theCUBE + NYSE Wired: Mixture of Experts Series | Nachiket Deshpande, Impetus Technologies

Clips
More from theCUBE + NYSE Wired: Mixture of Experts Series

Nachiket Deshpande

CEO

Impetus Technologies

play_circle_outline AI Platform Shift: How Rapid Innovation Is Outpacing the Internet and Cloud, Reshaping Every Industry

play_circle_outline CEO drawn to Impetus' enterprise data expertise and software-plus-services model

play_circle_outline Inflection point vs pivot: doubling down on strengths rather than abandoning them

play_circle_outline Shift from model-centric to systems-centric focus: workflows, architecture, orchestration

play_circle_outline Context gap: models lack enterprise data, industry semantics, and execution processes

play_circle_outline Governance, monitoring, and human-autonomy coexistence essential for adoption

Info

Nachiket Deshpande, Impetus Technologies

Nachiket Deshpande

CEO Impetus Technologies

search

John Furrier

>> NYSE Studio, of course. We have our Palo Alto studio connecting Silicon Valley to Wall Street. This is our Future of Agent Series. We unpack where AI's going from a business value standpoint. As the AI infrastructure continues to build out, we're seeing a huge demand for tokens. We're seeing a huge demand for agents in the enterprise and this is causing a lot of change and a lot of activity at the agentic layer. We got a great guest here who's going to unpack the industry trends for us here, Nachiket Deshpande, CEO of Impetus Technologies. Nachiket, thanks for coming on the cube. Appreciate you.

Nachiket Deshpande

>> Thanks, John, for having me.

John Furrier

>> Your career, you've seen a lot of different ways of innovation over the years. You've been involved in IT transformation. You've been involved in a lot of the tech build out over the years. This AI wave looks the same, but it's not the same. What pattern do you see evolving that's similar but different with agentic because there's a huge amount of demand. The hype is out there. I think it's worthy, but there's a lot going on.

Nachiket Deshpande

>> So in some ways it is similar to the previous transformation that we saw whether it was with the internet in late '90s or whether it was with cloud in early 2000s. I think the main difference though is number one, the pace at which the innovation is happening in this space is just unbelievable. You start this conversation at the end of the interview, the market or technology would have moved. It's like that pace of change, which is very unique. And I also feel that it is one sort of technology innovation that's going to impact every single industry and every single business around the world. So it's not isolated to some industries. It is not focused on certain segments, but it's going to disrupt and transform all businesses. Those are the two main differences I see with AI.

John Furrier

>> Yeah. The platform shift is... We've kind of crossed that threshold. You took over as CEO for the company. You could have done anything. You could have gotten multiple jobs. What was attractive to Impetus Technologies? What do you see going on there? Share why you were attracted to the company and what you got going on.

Nachiket Deshpande

>> So there are two things that actually really attracted me to choose Impetus over all other opportunities I had. One was data and being in the enterprise data is the biggest differentiator I feel which will make AI work. And Impetus as a company has been focused on working in enterprise data for the last 18 to 20 years. So we have been working in data long before it was sexy to be in data. And that's really the core how the engineering strength and the depth in the company is all geared up. And second, I feel with AI, softwares and services are coming together. And that's how even Impetus was set up. It was always a combination of software and services. Those two attributes is what all companies in the tech services space are wanting to get at. And here, Impetus was always set up that way. So for me, it was a great platform that we would be able to help our customers navigate through this transformation journey much better.

John Furrier

>> A lot of people talk about inflection point, platform shifts. We just mentioned that. Also, we hear the word pivot. I kind of like the word pivot, but on one hand, I don't like it. It implies stopping and turning. Inflection point means a kick up on a curve. I would think you guys are more positioned as an inflection point play because you already had the DNA. You were kind of doing the work going into the market, so it's not a pivot. And a lot of companies has been in that mode of in the cloud native world or coming into AI that have all the piece parts. Talk about that differentiation. It's very nuanced, pivoting versus inflecting.

Nachiket Deshpande

>> Correct.

John Furrier

>> Talk about your view.

Nachiket Deshpande

>> I think you're right about that. When I look at the software and the data, to me, that's how I would differentiate inflection point versus pivot is you are going to double down what you're good at and add a few new things to accelerate onto that curve. Versus pivot, you have to stop doing something and start doing something new. And for us, we're just continuing to be very relevant to our customers in their enterprise data journeys and then helping them use that enterprise data to make sense on their agentic AI journeys and get the ROI they're looking at. So that's the acceleration part that we are focused on.

John Furrier

>> That's awesome. Well, thanks for that setup. I want to get into the future of the agents because the conversations have formed in the industry around models and outcomes and the models really got all the attention. But you saw this year, already seeing it in the first half of the year, all the top players. The models are being decoupled from the actual stack, if you will. So choice is a big thing. And it's shifting towards workflows, architecture, orchestration. That's a systems place. So we're seeing a shift from models to systems. Explain what that means because you guys are... You work with models, but you have the system DNA. Talk about the shift from models to systems.

