Healthcare AI Innovation with Ganesh Padmanabhan, CEO of Autonomize AI

as an industry, we're probably the furthest behind using the data to better the human condition, right? Because you can blame it on regulation and stuff like that, but it's also because there's incentive models. You to understand how, you know, how healthcare works, who pays for it. There's so many different nuances into it. And our key focus today is about like the biggest way you can impact healthcare and like as an industry is if you change the economics of driving the business of care and that's what we focused on at autonomous, right? And I can unpack that a little bit more, but you know the idea that like you have to make sure that the right care is delivered to the right person at the right time. Hi, my name is Demetri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to

the AI agents podcast brought to you by Jot Form and featuring our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello everyone and welcome back to another episode of the AI agents podcast. In this episode we have Ganesh Pabmanabin from Automize AI. He is the founder and CEO. How you doing Ganesh? >> I'm good Dimmitri. Thanks for having me and um it's great to be here. >> Nice. How's that out there in Austin right now? Pretty good. >> It is pretty warm and I think it's going to be like 105° Fahrenheit today which is perfect summer for us. >> That's perfect. Okay. Um, well, not for me. Uh, uh, Chicago's actually got, in my opinion, what is good weather right now? It's like 60 70°,

which for us is kind of crazy. Usually it fluctuates between cold and colder. Um, but weather aside, let's talk about AI. Uh, appreciate you for making the time. tell us a little bit about how you got your start in uh AI and um how you got to the place to even you know start up a company like uh Autonomize >> you know uh thanks for having me again um excited you know I've watched a lot of your shows so it's really exciting to be here and u to talk to you about this and talk AI and AI agency so how how did I get into AI I mean through math probably right I mean most people did >> fair enough >> it's it's um I've been in the industry in and out of data, AI, machine learning, all that stuff for 15 to 20 years.

And the joke at the house is like my 11-year-old tells me, I'm like, "Dude, you know, I've been an AI for all this while." And he goes, "Dad, you must not have been very good because only the last two years it has taken off." >> And then the the reality is, but that's the that's the that's the thing about AI today, right? Everybody's talking about AI and agents. It's like a you know uh a huge moment that took 20 years 30 years to get here. So uh I got into AI initially through data. So I was actually spending a lot of time my initial part of my career was in in corporate. I was at Intel and Dell for a long time >> at Dell. Atlanta ran a pretty large business and one of the you know was the converge infrastructure business was at that

time the youngest billion dollar business that we built was a GM for it and one of the fastest growing part of that segment was big data at the time and this was the days of Hadoop clusters SAP HANA analytics and all that stuff and there was a definite trend in Mman where like you the the you know there was always there was this um the graph that everybody used to talk about where the amount of data produced in the last two years uh is double or far exceeds the amount of data that's produced ever in history. >> So it was very evident being in the midst of data that with all the amount of data that we have we need better tooling to analyze and understand what the hell is there in the data. >> So that's how I really got into early and it

was through machine learning. It was like statistical analysis, statistical models, regression models and so forth. Um I ended up you know after I left Dell after 11 years uh started a company and explainable AI. So I was my skill stack earlier on was like how do you go and help large enterprises really demystify some of this technology and build something meaningful and AI one of the biggest blocking points at that time it's a it was still new 201617 b you know for enterprises to adopt it so somebody has to explain it can't be a blackbox so I built an explainable AI company a little ahead of the time but one thing led to the other there was a later stage company called cognitive scale which was one of the original OG's of AI was founded by IBM Watson's original team >> and >> okay >>

I ended up joining them as a head of growth for the next few years so got really deep into AI we worked with several of the large fortune 100 companies across multiple industries have a lot of battle scars on my back even today of trying to launch AI in production and um packaged all that up and did another startup after that uh focused on data readiness for AI called >> Molecular >> and then founded about you know postcoid it was very evident that this you know um humans have far exceeded our capability to understand comprehend and make sense of all the data that's being you know uh generated around us so you need an augmentation layer uh of AI of human cognitive functions like being able to look at images and understand that reading text and comprehending that reasoning across multiple data sets um and

And no other industry was more hurting because of the lack of that capability in a highly constrained industry like healthcare. And that was the original story for launching autonomize AI about uh three and a half years ago. >> Interesting. Okay. Yeah. And um you know I guess just to speak back to your experience a little bit. You said you had some battle scars in there like what what was um what were some of those difficulties that you had uh early on? So you know look the the first generation of AI was about like how do I predict not the next token but the next number or digit right it was classification your classic machine learning models the biggest problem that we we still understand it because the those models were a lot more rigid in terms of its uh ability to deal with the variance

in the data itself right because unlike the more language models which are more understands the human nuance and stuff like that. There were you had to pre-process a lot of that structured or semi-structured data into you know the right data engineering output for these models to actually go and and put things in production. So the popular use cases in the 2000 mid 2010s in with AI were things around digital marketing segmentation analysis your next best actions um all of that stuff where you're predicting an action or you're classifying information and so forth. all of those problems when you go and like when it runs great we didn't have notebooks at that time but on a on a laptop right um uh it it runs great on a spreadsheet and you do a regression analysis but when you get in the real world that data was

