that's the way healthcare works. So what we have done, for example, is assemble a team that is cross-functional. I have folks who were like deeply on the technology side and deeply on the healthcare side. People who have run health plans, people who have run hospital systems, people who have run ⁓ divisions in FDA. So we have assembled a team that can pull the whole thing together. But I think that one of the major things that we actually go in with is like, it's... Healthcare, I mean, AI in general, there's a lot of noise, there's a lot of activity. One CIO told me like the best part about AI right now is like everybody thinks they can build, which is great for an technology organization. But there's a big difference between doing a great demo and running things in production and at scale, right? So it's very simple. go in, like we don't do a lot of pilots with our customers. We tell them, look, here's our results that we've delivered for, you know, multiple organizations, including the top three out of the top five in the United States. And these are complex enterprises, multiple systems. And it's not magic, right? I try to think AI is not the model. It's like to get to value. I used to call it like, know, there's the usefulness factor in AI, right? You can build a chat bot, but if it's not going to change anything, it's not going to drive better cost savings, better experience, any of that stuff. It's completely useless. And we approach problems that way primarily. So the reality is, even if you build a chatbot, you have to make sure it integrates with the right systems for people to use it. You have to make sure it has access to old school legacy systems and systems of record across that entire enterprise to make it useful. So there is a big jump between I'll instruct a model to create a demo to how do you really realize value. And our single biggest vector that we've been actually very successful with is like we go with the mindset of how do I deliver you ROI? Three to five X ROI within the first six months of engaging with us. And we've been consistently able to deliver that despite the delays we see, despite the legacy vendors or HR vendors who will put all kinds of roadblocks to actually delivering value, we still get them there because a lot of things in AI is also problem selection. Solve the hardest problem. It's not that difficult, right?

Opto: Mm-hmm.

Ganesh: Don't go solve a chatbot easy problem. Now, solve the hardest problem. And then you think through it. There's a little bit of design thinking on how do you think about it. Then you have the technology infrastructure which we built a platform that allows you to rapidly compose a particular workflow with a library of existing agents and then pull it together to go deliver value. So there's a lot of different things. You've got to understand it from a business value lens. You've got to understand ROI. What are the levers you have? And then do you have the right tools to pull it all together? So that's how we've been actually pretty ⁓ fairly successful in our journey with healthcare enterprises.

Opto: Yeah, I that's great. How long you been around now?

Ganesh: We are exactly four years old as a company. We're venture backed. We did a series a last year. raised about, we've raised about 32 million to date. We've been extremely capital efficient in general. ⁓ We tripled our annual recurring revenue two years in a row. I do think that there is like a 10X opportunity this year that we're seeing a lot of traction. The big thing that I would say is like, look, healthcare is very unique and very different. We didn't want to be a general AI company.

Opto: Bye. Yeah.

Ganesh: We are a healthcare native. We built for healthcare. And for that, it is like the story of Chinese bamboo, right? You got to sow in a lot. You got to actually pour the water. Like the Chinese bamboo story is like, ⁓ you if you have a little seed for bamboo seed, Chinese bamboo, you put it in this thing, you wait, you pour water, you wait for it, you wait for it, you wait for it. Three, four, five months go by and it doesn't even pop up. Then there's a little sapling that comes up. And then six months later, within the next two or three weeks, it shoots 40 feet.

Opto: Yeah.

Ganesh: Right? And it's incredible. Healthcare is like that. You got to put in a lot of investment. You got to make sure you're connected. We built hundreds of connectors to legacy systems. We built ⁓ foundation models that are fine tuned and ingrained for very narrow, nuanced problems within the healthcare dataset. We built an ontology and a knowledge graph that actually can cover ⁓ population health, care gaps, heaters, ⁓ all of the coding guidelines and all of that stuff. ⁓

Opto: I didn't realize. Yeah.

Ganesh: A lot of the foundational work is done and late last year, we opened up our platform for our customers to start building their own custom workflows and agents on it. And that's what really drove a lot of unlock for us. we're off to a great start this year in 2026.

Opto: And looking further ahead then, if you, you know, your best possible sort of outcome over the next three, five years, where do you think we're at? Like, what's changed? How does it look like in the future?

Ganesh: Yeah. Look, I mean, there's a, there's a, so it's a very, I mean, in the age of AI, it's hard to predict anything more than three to six months, right? But, but, but I think, you know, structurally, there are a lot of things that won't change, right? You're still going to take care of people in healthcare. You're still going to need caregivers and doctors and physicians and folks, because healthcare is not just about the data and the software and the pieces. It's about delivering care. Now we like to say that like, you know, AI should run healthcare.

Opto: Yeah, for like two weeks and a bit.