Nachiket Deshpande

>> As the whole generative AI wave evolved over the last three, four years, attention was on models, and every single day newer model or better capable model keep coming. It almost feels magical in terms of what those models can do. But as you start to get into the enterprises, what those models can do in isolation, they're not able to scale that within the enterprise. And the reason they're not able to scale in the enterprise is because that lack in the system's thinking. Because for something to work in enterprise, you need to be able to understand cost equations and are able to scale with ROI. That's one challenge that the models alone are facing. Second, it has to be integrated in your enterprise own ways of working. And you cannot put a model on top and say, "Okay, now it starts to make your systems or your processes intelligent." It has to be ingrained in your way of working, and hence it has to be more organic and hence the system's thinking is needed. The third, I think every enterprise is unique in its own way. They have their unique customers. They have their unique products. They have their unique processes. Some of them are good, some of them are bad, but they're unique to them. And if you don't take into account that uniqueness and you don't use that uniqueness to your advantage, you are making a commodity play where your competition has access to the same model. So how do you unique? You have to preserve your uniqueness and enhance it with model and how to replace it with model. Again, that requires you to have a systems thinking where your uniqueness is engineered into how the models work.

John Furrier

>> Yeah, that's awesome. Also, we've been hearing a lot about context, agent evaluation. Observability is coming up into the agent layer, which is a cloud native concept. You got to observe, got to evaluate. So context has been a discussion. You talk about the context gap, which I'm interested in. What does that mean? Explain what the context means because people are like, "Hey, just throw more compute at the models. We got inference." Inference is getting cheaper, but the compute's going up.

Nachiket Deshpande

>> We see that in enterprises all the time. Most of the enterprise leaders today face that dichotomy, that on one side they see the models that are becoming so powerful and do so many things with every new release, every new capability. They come to their enterprises and they're not able to get the ROI on those use cases that seems to work in a lab or in a proof of concept environment. Every single company over the last maybe 18, 24 months have done hundreds and thousands of POCs and all of them seems to be attractive at the POC stage. But if you look at not even two, three, 5% of them at best have made it into production. And that gap is because the solutions that work in POC does not work when you try to expose it to real data, try to scale it to all the users in your organization. And as you double click on why that is happening and that's where that context gap comes into play. I feel that's happening because models are capable, but they don't know you. And how do you make the models understand you is the trick to make those POCs into a real value production system. And what is you is made up of, in my mind, three layers of information. First is, of course, the structured data that's unique to you, which is your customers, your products, your services, your transactions, your supply chain. That's the structured data that's...

John Furrier

>> And that's not in the models.

Nachiket Deshpande

>> That's not in the model. That's in your system.

John Furrier

>> That's locked down.

Nachiket Deshpande

>> That's in your system. The second piece is there are nuances to every industry. There are certain ways of working. There are certain regulatory constraints. There are certain uniqueness to how every industry works. So that's your second layer of the context, which is probably not unique to you, but it's unique to the business you are in. So probably shared by 10, 15 or 20 others, but it's still unique. Model still doesn't know that because model is built for millions and billions of people and companies, but that's the second part. And the third, which is very unique, is your way of working. See, how things get done in Impetus is different from any other company and so is true with every other bank and so is true with every other life sciences company and manufacturing company and so on and so forth. And it's important for you to understand or codify how things get done in your organization. And we call those are three layers of gap, right? So the first one is a data gap. The models don't know your enterprise data. How do you unlock your enterprise data and make it accessible and available to the model at the right place at the right time is your first gap problem. The second is the context gap because that data has to be used in a particular industry business process in a semantic context. And the third I feel is your execution gap because that's where you... How get things done in your organization will vary. And all of these put together will bridge that gap between a model that is capable and the ROI that you see within your enterprise.

John Furrier

>> Yeah. I mean, basically what you're saying is that models deliver good outcomes. But if you have the right context, you get better execution that yields to ROI and business performance.

Nachiket Deshpande

>> Correct.

John Furrier

>> All right. So let's unpack the challenges for the enterprise because I like the way you lay that out. Because one thing that we were talking about before we came on camera here was that we both, and you've done extensive work in this area in your career around IT transformation projects. You're smiling because we probably got a lot of scar tissue. IT transformation isn't the same as it is happening today because those layers point to the fact that this is business model, business process up and down the business. Because if data on how the business is run, data on what's the business about, you can use that. So this is a business transformation. This is the top theme in all the narratives we're having on theCUBE. How is that different from IT transformation and what is the challenge for the enterprise?