so much more varied you had to do so much more pre-processing on that data to make the machine learning models absorb it the right way to give you the right output so those were the class of problems then wherein you know most organizations and we work with like several big ones including like JP Morgan Chase Goldman Sachs and stuff where it was great to go into a pilot and solve a problem, you know, including some large healthcare organizations back in the day, the initial problem, initial pilot always looked great. >> Yeah. >> When you go into production and the entire thing, it's like the Mike Tyson statement, right? >> Everything everybody has a plan until they get punched in the face. >> Punched in the face. Yeah. >> Yeah. And and the most of the calories that you have to spend to make sure that

it worked. Was it on data readiness? you know uh things like it's different problems today but those days the big problems were like you go into production and then you realize that your model doesn't conform to the shape of the new data the drift in the data is way too much for the model to be useful a lot of caring and feeding that you have to do it was so you know unforgiving for machine learning systems to actually handle the variance of the data that's changed now and now we have different problems >> you know um yeah I feel like we have different problems, but maybe it's there there is kind of more of the same of that. People still have that concern of like uh data integrity and accuracy. And my question is, you know, you're getting into an industry that a lot of

people may be a little bit afraid of um in healthcare like right now. And you know, because there's so much I mean, I think healthcare I used to work in uh paid ads uh prior to to doing what I do now. And um one of my clients was a healthcare company. >> Compliance, compliance, compliance was all I thought about. So what what was it what is it like and how was it like getting started in in healthcare with AI because I can imagine there there's probably some interesting roadblocks uh and like hurdles to come over there too. >> No. Yes, there is. I mean there is well two two things right when you're building a company no matter how hard you try it's going to be hard right you're always going to have to put in the work you're always >> so the big thing

for us was about right look you know while we're doing something but do something meaningful that has got lasting impact makes a difference to the world all that stuff so that was like healthcare right after co it was almost um I'm a engineer technologist And it's almost a sin not to apply all that expertise and experience into something that we just saw during COVID how the health crisis can completely bring populations down, governments down, you know, the entire human global ecosystem down. >> So it was almost a sin not to apply all of that experience and expertise into something that makes a difference. So that was one thing. Now >> to your point on, you know, yes, healthcare is different. It's very different and I think it's very different in so many different ways. one >> you know it is very you know the the

the value in health care is you know the the re healthcare works because of the people that's in healthcare right and it's a huge nuance that people don't understand wherein you know you can oh I can just automate this particular workflow well you can automate any workflow any which way you want >> sure >> but the big part of health care is the care side of it and the care comes from the human touch right >> um so there's a lot of that nuances in terms of you have to understand you have to get deeper into the workflows and so forth in terms of compliance and all the other things right so every industry is built or like healthcare being one of the most regulated industries there's a reason we're in there now there's argue you can argue whether you need a lot lot of

that regulation should it be opened up so that aside right you have it's a it's a it's a constrained system because of regulation because of rulings because of aduranceances and because of the cost of compliance all of that But to me that is all like constraints you know drive innovation. So you have to just engineer the solutions for that problem. So things like >> Right. So so big things like you know we focused on right day one and one of the thesis of the founding thesis of autonomized AI is like you cannot retrofit um a vanilla horizontal platform for AI or a horizontal model for AI into healthcare because it doesn't understand the nuances. It doesn't have the right scaffolding for compliance for enablement for building trust with the users you know c like you said accuracy versus you know uh variance of like you

know data. How do you actually be less forgiving on the things of data? How do you make sure the quality of the output is always good. All of that constraints were like engineered in and baked in in the way we thought about healthcare. Right? So we don't do a lot of work outside healthcare. We just focus on healthcare and we built it from the ground up in healthcare. Understanding the nuances, the workflows, the data types, the users, how healthcare operates. You know, you know, if you really think about the there's so many problems in every industry, but in healthcare, the biggest thing is like healthcare as an industry produces more than twothirds of all the data in the world because most two-thirds of the data >> really okay. There's a and it's it's I think they also count all the variables and all the other

data which has completely exploded over the last you know 10 years or so. >> Sure. Yeah. >> Um yet as an industry we're probably the the human condition right because >> you can blame it on regulation and stuff like that but it's also >> because there's incentive models you to understand how you know how healthcare works who pays for it. There's so many different nuances into it and our key focus today is about like the biggest way you can impact health care and like as a as an industry is if you change the economics of driving the business of care and that's what we focused on at autonomize right and I can unpack that a little bit more but you know the idea that like you know you know you have to make sure that the right care is delivered to the right person at

the right time you know and they get cared for paid for the right way and all of that stuff because if you drive and optimize for at and you lower the admin burden, you lower the cost of care, you increase access to care, you increase the experience, everybody wins, right? And you cannot fight the system of oh this shouldn't be this way. You cannot have a you know capitalist economy running healthcare. It has to be nationalized versus not. All of those are different models, different constraints. >> But I think you know the opportunity we have in front of us especially in the age of agentic AI right now is to how do you go and rearchitect the operating system needed to deliver care. >> Yeah. So it feels like um you touched on a couple different points there where you know we can talk about