Ganesh: the business of care, the operational layer, so humans can heal because that's what they do. Right? So if I fast forward two to three years, the future we want to build and we're helping our customers build is a hybrid health enterprise, the agent health enterprise where you have humans and agents working in concert. It will not be a completely automated environment. It'll be autonomous in a lot of different ways and workflows, but the operational layer, how you move data around, how you make know, collect information from multiple systems, how you communicate, how you collaborate, all of that will be done autonomously with agents. And where judgment, guardrails, know, regulation, compliance, all of that stuff will be oversight, will be delivered by humans. And a simple way to think about it is we walk into a hospital, you know, you don't have to go and spend, you know, I don't know whether after you move to the to the United States, you've gone to a doctor's office, you're spending 20 minutes filling forms as you go in. None of that stuff. You walk into the thing, you see a doctor, you get an experience, you walk out without having to think about, now I have to call my pharmacy to get the medication. Do I have to go pick it up? And then there's physical AI, there's robotics coming in. So I think we're in this age of abundance that it's very clear to a lot of us that it's actually coming. It's inevitable, but it'll require very careful consideration on how we shape it. for critical regulated industries like healthcare. And I think we see a world where like it'll be a hybrid enterprise. like if you're a health, if you're ⁓ an employer like Dell or FedEx or one of these large Amazon for that matter, today you depend on ⁓ multiple brokers and insurance companies to help manage care for your employees. Or if you're a health insurance company, you want to start a health plan, you should be able to do it with one risk actuary, a few doctors. 10 nurses and an operating layer from Autonomize AI, right? That's the future we want to build. And I think we're well on our way to do that. And we're helping our customers get to a point where, you the clinic of the future should have, you know, like a lot of very meaningful people who are having meaningful interactions, but not a bunch of assistants who are just doing paperwork, right? So all that roles will reverse all of that, you know, expectation because, you know, the, it's not going to be static as it is right now. people are seeking care from chat GPD right now. How do you give that experience in an actual real environment? So there's a lot of things that's gonna change. And I think we're super excited about the future we're

Opto: the healthcare operating system everyone's going to know about.

Ganesh: the agent, the healthcare operating system, that's right.

Opto: Thanks, Ganesh. I thought if you're up for it, they've got a lightning round just to finish, which is just five quick questions, not looking for a long answer, just to wrap up the interview. So number one, biggest myth about AI agents.

Ganesh: Okay. The biggest myth is like, you know, an agent is like a flip of a switch and it just works. It requires engineering to what goes in. It actually requires you to shape it the way you want. So it just, it's not a light switch.

Opto: that's a good one yeah. One workflow that will be mostly agent run sooner than people expect in health care.

Ganesh: So I would say anything that is purely administrative in nature. So revenue cycle management, claims adjudication are two examples like that.

Opto: One workflow that should never be fully autonomous.

Ganesh: clinical decision support. think it's going to be, like there's the argument that you can actually have, hi, there's no doctors in East Africa, so you should have an AI doctor delivering that thing. Well, what you should be thinking about is like, how do you make the doctor in the United States or in London have the ability to go reach that through digital means so you can deliver care? Because healthcare is not just about the data or the interaction, it's about people delivering care to people.

Opto: And finally, one red flag when evaluating AI vendors. Any AI vendors.

Ganesh: Yeah, ⁓ there's a lot of red flags, but I would say, you know, challenge the claims, right? Challenge what the claims they're making. Everybody can make up a lot of stuff right now, including like we actually ask our customers, like we have published case studies. We actually ask them to call our customers and talk to them and stuff. And I think there is, you know, a lot of engineering between, Hey, I want to do something to actually doing it. Right. and doing it at scale and doing it well.

Opto: That's a really good point, actually. They can also make up a lot of stuff and there's a lot of fluff that you can put behind in marketing, but you know, have actual customers.

Ganesh: Yeah. Yeah. And it's easy to sniff out to, right? It's easy to sniff out. You just have to ask a few questions like, wait, tell me more about it. How did you do it? And then you'll find out that any agent take a workflow that most people describing either a rag workflow, which by the way is great for a search use case, but for nothing else, or it is a chatbot use case, like as a QA on a document that is being there. It's like, ⁓ that's great. How do you store that data? Where do you store it? How do you know this actually agent is making the right decision? Ask a few questions. You will. It'll completely disintegrate very, very quickly if somebody's making a claim.

Opto: Well, Ganesh, thanks very much for your time. It's been really good to go through healthcare and AI, the new OS that everyone's going to be looking out for. Really appreciate your words.

Ganesh: Thank you, thank you Ed. Thanks for having me and I hope like, we can together build a great future for, you know, healthcare for all. All right, cheers.

Opto: Yeah, I'm looking forward to it. Thanks again. Bye.