Nachiket Deshpande

>> I think the fundamental shift with the AI that takes it away from being IT transformation to business transformation is in the past when you were building rule-based systems, defining the system was a small part of the work. A large part of heavy lifting was happening in actually building that rule-based system and making that robust and making it run in an enterprise. Hence, IT transformation was the bulk of the heavy lift in that journey. Now, when building has become trivialized with models, you can build it really fast, really quick, and that's not the bottleneck. The code is not the bottleneck anymore. Then the bottleneck shifts to the intent of what are you building, what are you solving? And that question essentially is a business question is where you need to really understand what is that you're trying to transform, whether it's a business process, whether it's a customer experience, whether it's a product innovation, whether it's a channel innovation. Whatever that you're trying to solve essentially is a business problem and understanding what to solve is important. Another difference is when the technology implementation cycle was wrong, the clarity emerged along the way. So you started with a particular vision in mind. But with every step of your IT project, better clarity, better changes emerged and you got to the end point. Now, when that cycle is largely done by models, you have to have all that clarity upfront because you need to know what otherwise gave you time to evolve that clarity. You got to now sit back in your chair and think about all of those things upfront because that's the intent you will need to go to and give to an AI system to be able to execute.

John Furrier

>> And there's more stakeholders involved because the business includes CFO, Chief People Officer, CIO. I mean, everybody's involved. It's not like the IT guys. So they're involved certainly because the CIO and CISO have a lot of challenges. I want to get your thoughts on shadow AI because I think it's coming together for me. I like where this is going because shadow AI is really built on shadow IT concept where we all live through that. Hey, the IT guy goes around the boss, puts his credit card down on AWS, gets a prototype up and running, gets his hands slabbed and gets promoted. That's kind of grew the cloud native. Shadow AI is happening everywhere. I've seen CFOs closing the books and doing a lot of shadow IT. Shadow AI, I mean. Shadow AI is becoming a feature because what it's doing is empowering people to feel the velocity and execution of agents. And so that's going on. And it's kind of like a kid who's 16 years old who gets his driver's license and gets the parent's car and starts driving around town. It's very motivating. So when you get the keys to the car, you get a lot of power. This is really kind of becoming an operational thing. What's your thoughts on this? Because you don't want reckless driving of AI because people are going to start doubling down on it. Because when they see the value, they just do more of it.

Nachiket Deshpande

>> Yeah. I look at it less as shadow IT, but I would look at it more as decentralized or empowered IT where you're empowering people who are actually delivering value in front of your customer or any of your stakeholders to actually solve the problem then and there, rather than having to translate it to a back office, which is then converting it into this and then coming back. And the pace of change is such that that cycle is too long. Hence, now the empowered IT or citizen developers, as some people call it, is becoming more of an intentional model than something that's happening under the radar. And hence, the role of the IT or technology in an enterprise has changed. The role of the technology in an enterprise is to now enable that, enable that with right guardrails, enable that with the right financial ROI metrics. So that, continuing your analogy, when a 16 year old is driving that car, he or she now knows what are dos and don'ts, what are the best practices, what would keep them out of trouble, so to speak, while not taking away their flexibility and saying, "I will drive with you every time you want to go." But saying that here are the...

John Furrier

>> You got to pass the road test.

Nachiket Deshpande

>> You got to pass the road test and beyond even just the rules, here are things that you got to keep in mind in being a successful driver and protecting yourself and the others that are...

John Furrier

>> I love the car analogy. I was watching F1 in Miami last weekend and of course, all the tech companies sponsor the car. So highly interested, and plus my son's a big fan. I'm somewhat of a fan, but the technology is the car, humans are the drivers. I love that analogy for AI. I know you and I used that before. Everyone pretty much agrees in the industry today that I talk to that technology's not the problem. Explain how that pivots out because now we're getting cars for everybody, not just the IT guys. So talk about that technology car analogy. It's going to get better. We see the improvements on the roadmap. You guys have stuff coming out. But the human driving, that ultimately is what you want.

Nachiket Deshpande

>> So I think there are two parts to that story and I'm taking that car analogy forward. If you are driving a normal car on the road and if you go to an F1 track and sit in an F1 car, then very likely in that first three seconds you're going to crash that car somewhere. And not because the car has a problem and not because you are a bad driver, it's just that as a driver, you're not trained to drive that car on that track. So a lot of this about AI and AI agents is like that. It's a very, very powerful machine, but most people and the processes and the systems aren't trained to handle that power. So if you don't make that change management work first, you are likely to get a lot of crash. And learning it by doing and bumping it a few times is a very expensive way of learning, especially in an enterprise when you have so many stakeholders they're accountable to. For me, the adoption of AI, one is, of course, the car itself has to be tuned for the track, which is where we talked about the enterprise context and getting it right so that it's not just that the car which is in the manufacturing plant works in the lab, it also has to be tuned for that track. And second, you've got to do change management and training of drivers to be able to handle that power, so that you are able to use that power to get the speed and the timing you want. Otherwise, it will be a very powerful car, but that same distance of a mile will take you forever because you will probably...