the how you are impacting uh healthcare itself. But first, I I think maybe a more specific thing that we could get into is the the the nature of uh how your AI is maybe built to be more specific cuz uh a lot of the concerns that people have about quality. It seems like most of the answers that I'm getting from founders here is it's like well we're building this from the ground up catered towards X use case or X industry, right? Um, so that seems to be a consistent thread and I and I just want to pull on that a little bit more. And it's like it's just interesting to me that the the concern is obviously there and and rather than like you're saying people try to be general, it seems like a lot of people are realizing, okay, well, we'll just be more

specific and that's the constraint we're working with. We'll pick our vertical. And um, how, you know, are you seeing uh, in in the tech though? because I I'm just curious generally speaking um it improved to be more general, right? Because it seems like at this point I've only heard over and over again it's like well we're trying to be specific because we want it to be good. Are you seeing though that there's like a trend that is getting better generally as well at the same time though after you already like I don't want to say pigeon hold yourself but you know what I mean. >> Yeah. No, you know it's a it's a very interesting question and this has been I think it's going to be the debate for a while in the industry on like you know >> Yeah. general or >> yeah you

know and and the whole idea of AGI right and and >> definitions can be you know very unique to how people look at it instead I don't think it's going to be one AGI in general going to be multiple AGIS in the industry right you're going to have industry specific AGI but I think again the definition of what is general versus specialized kind of thing I think that there's two ways to think through it and here's how we think through it >> one is you got to solve specific problems that deliver high quality output So be it accuracy, be it latency, be it timelines and stuff like that. So you know in the 1960s AI conference, the big Dharmat conference that coined the term artificial intelligence, all the scientists and all the great people at that time, you know, got together and they agreed on

a definition of AI and they said artificial intelligence is a bag of tools for a bag of problems. That was the only definition they agreed on. Right. >> It's pretty good. It's actually kind of interesting. Yeah, >> it is more true today than ever before because you know the the default default mode today is you go and pick an LLM and start looking for nails and the LLM is a hammer. You just go try to you know bridge that thing down, right? It gives you the what really generative AI in this age of the GPD era really showed us is like one big thing was people didn't expect AI to be as good as it is for doing atomic tasks like looking at a picture and identifying what's in there. You know looking at text and understanding tone and donation or meaning out of that

or looking at you know time series data and predicting the next thing. All of those individual atomic things like people didn't realize how good it was and people can't really put it all together. And then a generalist model came in that can put all these things together and suddenly the expectation gap where people thought AI was and where AI really was just collapsed right and that created and there was a user experience paradigm that also changed where like you now can interact with intelligence have a dialogue and conversation which is great I mean it's it's great for very generalist kind of use cases but you're not going to be able to move the needle with chatbot in most industries right you can in some cases >> because That's no worry like healthare for example I'll give you two examples right one and I'll tell you

what we learned as well in this process and I'll come back to I'll try to link it back to the question you were asking on general versus specific a general solution for let's say we're dealing with payment integrity for health plans or like looking at hey whether a claim was paid out at the right level where it was coded right and so forth we did a solution back like two three two years ago three years ago almost to create this payment policies are complex there are like hundred hundreds of PDF documents sitting in a SharePoint site. So we created a curated all of that in a rack pattern, gave a chatbot to the users in the help plan and say go ask questions. Nobody asked questions. it was changing their existing workflow to go do something else that may or may not give them value

but they didn't buy into right >> we same solution we anticipated it and go and say look what I want you to do is when you click on that you know look review claim what you going to do next I'm going to anticipate what question you're going to ask and give you the insight at that point of interaction that user experience chains same rack capability applied in a workflow context text changed the game. 100% utilization within the first 3 months, right? Huge value savings and all that stuff. You're generalist versus specialist kind of a model. So there's a lot of these nuances in there. So there is no one answer versus not. I think today to deliver solution values to particular workflows and stuff whether you use a generalist LLM model underneath it but you have to still adapt it to a particular specific workflow.

And I think I fundamentally believe that you know context is where the magic is. It's not the data, it's not the model, it's the context. And the context is often sitting inside a workflow. Right? If you look at, you know, healthcare for example, there's medical charts. Everybody talks about medical charts, but the visiting physician who's reviewing EMR for medical charts looks for very different things on that chart. Understanding a patient's trajectory and treatment protocol and so forth, versus a clinical researcher looking at the same chart and trying mental math. They're trying to actually map it to 16 inclusion exclusion criteria for a clinical trial. The same charts when reviewed by exact same chart reviewed by um a prior authorization nurse and a health insurance company. They're reviewing for medical necessity against payment policies and so forth. So what we learned very early on is like

you have to separate that content layer which is like transforming data from unstructured to structure and so forth to the context in which you're operating in. And that works really well when you have very specific a lot of small language models lots of augmentation of large language models with small language models but it's all the magic is in the context which is inside the workflow. So I don't think there's one answer where generalist versus specialist but I would say a lot of the use cases when you're the general moniker we use is if your use case or the solution the problem you're solving has to do with how can I mimic a human communication paradigm. I mean responding to queries, customer service, all of that stuff. These large generalist models because people are very um unpredictable in the way what they tell you what they