John Furrier

>> You don't want the Ferrari to go down and get the milk at the corner store. You want to use that on the open road. Love the car analogy. Let's extend the car analogy to agents because I used to drive my car to get food. Now the food comes to me. The agents are Uber and DoorDash and all these services. So autonomous is coming. We see it in cars coming in. Agents are autonomous humans working, digital workers. What's the agentic reality? Because now you're going to have, okay, let's get trained on how to drive the technology. That's the AI. So talk about the human AI operating model for leadership. Because as you guys roll out your platform and as the platform gets more powerful, you'll have agents working on your behalf driving their car. So talk about that connection and what's the reality of where that sits with agentic.

Nachiket Deshpande

>> So there are two components to that autonomous agents related challenges and opportunity. I think number one is it's very important that within the enterprise, you have a very clear idea that what tasks require a deterministic approach and what task requires a non-deterministic approach. The models are extremely useful and powerful in a non-deterministic world. But within an enterprise flow, there will always be certain things which will need to be deterministic. They could be influenced by regulatory. They could be influenced by the risk. They could be influenced by many other factors, but you will have that. So when you are designing your agentic strategy, it's very important for you to know what parts of my agent execution and capability is going to be deterministic, non-deterministic. That's the first part of that. And then the non-deterministic part that you want it to continuously improve and learn, that's when do I switch it to become autonomous where I have enough trust in that, enough controls to maybe able to do that. They will vary from industry to industry and process to process because there's certain industries where segregation of duties is extremely critical. And it is for the protection of all the consumers in that. And hence, you would not want to put autonomous in such scenarios, but there are lots of back office tasks that actually you would be very benefited if it was done autonomous. Because in a way, no humans should actually do them. It's just that because you didn't have an option, humans were doing it. It's not a very value added service from a business perspective. So you would need to look at that. So I think the selection between autonomous, non-autonomous is important. And it then comes to the same thing, the FSD driving versus manual driving in a car. One, it will based on your own risk appetite. You will not do that on a New York street, you will do that on a freeway. And that's the difference where I said where you apply and where you don't apply is how I look at it.

John Furrier

>> Nachiket, talk about the industry implications as to where we are on the progress and what's next. Because as agents comes, where are we on the progress bar of agentic reality and what's next? How do you see the industry evolving?

Nachiket Deshpande

>> So I feel that the technology is very much there. It'll keep improving, but I think the technology that's available is good enough for the agentic adoption to scale. But where the gap is, one, as we spoke about, is the context because the enterprises need a lot more preparation and background platform work to be able to make use of that technology in a meaningful way. So there is that curve and that wave of investment that has to happen. Second, I think we also don't know how to govern, monitor, and work with these autonomous systems that come in. So it's like, if you take the same car analogy, your road infrastructure, your signaling infrastructure, all of that also has to improve if every car is going to be as fast as a F1 car on the road, just today not. So that's the other piece that the governance of the systems and how do we really manage and monitor agentic systems that needs the evolution. And third I feel is, like we said, all the drivers who are going to drive those cars or even if there is a part of the traffic is going to be autonomous drivers and part of the traffic is going to be human drivers, I think the humans need to know as much about the autonomous cars and the autonomous cars also need to know enough about human behavior to coexist. So there's a lot of change management, training, familiarization, a lot of that is needed on the humans as well for that to happen. So those would be the three things that I believe would be the focus over the next 12 to 24 months before we see a true adoption of agentic in the enterprise.

John Furrier

>> Well, certainly a lot of demand. What's on your roadmap? What are you focused on now? What do you got coming up?

Nachiket Deshpande

>> So for us, I think our superpower has always been in the enterprise data world and we want to use that superpower to solve that context gap problem that I talked about because we understand enterprise data intimately well. We understand models very well. So we are in best position to be that bridge to be able to help enterprises close that gap. And as I said, our approach has always been software and services together. So what you would see from us over the next year or so is a lot of software and services capabilities that we will roll out that will significantly help enterprises bridge that gap.

John Furrier

>> Any teasers for the big news? Try to get your pin down here. Got some big news coming up.

Nachiket Deshpande

>> I think yes, we have a big launch coming up. Our set of solutions that we believe will significantly accelerate that journey across the data gaps, context gap, and the execution gap for our clients. We would also launch a lot of industry aligned examples and use cases that will show it, not just give them tools, but also give them examples of how it actually can be realized in specific industries. We've been working with some of our beta customers already in the background. So I'm hoping all of that would unveil and we would be able to see a lot of traction with our enterprise customers on the journey.

John Furrier

>> Well, the conversation has shifted from models to architecture. Congratulations on the momentum. Thanks for coming on. Appreciate it.

Nachiket Deshpande

>> Thank you.

John Furrier

>> All right. I'm John Furrier with theCUBE here. The future of agents and the agentic infrastructure is emerging as AI infrastructure continues to build out, a lot more systems thinking, a lot more platform shifts continuing to happen. We're doing our part to bring that to you. Thanks for watching.

Nachiket Deshpande, Impetus Technologies

search

John Furrier

>> NYSE Studio, of course. We have our Palo Alto studio connecting Silicon Valley to Wall Street. This is our Future of Agent Series. We unpack where AI's going from a business value standpoint. As the AI infrastructure continues to build out, we're seeing a huge demand for tokens. We're seeing a huge demand for agents in the enterprise and this is causing a lot of change and a lot of activity at the agentic layer. We got a great guest here who's going to unpack the industry trends for us here, Nachiket Deshpande, CEO of Impetus Technologies. Nachiket, thanks for coming on the cube. Appreciate you.

Nachiket Deshpande

>> Thanks, John, for having me.

John Furrier

>> Your career, you've seen a lot of different ways of innovation over the years. You've been involved in IT transformation. You've been involved in a lot of the tech build out over the years. This AI wave looks the same, but it's not the same. What pattern do you see evolving that's similar but different with agentic because there's a huge amount of demand. The hype is out there. I think it's worthy, but there's a lot going on.

Nachiket Deshpande

>> So in some ways it is similar to the previous transformation that we saw whether it was with the internet in late '90s or whether it was with cloud in early 2000s. I think the main difference though is number one, the pace at which the innovation is happening in this space is just unbelievable. You start this conversation at the end of the interview, the market or technology would have moved. It's like that pace of change, which is very unique. And I also feel that it is one sort of technology innovation that's going to impact every single industry and every single business around the world. So it's not isolated to some industries. It is not focused on certain segments, but it's going to disrupt and transform all businesses. Those are the two main differences I see with AI.

John Furrier

>> Yeah. The platform shift is... We've kind of crossed that threshold. You took over as CEO for the company. You could have done anything. You could have gotten multiple jobs. What was attractive to Impetus Technologies? What do you see going on there? Share why you were attracted to the company and what you got going on.

Nachiket Deshpande

>> So there are two things that actually really attracted me to choose Impetus over all other opportunities I had. One was data and being in the enterprise data is the biggest differentiator I feel which will make AI work. And Impetus as a company has been focused on working in enterprise data for the last 18 to 20 years. So we have been working in data long before it was sexy to be in data. And that's really the core how the engineering strength and the depth in the company is all geared up. And second, I feel with AI, softwares and services are coming together. And that's how even Impetus was set up. It was always a combination of software and services. Those two attributes is what all companies in the tech services space are wanting to get at. And here, Impetus was always set up that way. So for me, it was a great platform that we would be able to help our customers navigate through this transformation journey much better.

John Furrier

>> A lot of people talk about inflection point, platform shifts. We just mentioned that. Also, we hear the word pivot. I kind of like the word pivot, but on one hand, I don't like it. It implies stopping and turning. Inflection point means a kick up on a curve. I would think you guys are more positioned as an inflection point play because you already had the DNA. You were kind of doing the work going into the market, so it's not a pivot. And a lot of companies has been in that mode of in the cloud native world or coming into AI that have all the piece parts. Talk about that differentiation. It's very nuanced, pivoting versus inflecting.

Nachiket Deshpande

>> Correct.

John Furrier

>> Talk about your view.

Nachiket Deshpande

>> I think you're right about that. When I look at the software and the data, to me, that's how I would differentiate inflection point versus pivot is you are going to double down what you're good at and add a few new things to accelerate onto that curve. Versus pivot, you have to stop doing something and start doing something new. And for us, we're just continuing to be very relevant to our customers in their enterprise data journeys and then helping them use that enterprise data to make sense on their agentic AI journeys and get the ROI they're looking at. So that's the acceleration part that we are focused on.

John Furrier

>> That's awesome. Well, thanks for that setup. I want to get into the future of the agents because the conversations have formed in the industry around models and outcomes and the models really got all the attention. But you saw this year, already seeing it in the first half of the year, all the top players. The models are being decoupled from the actual stack, if you will. So choice is a big thing. And it's shifting towards workflows, architecture, orchestration. That's a systems place. So we're seeing a shift from models to systems. Explain what that means because you guys are... You work with models, but you have the system DNA. Talk about the shift from models to systems.