they themselves don't know what they're going to tell you next. So you need a larger model that is generally capable to go have that writing content all of that stuff. If it's a specific task I am looking for signals inside medical charts to identify if the patient needs or has a care gap of how they manage their diabetes. You don't need a trillion tokens to be, you know, ran through and recent across to actually answer that question. That's like stupid, right? So, I think there is no one answer to it. But what we have actually seen it work is like solve specific problems and then generalize it across. Now, one last thing I'd say on that same topic and this is how we see the future when you start capturing context across multiple workflows in an enterprise. This is what we do really well at

on ice, right? We don't just do try rod versus care management versus this. We're like we go down and help them assemble the building block agents to go solve specific problems but then in the process you're capturing the nuance the data that's not in digital form. Sally knows how to actually do you know prior authorization for DME equipments because she has done it for 30 years. It's not in a job aid. It's not in a SOP document anywhere. We capture that because you build an interactive paradigm to capture that. Let Sally share that with you. Now all of a sudden you're capturing that context from that workflow and you do that for the next workflow and the next workflow and the next workflow. Now we have probably the best data set of understanding the context switches that happen within a health enterprise and that becomes

the the the core data set to build a true healthcare AGI if you really think about right. So yes, the generalist models will work, but you have to start the, you know, it's a trade-off between am I wasting tokens and electricity and power to actually do it and eventually those models will get better, but you won't need that. It'll be an overkill. You will need AGI, you'll need general capabilities, but that is going to come out of you build a lot of specialty and then you build a framework for it to actually share that specialty across to become a generalist model, right? where you get into a coordinator kind of mode where you have a you know generalist agent that is calling the specialist agents to go do this. So those are the kind of you I would say architectures that we actually we think

through and how we have deployed at autonomized and it also makes it a very practical way to generate and deliver value because the biggest AI never had a problem with models and capabilities. It always had an adoption problem right people won't adopt it fast enough. You have to get them to adopt it. And all of that is like you give bite-size value. You get them to interact with you, share data all of a sudden. And then this becomes a generally capable capability over time, >> you know. Uh I I think that's actually pretty interesting what you mentioned about the uh architectures there like this is something that um agentically we are seeing a lot the director agent followed by the specific sub aents uh that are doing specific tasks. Well, like I use uh relevance AI a lot um which has like this workforce capability

for the terms that I'm or the the workflows that I have um and it's it's very effective for general business purposes. But maybe explain to us a little bit more about how you're utilizing that kind of stuff at Autonomize. >> Yeah. So, you know, think about it. We believe like every enterprise, especially health enterprise, but even you know, um take a take I'll take a step back like think about healthcare. We have 336 million people in the United States. We have 1 million physicians that are there supporting 336 million people in the US. And we can't produce doctors fast enough. You know, it takes 10 years, a lot of money. There are campuses and organizations now that are providing med uh you know the med program, medical program free of cost for students because we have to produce more. There is 5 to 6 million

nurses supporting 336 million people. There is no way we can produce enough human capital to support the needs of an aging fast growing population in the United States around the world or around the world for that matter. So you need to rethink this in terms of an enterprise that is in the health services business need to have an army of agents that are spe specialized good in augmenting the human workforce that's in there. So that's one framing of this right. how we do it in this in our thing like what we have built is autonomize is we built a library of a 100 odd agents and these agents are very I mean there's a little nuance I also you know um a detour on agents here agents also we're we're in a world of agent washing everything is an agent and when everything is an

agent nothing is an agent >> oh absolutely yeah absolutely yeah yeah >> so what we have actually when we define an agent it could be it doesn't have to be an LLM agent that's conversational in nature. It is an atomic capability that can stand on its own. It solves a particular problem. It optimizes for a goal and it delivers value. Now, language models or SLMs are better fit for really than than a GPD4 and so forth. So, what we did was actually we built a library of agents after we studied the workflows within a health enterprise. A provider, a life sciences company, a health enterprise, a digital health company. We understood that like for example, if you're dealing with patients, you deal with intake. You deal with intake of information. You deal with claims processing. You deal with revenue cycle management. You deal with care

management, care quality coordination, care coordinate, all of these different things. And we mapped each of those process to understand where are the bottlenecks. We built agents to solve those bottlenecks. And what I mean by that is saying I don't have an Uber agent that does prior authorization end to end for a thing. I mean we we could build that but most likely it's not going to be successful because the prior authorization is made of five or six subprocesses that are very specific in nature and that differs from one customer to another all of that stuff. So what we did was actually we built a library of this 100 odd agents that does everything that a healthcare will need and they can all be independently called as an API plugged into your existing infrastructure or existing IT systems of record whatever. Um and then we built

a configure or or orchestrator on top of that multi- aent workflow engine. We're launching our uh we have a studio product that is in beta right now with our early customers. Um we're going to GA that uh later this year and that allows us to because what we also learned is like workflows are bespoke. Everybody in healthcare no two healthcare organizations have the exact same workflow and anybody who tells you otherwise is BSing. And I I and the reason the workflows that matter. I mean obviously there are some specific workflows standard like you know everybody uses Salesforce but everybody's data model in Salesforce is different. >> Everybody uses epic the way your data is organized within Epic is very different for every single customer. Right. >> So then how do you personally like uh make sure that you do it because you obviously you don't