Nachiket Deshpande

>> As the whole generative AI wave evolved over the last three, four years, attention was on models, and every single day newer model or better capable model keep coming. It almost feels magical in terms of what those models can do. But as you start to get into the enterprises, what those models can do in isolation, they're not able to scale that within the enterprise. And the reason they're not able to scale in the enterprise is because that lack in the system's thinking. Because for something to work in enterprise, you need to be able to understand cost equations and are able to scale with ROI. That's one challenge that the models alone are facing. Second, it has to be integrated in your enterprise own ways of working. And you cannot put a model on top and say, "Okay, now it starts to make your systems or your processes intelligent." It has to be ingrained in your way of working, and hence it has to be more organic and hence the system's thinking is needed. The third, I think every enterprise is unique in its own way. They have their unique customers. They have their unique products. They have their unique processes. Some of them are good, some of them are bad, but they're unique to them. And if you don't take into account that uniqueness and you don't use that uniqueness to your advantage, you are making a commodity play where your competition has access to the same model. So how do you unique? You have to preserve your uniqueness and enhance it with model and how to replace it with model. Again, that requires you to have a systems thinking where your uniqueness is engineered into how the models work.

John Furrier

>> Yeah, that's awesome. Also, we've been hearing a lot about context, agent evaluation. Observability is coming up into the agent layer, which is a cloud native concept. You got to observe, got to evaluate. So context has been a discussion. You talk about the context gap, which I'm interested in. What does that mean? Explain what the context means because people are like, "Hey, just throw more compute at the models. We got inference." Inference is getting cheaper, but the compute's going up.

Nachiket Deshpande

>> We see that in enterprises all the time. Most of the enterprise leaders today face that dichotomy, that on one side they see the models that are becoming so powerful and do so many things with every new release, every new capability. They come to their enterprises and they're not able to get the ROI on those use cases that seems to work in a lab or in a proof of concept environment. Every single company over the last maybe 18, 24 months have done hundreds and thousands of POCs and all of them seems to be attractive at the POC stage. But if you look at not even two, three, 5% of them at best have made it into production. And that gap is because the solutions that work in POC does not work when you try to expose it to real data, try to scale it to all the users in your organization. And as you double click on why that is happening and that's where that context gap comes into play. I feel that's happening because models are capable, but they don't know you. And how do you make the models understand you is the trick to make those POCs into a real value production system. And what is you is made up of, in my mind, three layers of information. First is, of course, the structured data that's unique to you, which is your customers, your products, your services, your transactions, your supply chain. That's the structured data that's...

John Furrier

>> And that's not in the models.

Nachiket Deshpande

>> That's not in the model. That's in your system.

John Furrier

>> That's locked down.

Nachiket Deshpande

>> That's in your system. The second piece is there are nuances to every industry. There are certain ways of working. There are certain regulatory constraints. There are certain uniqueness to how every industry works. So that's your second layer of the context, which is probably not unique to you, but it's unique to the business you are in. So probably shared by 10, 15 or 20 others, but it's still unique. Model still doesn't know that because model is built for millions and billions of people and companies, but that's the second part. And the third, which is very unique, is your way of working. See, how things get done in Impetus is different from any other company and so is true with every other bank and so is true with every other life sciences company and manufacturing company and so on and so forth. And it's important for you to understand or codify how things get done in your organization. And we call those are three layers of gap, right? So the first one is a data gap. The models don't know your enterprise data. How do you unlock your enterprise data and make it accessible and available to the model at the right place at the right time is your first gap problem. The second is the context gap because that data has to be used in a particular industry business process in a semantic context. And the third I feel is your execution gap because that's where you... How get things done in your organization will vary. And all of these put together will bridge that gap between a model that is capable and the ROI that you see within your enterprise.

John Furrier

>> Yeah. I mean, basically what you're saying is that models deliver good outcomes. But if you have the right context, you get better execution that yields to ROI and business performance.

Nachiket Deshpande

>> Correct.

John Furrier

>> All right. So let's unpack the challenges for the enterprise because I like the way you lay that out. Because one thing that we were talking about before we came on camera here was that we both, and you've done extensive work in this area in your career around IT transformation projects. You're smiling because we probably got a lot of scar tissue. IT transformation isn't the same as it is happening today because those layers point to the fact that this is business model, business process up and down the business. Because if data on how the business is run, data on what's the business about, you can use that. So this is a business transformation. This is the top theme in all the narratives we're having on theCUBE. How is that different from IT transformation and what is the challenge for the enterprise?

Nachiket Deshpande

>> I think the fundamental shift with the AI that takes it away from being IT transformation to business transformation is in the past when you were building rule-based systems, defining the system was a small part of the work. A large part of heavy lifting was happening in actually building that rule-based system and making that robust and making it run in an enterprise. Hence, IT transformation was the bulk of the heavy lift in that journey. Now, when building has become trivialized with models, you can build it really fast, really quick, and that's not the bottleneck. The code is not the bottleneck anymore. Then the bottleneck shifts to the intent of what are you building, what are you solving? And that question essentially is a business question is where you need to really understand what is that you're trying to transform, whether it's a business process, whether it's a customer experience, whether it's a product innovation, whether it's a channel innovation. Whatever that you're trying to solve essentially is a business problem and understanding what to solve is important. Another difference is when the technology implementation cycle was wrong, the clarity emerged along the way. So you started with a particular vision in mind. But with every step of your IT project, better clarity, better changes emerged and you got to the end point. Now, when that cycle is largely done by models, you have to have all that clarity upfront because you need to know what otherwise gave you time to evolve that clarity. You got to now sit back in your chair and think about all of those things upfront because that's the intent you will need to go to and give to an AI system to be able to execute.

John Furrier

>> And there's more stakeholders involved because the business includes CFO, Chief People Officer, CIO. I mean, everybody's involved. It's not like the IT guys. So they're involved certainly because the CIO and CISO have a lot of challenges. I want to get your thoughts on shadow AI because I think it's coming together for me. I like where this is going because shadow AI is really built on shadow IT concept where we all live through that. Hey, the IT guy goes around the boss, puts his credit card down on AWS, gets a prototype up and running, gets his hands slabbed and gets promoted. That's kind of grew the cloud native. Shadow AI is happening everywhere. I've seen CFOs closing the books and doing a lot of shadow IT. Shadow AI, I mean. Shadow AI is becoming a feature because what it's doing is empowering people to feel the velocity and execution of agents. And so that's going on. And it's kind of like a kid who's 16 years old who gets his driver's license and gets the parent's car and starts driving around town. It's very motivating. So when you get the keys to the car, you get a lot of power. This is really kind of becoming an operational thing. What's your thoughts on this? Because you don't want reckless driving of AI because people are going to start doubling down on it. Because when they see the value, they just do more of it.

Nachiket Deshpande

>> Yeah. I look at it less as shadow IT, but I would look at it more as decentralized or empowered IT where you're empowering people who are actually delivering value in front of your customer or any of your stakeholders to actually solve the problem then and there, rather than having to translate it to a back office, which is then converting it into this and then coming back. And the pace of change is such that that cycle is too long. Hence, now the empowered IT or citizen developers, as some people call it, is becoming more of an intentional model than something that's happening under the radar. And hence, the role of the IT or technology in an enterprise has changed. The role of the technology in an enterprise is to now enable that, enable that with right guardrails, enable that with the right financial ROI metrics. So that, continuing your analogy, when a 16 year old is driving that car, he or she now knows what are dos and don'ts, what are the best practices, what would keep them out of trouble, so to speak, while not taking away their flexibility and saying, "I will drive with you every time you want to go." But saying that here are the...

John Furrier

>> You got to pass the road test.

Nachiket Deshpande

>> You got to pass the road test and beyond even just the rules, here are things that you got to keep in mind in being a successful driver and protecting yourself and the others that are...

John Furrier

>> I love the car analogy. I was watching F1 in Miami last weekend and of course, all the tech companies sponsor the car. So highly interested, and plus my son's a big fan. I'm somewhat of a fan, but the technology is the car, humans are the drivers. I love that analogy for AI. I know you and I used that before. Everyone pretty much agrees in the industry today that I talk to that technology's not the problem. Explain how that pivots out because now we're getting cars for everybody, not just the IT guys. So talk about that technology car analogy. It's going to get better. We see the improvements on the roadmap. You guys have stuff coming out. But the human driving, that ultimately is what you want.

Nachiket Deshpande

>> So I think there are two parts to that story and I'm taking that car analogy forward. If you are driving a normal car on the road and if you go to an F1 track and sit in an F1 car, then very likely in that first three seconds you're going to crash that car somewhere. And not because the car has a problem and not because you are a bad driver, it's just that as a driver, you're not trained to drive that car on that track. So a lot of this about AI and AI agents is like that. It's a very, very powerful machine, but most people and the processes and the systems aren't trained to handle that power. So if you don't make that change management work first, you are likely to get a lot of crash. And learning it by doing and bumping it a few times is a very expensive way of learning, especially in an enterprise when you have so many stakeholders they're accountable to. For me, the adoption of AI, one is, of course, the car itself has to be tuned for the track, which is where we talked about the enterprise context and getting it right so that it's not just that the car which is in the manufacturing plant works in the lab, it also has to be tuned for that track. And second, you've got to do change management and training of drivers to be able to handle that power, so that you are able to use that power to get the speed and the timing you want. Otherwise, it will be a very powerful car, but that same distance of a mile will take you forever because you will probably...