work with one company right? So >> yeah. Yeah. Yeah. >> Yeah. So what we do is like we have this library of agents that are pre-trained, pre-built for a particular thing with a configurator layer that we add on top of that that'll actually allow them to customize it for their particular thing. That's why the studio product is important because we're seeing more and more healthcare customers even their technology teams or product owners even business users right you know if you're a nurse who is reviewing all you're doing is reviewing uh muscoskeleletal cases every um you know to look at care gaps and call your patient and do something else and you have very unique knowledge of what you need to do. If I give you a an MSK chart review agent, you can give a little bit more extra prompting on it to make it

your own and that becomes your AI sidekick to go do your job more efficiently. Right? We see that is the world that we want to get. It's not going to be AI is not going to be a centralized capability. It's going to be decentralized within organizations because knowledge and context is often at the edges not at the core. >> Right? And when I say edges, it's human beings and people careers whatever. So what we have enabled is like we while we have a library of agents we we allow a layer of configuration on top of that and there's an age-old aden platforms when you build it it's our platforms are our c agents are open for configuration closed for modification right because we do and that closed for modification ensures that we're actually you know uh um we can bring in the rigor of actually

it's pre-trained on certain data sets it will not you know give you any um um uh any output for less than if the confidence score is less than 95% we won't give you the output we'll decline it. So things like that we built in. So that's you can modify those things right that's you know we do go through a extensive process of uh validation user acceptance you know clinical validation eval all of that stuff so we go through that process to get that thing but it can then be configured or customized to your unique workflow needs that's what makes it exciting how we do it >> yeah I mean it is um it's almost like so to speak and and this is where the agentic word kind of gets uh thrown out a lot, but if you're doing it right, I guess it it does

make sense. I feel like when people are doing it properly, like yourself, you're essentially building out subfunctions of of jobs, so to speak, or like uh capabilities of a person, right? Um like I I and this is where the next question naturally leads. I'm seeing a lot of trends from answers on the podcast about this and I'm curious where you think it's going to go like you are obviously making these functions that'll help save time so that and I think in the healthcare industry this is going to be a different answer that's why I'm curious it's saving time for for people it's helping them give better care because let's be honest a lot of like the time that unfortunately doctors nurses etc spend uh is just on like documentation that's like nonsensical um how much time it takes for um and it stresses them out

and then they they don't do as good of a job as honestly AI would. Um so do you feel like this is just going to help in your case maybe differently than some of the other industries cuz like in marketing or like sales all these areas people are just concerned about like job loss but I feel like it could actually be an interesting thing where it's just like we just get better care right in healthcare. Maybe I'm off base, but >> No, no. I I think it's a good question, right? I think there's there's there's two things we if if you extrapolate how this world progresses, right? There's one the classic argument that well, a lot of the mundane jobs will be lost. I don't believe that. I think it just it will evolve and it'll change, right? You know, we we it's history has

shown us, I don't want to go into it, like horse riders versus cars and drivers and all of that stuff too. you'll create a class of new um you know jobs that didn't exist like you have AI engineers, prompt engineers. One of the things we actually do is like our ML engineers don't write prompts. >> We give them prompts and and it is a very nuanced thing where we say look the prompts should be written by the folks who are closer to the problem not the one building right prompts in a programmatic fashion and prompt engineering practices and stuff. I think those will things will evolve quite a bit. the best one to write that prompt on a model how it actually should behave is the one who's closest to the problem or the knowledge or the context right so that's one thing so I

think in any industry so you're going to have some labor displacement which is a temporary thing you know which because people as they say right like it's not the AI is going to take the doctor's job away but the doctors who don't use AI will actually now be displaced by the doctors who use AI because there'll be more things in healthcare is a little bit more nuanced because we still need every doctor we get up. So that's a different problem. So I think one is the most industries will see AI as an augmentation layers like I like to call it the Jarvis suit right everybody has the opportunity to become Iron Man today because you have an AI powered Jarvis suit that you can actually use to go do your jobs better. Some of the more mundane industry I would say not mundane but like

some of the more industries that are more standardized in nature like marketing for example like digital marketing it's a class of things there's only certain things you do and stuff you will get the vanilla treatment with just AI and not having any human beings but then then it becomes more competitive how do you stand out you stand out by having that human touch you stand out by doing things that everybody else is doing which is basically what's on the internet that's train strain on these large language models then if you are more creative you now have an edge over everybody else right and I think >> yeah there's a significant problem with like um if you look at even I'll let you sorry I didn't mean to interrupt this there's a significant problem in marketing right now where it's like a uh a feedback loop

where content is being written by AI which is being written by content which is being written by AI and then now with the advent of the web search functionality uh naturally into all of the models it like feedback loops itself even more so Like something came out the other week where which is like atrociously bad uh 40% of the uh web search results coming from like a Google bard or not sorry I can't call the bard uh Gemini's like AI preview is coming from Reddit and I'm like >> that's bad. I was like okay I was like all right cool. So then but then the other percentages are from blogs >> which are being w written by what? which are being written by AIS that are referencing Reddit. >> You know, it's interesting. It's >> like, so everything's just some it's I'm like, okay, so

now everything's a Reddit tier argument. Great. That sounds that sounds like that sounds awesome. >> I mean, you know, the the silver lining in all this Dimmetry is the fact that look, you know, I think it is an opportunity. All of this noise will make it all the more human again to perform really great work, right? I mean, think about what the the output of this is like. We will probably I mean we see this in fraud that is happening with video fraud like people making calls to other people. It's not actually people as their AIs that's being trained on their avatar of public places you know public figures. You're seeing voice calls that voice emulates everything based on you know you know you and I you post this podcast I'm sure your grandma will probably uh get a call saying it's Dimmitri who sounds

like you because you have enough content on YouTube to actually go train that on. Right. my uh my my my cena old Greek grandmother is not going to be happy to hear this. So, um >> No, but seriously, I I I haven't had to tell my parents before like, hey, um I you have to really we have to have like code words or something. >> Exactly. And I think >> like Yeah. >> So, what this will all lead to if you really extrapolate this Demetri is that you know it will become more meaningful to have those human connections. for example, more business will be done in face toface settings than on Zoom. I mean, it's going to happen very soon because you won't trust what you're going to see on the other side of the screen, right? Because there's a, you know, you can have

code words all you want, but you know, you're having a what used to be an amazing meeting and you go sit in there, it's like, I don't know whether there's an AI that's just saying what I want to hear. And some of these, you know, all these u recursive self attention models are really good at, you know, uh, psychopanny. So, he'll actually tell you what exactly what you want to hear, right? And this creates this bubble. All of that now it's just like okay you want real business to happen go meet person meet in person go make this happen. You want meaningful relationships to be built. It just happens to happen on person. So all of this I think this this whole noise levels going up because of all the content being generated on the internet. Um I don't think it's going to make the

internet any less effective. It's just going to be a different way we we evolve into a hybrid world where you know human connection becomes even more meaningful than it was because you know what's because that becomes the scarce thing. So whenever humans have some scarcity like >> it has more value. >> I mean that's a good point I think for me like um been trying to do like sales work and stuff and I I was just thinking about the other day I'm like I should like I should go in person. People would just be so shocked or like I have friends who are trying to get jobs right now and I'm like you should just go like when it's like it's like I want your writing samples just show up. They'll be so gobsmacked >> becomes the exception not the rule. Exactly. >> Yeah. And

it's a very it's a very weird environment we're in now. So um but you know like healthcare is obviously even had that trend a little bit too. like I um I'm asthmatic or at least I have like a athletic induced asthma and uh or exercise induced asthma and like I got a got another aluterol prescription like over tellahalth and that was just weird to me you know the experience there was odd and I'm like I don't I kind of miss I was like can I go to the doctor and I'm like ah but I don't want to it was a very the experience was odd right? Yeah. Yeah. Yeah. >> But um yeah, it's a it's an interesting >> interesting world we're living. Let's also you know the other thing the other side of this is like you know we talk about this from from

you know healthcare this I always believe like healthcare is a lot about care than health right so that whole human condition there's enough research and done like historically how know you had these healers in communities before civilization took off and became very scientific right you had all >> raiki healers if you remember right so all of that stuff >> there is some there's not enough research done there's like quantum vibrations and all of that stuff that actually leads to how people get healed. The the you know evidence-based medicine and all that stuff came in much later in this civilization history. So right so there's there's an element of that. The other thing is like I think the other the positive side of this whole thing is right right now with AI >> you can have a really meaningful conversation with your doctor >> right because

you can just get you get a lab report you throw it up on plexity or chat GPD and at least you'll know what questions to ask. Now the problem is you cannot cross that line and saying you have to know when it is actually bullshitting which is not which is the hard thing that's why you get trained on at least hey what question should I ask so you're going to get into a world where the educated healthcare consumer is this I think the the other you know part we don't talk about enough is like consumers are now in a much more powerful position in healthcare than they traditionally have been and we are seeing that when we work with healthcare businesses how much they are seeing the change in behavior on the patients right earlier it was like you get an appointment when you get

it right now they're like okay they can have means for it they'll get concare they'll get teley health they'll go to your competitor they'll go to so you have to start thinking about and and you can't just say you still hear stories where you know providers or doctors who are like they're just so busy they spend 60 70% of their time doing administrative bottleneck they're frustrated they do pajama time on the computer after the kids >> yeah they don't want to do all that like that's I don't think that's been any of their >> I don't think people get into healthare for that reason. >> No, you know, in fact, I I you know, nobody like if you really think about people who get into healthcare, especially in the US healthare system. Software engineers probably get paid more. They don't have to work the odd

hours that these guys do. They listen to sick people all day. They, you know, you know, you're people are constantly complaining on their face every day and you still have to go and you get calls in the middle of the night. All of that stuff. They do it for mostly altruistic reasons. They're like, I want to make a difference to my community, to the patients, solve problems, that kind of stuff. And as a system, we've kind of failed them a little bit, like especially the healthcare frontline workers to to do it. I think, you know, AI is the opportunity to go rewrite that that uh that narrative for everybody and for healthcare knowledge workers. And do you feel like that's, you know, I mean, if that was your ultimate goal for what you're building, that would be helping to make that like more of a