John Furrier

>> You don't want the Ferrari to go down and get the milk at the corner store. You want to use that on the open road. Love the car analogy. Let's extend the car analogy to agents because I used to drive my car to get food. Now the food comes to me. The agents are Uber and DoorDash and all these services. So autonomous is coming. We see it in cars coming in. Agents are autonomous humans working, digital workers. What's the agentic reality? Because now you're going to have, okay, let's get trained on how to drive the technology. That's the AI. So talk about the human AI operating model for leadership. Because as you guys roll out your platform and as the platform gets more powerful, you'll have agents working on your behalf driving their car. So talk about that connection and what's the reality of where that sits with agentic.

Nachiket Deshpande

>> So there are two components to that autonomous agents related challenges and opportunity. I think number one is it's very important that within the enterprise, you have a very clear idea that what tasks require a deterministic approach and what task requires a non-deterministic approach. The models are extremely useful and powerful in a non-deterministic world. But within an enterprise flow, there will always be certain things which will need to be deterministic. They could be influenced by regulatory. They could be influenced by the risk. They could be influenced by many other factors, but you will have that. So when you are designing your agentic strategy, it's very important for you to know what parts of my agent execution and capability is going to be deterministic, non-deterministic. That's the first part of that. And then the non-deterministic part that you want it to continuously improve and learn, that's when do I switch it to become autonomous where I have enough trust in that, enough controls to maybe able to do that. They will vary from industry to industry and process to process because there's certain industries where segregation of duties is extremely critical. And it is for the protection of all the consumers in that. And hence, you would not want to put autonomous in such scenarios, but there are lots of back office tasks that actually you would be very benefited if it was done autonomous. Because in a way, no humans should actually do them. It's just that because you didn't have an option, humans were doing it. It's not a very value added service from a business perspective. So you would need to look at that. So I think the selection between autonomous, non-autonomous is important. And it then comes to the same thing, the FSD driving versus manual driving in a car. One, it will based on your own risk appetite. You will not do that on a New York street, you will do that on a freeway. And that's the difference where I said where you apply and where you don't apply is how I look at it.

John Furrier

>> Nachiket, talk about the industry implications as to where we are on the progress and what's next. Because as agents comes, where are we on the progress bar of agentic reality and what's next? How do you see the industry evolving?

Nachiket Deshpande

>> So I feel that the technology is very much there. It'll keep improving, but I think the technology that's available is good enough for the agentic adoption to scale. But where the gap is, one, as we spoke about, is the context because the enterprises need a lot more preparation and background platform work to be able to make use of that technology in a meaningful way. So there is that curve and that wave of investment that has to happen. Second, I think we also don't know how to govern, monitor, and work with these autonomous systems that come in. So it's like, if you take the same car analogy, your road infrastructure, your signaling infrastructure, all of that also has to improve if every car is going to be as fast as a F1 car on the road, just today not. So that's the other piece that the governance of the systems and how do we really manage and monitor agentic systems that needs the evolution. And third I feel is, like we said, all the drivers who are going to drive those cars or even if there is a part of the traffic is going to be autonomous drivers and part of the traffic is going to be human drivers, I think the humans need to know as much about the autonomous cars and the autonomous cars also need to know enough about human behavior to coexist. So there's a lot of change management, training, familiarization, a lot of that is needed on the humans as well for that to happen. So those would be the three things that I believe would be the focus over the next 12 to 24 months before we see a true adoption of agentic in the enterprise.

John Furrier

>> Well, certainly a lot of demand. What's on your roadmap? What are you focused on now? What do you got coming up?

Nachiket Deshpande

>> So for us, I think our superpower has always been in the enterprise data world and we want to use that superpower to solve that context gap problem that I talked about because we understand enterprise data intimately well. We understand models very well. So we are in best position to be that bridge to be able to help enterprises close that gap. And as I said, our approach has always been software and services together. So what you would see from us over the next year or so is a lot of software and services capabilities that we will roll out that will significantly help enterprises bridge that gap.

John Furrier

>> Any teasers for the big news? Try to get your pin down here. Got some big news coming up.

Nachiket Deshpande

>> I think yes, we have a big launch coming up. Our set of solutions that we believe will significantly accelerate that journey across the data gaps, context gap, and the execution gap for our clients. We would also launch a lot of industry aligned examples and use cases that will show it, not just give them tools, but also give them examples of how it actually can be realized in specific industries. We've been working with some of our beta customers already in the background. So I'm hoping all of that would unveil and we would be able to see a lot of traction with our enterprise customers on the journey.

John Furrier

>> Well, the conversation has shifted from models to architecture. Congratulations on the momentum. Thanks for coming on. Appreciate it.

Nachiket Deshpande

>> Thank you.

John Furrier

>> All right. I'm John Furrier with theCUBE here. The future of agents and the agentic infrastructure is emerging as AI infrastructure continues to build out, a lot more systems thinking, a lot more platform shifts continuing to happen. We're doing our part to bring that to you. Thanks for watching.