easy reality. >> Yeah. I know. So, look, we we're we're building the infrastructure needed to make healthcare run better, right? And that includes how businesses can optimize their business to deliver best care, how patients can actually use the available information and access to care and make it easy for them to go understand it. Like simple things like you get EOB bills. It's like a maze. How much do I really owe my provider? Should I pay this bill or should I not? It's a maze, right? Simple problem. >> If it wasn't from my parents and the internet, I have no idea how anybody would sol I I let alone anyone would solve that. Like it it is actually the most frustrating experience getting those. >> What do I owe? >> What do I owe? So, and then on the provider side, like you said, you don't

want them doing pajama time with their kids after they go to bed. Every doctor in the United States spend at least 2 or 3 hours of time after the kids go to bed finishing up the notes and administrative work for the day. >> It's ridiculous. It is >> that's really sad. >> Bonkers. >> And and nurses and PAs and nurse practitioners and stuff too, right? I mean, it's just like it the all of these folks, all of the folks in care who are at the point of care, they're so administratively burdened that the last time they read read a research on how to manage a chronic condition like diabetes was when they were in med school. So, it just doesn't help. >> That doesn't sound ideal. >> It does not ideal at all. So, I think that's the opportunity. So our point is how do

we go give the first order business for us is like give Jarvis suits to everybody in healthcare make them more productive give them the right things give them the right infrastructure to go run business care second help the businesses have the right operating system to deliver care effectively efficiently and you know don't have to spend a lot of money doing dumb things to be honest right final frontier for us >> they are and the final frontier for us is going to be then How do you go empower the healthcare consumer with all of the work we're doing to make sure they can make better informed decisions? And you know, I want us to really get to a future where we don't talk about health care as being taking care of the sick people. Healthcare should be a conversation about longevity, quality of life, how do

you live your full life, how is more, you know, fulfillment across beyond just like, you know, I thank God I stayed out of the hospital this year, right? That is this true north for what we can do if you just apply technology the right way, empower the right people and go deliver u you know deliver a difference with AI. >> Yeah, I that's totally fair. I think I think that's that's a that's a good plan. Um I like that for you. I think that there was a company that I worked with as a client a while back where it first started hitting me that this uh that the AI trend could really help um healthcare a lot. Um it's called soap note AI and they just very much uh harped on uh I mean it's a it's a note-taking methodology. I didn't know this. I'm

not in healthcare. Um, and I don't know. I I just feel like this is this is a a realm where a lot of people who are skeptical skeptical about AI should should take a look because I'm sure there's more industries just like it that we're not thinking of it in that way. Um, and everyone's obviously always concerned about what's going to happen with the job market and maybe the more traditional less uh like more standard job situations, but with this everyone's impacted by healthcare. Everyone's impacted by the costs. Everyone's impacted by poor health care. Everyone has a story. >> Everyone has them or their family >> or whoever, their friend that has have had a poor healthcare experience. And and I wouldn't doubt that it's due to the administrative bloat. And that could be either from a cost standpoint where there's too many administrators so

the cost of health care rises or someone was too tired to to properly research something because they're doing all this stupid paperwork. So >> no, it's exactly right. It's either like as you said like all the cycle time, all the cost of care and the quality of care that we receive can be rooted back into everybody's overburdened with administrative processing. >> People get tired and make errors and emissions that they shouldn't do. And then you know it just like alignment of uh incentives is another big problem in in healthcare industry but increasing transparency um can go a long way and that's where like things like AI can come into play to your question on like I think more more organizations it's very rewarding to see are taking the step right now like when we started the company three and a half years ago it was

uphill battle trying to get you know um organizations to talk to us about like hey I'm not sure we're not ready for AI Right now it's a boardroom conversation. CEOs are actually being told by the Google and Microsoft executives that they talk to say you can cut your 50% of your IT budget by using AI instead of engineers. That's what they usually tell them. >> Then they get excited. Yeah. >> Then they get excited. At least it starts the conversation. Right. I think the big thing I would say what is happening right now and it's going to be more and more critical for us to continue to do this as an industry especially in healthcare is separate the the signal from the noise. There is a lot of noise. There's a lot of folks who come in and say because everything looks good on a

demo because it's easy to build a demo with, you know, that 50% cost reduction on dev that applies to building demos, right? You can build a demo in a few minutes or like a 30 minutes using Google Gemini or build a GPD to show something and stuff like that. So, but the the proof is in the pudding. So, for example, we like we work with se three of the five largest healthcare enterprise in the United States. one of them in the last 18 months or so that we've been working with were able to do a hundred million of GNA compression their general administrative costs in terms of operational cost savings >> not just because of us but the all the operational efficiencies that we brought in and that allows them to reimagine some of the workflows because the first order of business dementry is like

oh I'll go take down inefficiency oh I'll just compress this workflow I'll solve this problem here solve this problem there all of a sudden you'll see that oh Why am I actually solving this in, you know, band-aids where I can just completely now I've just exposed it to the bones and I can reimagine this entire workflow? Why should actually go through 15 touch points before it gets to that point? So then you start having those conversations and now we're helping them reimagine some of these workflows which will be huge cost takeout opportunity, redeployment opportunity, right? Like it's a big thing in healthcare like we don't have enough people to deliver care. So in those people, those nurses who are doing administrative processing and health plan should be at the point of care delivering care to patients should be at the care management talking to patients

and helping them navigate their care plan or updated and you know all of that stuff. I think that's what we're unlocking for them. So the the so the the the positive double whammy here is that on the one hand you reduce your general administrative expense on your your balance sheet looks better and everything. On the other hand, you can redeploy that to high value areas like care management, population management that gives you better outcomes, health outcomes of patients. Patients like hearing from doctors and nurses, right? I mean, you know, we don't have time to go. I have so many stories when we talk to provider groups and or health plans or patients even where simple things you know just the fact that like a a GI uh group like a gastro group if you will yes >> there's some of these >> diseases where like

you know if you have like IBD or IBS kind of diseases like irritable bowel syndrome >> the follow-up protocol is very very very critical >> because it can lead to really bad outcomes and it's as simple as calling the patient and saying Hey, how are you feeling? Do you feel bloated or not? But the nurses or the folks don't have the time to do that. So what happens? The patient goes into a protocol. The next time their appointment is 3 months later, they've already gone to a situation and 3 months later when they come in and they have something happen, they redo the appointment. Now they've lost the window. Now they're waiting for another four weeks. All of a sudden situation degrades. It becomes a a bigger episode than it should have been. All it took was the time to make one call, have that

conversation, check in on that patient, but they don't have the bandwidth to do that today. So, I think, you know, all of the things pulled together, healthcare is in a better place today to do that. Um, I think more and more organizations, they should, you know, we're talking to several organizations and others who are listening to this should reach out to us. We'll show them how to actually think about it. my point and signal to the noise. We're one of the very few organizations in healthcare AI today that has generated value for our customers. You'll see the metrics on our website. We have been able to and then we because we're focused on hey the business of care kind of an ecosystem. We're not trying to solve for hey how do I find the next drug or how do I actually go you know do

the best um you know um um you know point of care solution to do a diagnostic better and stuff. We're focused on making the system less painful for patients and providers and okay that is that focus has given us a lot of early lessons and what do what not to do and we've been able to generate a lot of value for for our customers and organizations. So I think you know I'm more bullish and more positive about healthcare than I was three and a half years ago. So um and and um it's it's really great to see and I think there's a lot to be thankful for all the generalist movement the AGI movement because all of that raises the conversation the awareness of the s situation which makes everybody pay attention which then now you ever get into the mode of actually not just

learning and listening and hearing about this and listening to your podcast but then I want to try it. I want to see what's right for me as an organization as an individual. Do more of that. Get the value. So now we're we're we're passing from that phase of it's still hyped up and earlier it was geni now it's agents hype and stuff like that but that that that switch from just being the hype and talking about it to say like how do I do some things generate value we're hoping that in the entire industry gets to that point very quickly and at least our customers are getting there so which is great to see. Absolutely. And I think uh that's something that um I I want to just tell everyone as a as a close out to this episode, you know, please go check out

what um they are doing at Autonomize. Please check out um and and if there's anything else you want to plug, please feel free to Ganesha has a has a LinkedIn. Obviously, everyone's got a LinkedIn, but he's got a decent following on there. You want to go check him out there. Um posting some cool stuff, it seems like, too. Um so, yeah. And and you know I would the last thing I would say I would add on to that Demetrius and thank you again for having you on the show. You know AI is such a critical conversation for you know it's pop culture now. It's part of how we actually you know you know our kids are going to grow up all of that stuff. >> Sure. >> Don't leave the conversation to only people who think there can be in the boom. So everybody needs

to participate and take part in it. Right. And uh for that reason I like four years ago three four years ago now you have a lot more podcasts but I started this podcast called stories and AI and we're going to you know start the next season you know in a couple of weeks actually we want to launch this next season. So I had 200 plus episodes on it and you know like your podcast is amazing and we more focused on folks who are actually in the throws of you know using AI building AI in large organizations small organizations and stuff ethic researchers and stuff like that to go and share what they're seeing beyond just what you read on public news right so um we try to actually uh gauge the conversation but the goal idea was actually inspire everybody to give take part in

it so Great job to you. What you're doing is elevating the awareness and the education, the engagement for people. I think everybody should take part in it and participate in this movement. Um, and make it your own. I think, you know, uh, back in I was in a, you know, my one of my previous companies I was in, we had this phrase that we used to say and it's more apt today. Uh, you know, take part in it. you and AI can change the world and you and I can change the world kind of a thing wherein you can actually make it your own or you now have the tools to do it then you didn't have before 5 years ago six years ago right so I think you know it's a very important aspect of um you know how we evolve as a civilization

and I think it's a huge call to action for everybody but thank you for having me >> absolutely of course no thank you so much for being on here uh everyone please make sure to go check out autonomize.ai. Thank you so much for watching and we'll see you in the next one. Bye. >> Thanks, Mick. [Music]