Data Analytics Chat

How Gen AI and Agents Are Reshaping The Game

• Ben Parker • Episode 58

Share episode

Send us a text

In this episode of Data Analytics Chat, we sit down with Dushyanth Sekhar, Head of AI and Data Platforms at S&P Global. Dush shares key insights from his career journey, emphasising the importance of learning from failures in the AI space. 

He discusses his transition from a non-technology background to leading AI and data initiatives, the role of continuous learning, and the significance of setting a vision and promoting a risk-taking culture in leadership. 

The conversation highlights the strategic implementation of generative AI and agents in transforming data workflows and enhancing company growth. 

00:00 Embracing Failures in AI
01:07 Introduction to the Podcast and Guest
01:50 Career Journey and Early Beginnings
03:03 Transition to AI and Key Learnings
04:19 Defining Moments and Career Decisions
05:33 Moving to the US and Professional Growth
10:16 Leadership Insights and Risk-Taking
17:19 Staying Updated in a Fast-Paced Field
19:36 Excitement for the Future of AI
21:10 Overestimating Risk and Underestimating Opportunity
21:54 The Evolution of Language Models at S&P
23:49 Integrating AI into Workflows
28:31 Enhancing and Creating New Products with LLMs
30:54 Future-Proofing and Optimising with AI
37:19 The Role of Agents in AI
40:19 Lessons Learned from AI Integration
41:40 Conclusion and Final Thoughts

Thank you for listening!

dushyanth:

Failures are a part of, part and parcel of AI and transformation. I think the key here is to learn fast from your failures. And so that's been very critical to this journey that, you will fail. I think in the AI world, the way I see this is you'll have to make many bets and maybe a lot of those bets will fail, but the ones. Which succeed, have much larger payoffs. And you'll fail, but you have to pivot fast and learn fast. I feel leaders in the future, or even today, have only three jobs, right? They set the vision, they remove roadblocks, and they stay out of the way. Absolutely. I think I'm not a big fan of fail fast. I think the key is to learn fast. If you fail fast and not learn anything it's not, it doesn't help much. Your job is to set the vision and remove roadblocks. And lastly. You, not literally, but metaphorically, you'll have to go to school many times in your life, right? So learning has to be continuous,

ben parker:

Welcome to Data Analytics Chat, the podcast where we discuss the world of data, ai and the careers shaping it. My guest today is Damp Sekar, head of AI and data platforms at s and P Global, where he is playing leading role in transforming their data platform. In this episode here, we'll unpack his career journey, the moments that have shaped his career and share the insights he's gained to anyone looking to grow their own career. And also, the date topic for this week is how gen AI and agents are reshaping the game damp. Welcome to the podcast.

dushyanth:

Thank you for having me, Ben. Happy to be here.

ben parker:

Pleasure. So do you start off with yourself giving a background?

dushyanth:

Sure. My name is D Shekar work for s and p Global. At s and p Global as a part of the Enterprise Data Organization I lead the data platforms and the ai. Which basically means that from the time data is available in its raw form until it's has some commercial value, the entire pipeline and all the tooling around it my team takes care of that.

ben parker:

So then what?

dushyanth:

Oh, it's actually, it's, that's interesting, right? I never came from a traditional technology background, so I started my career, actually, my first job was I was replying to. Amazon customers back in 2000. So I was like I was working for a company which were doing customer service for Amazon in India. So that's where I started my career. So I don't come from a technology background. And over time I joined s and p in 2009. Prior to that, I used to work with an insurance company, have some background in actuarial science. And 2009 I joined s and p as a part of their ratings group. Was doing multiple roles across data and analytics. And around 2015 when, the first generation of machine learning and robotic process automation was coming in, that's how I got interested in this space. I always had an interest in math and statistic. Which was a good, which is helpful because a lot of what happens in this space under the hood is all math. And over time that role transformed and here I am.

ben parker:

Yeah. And what, what drew you into the field? Was it just any particular interest or did you fall into it?

dushyanth:

That's very interesting. So actually what happened was around 2014, 15, I was used to manage a large data team and I was and I was very interested with the new age technologies, which were coming up. And fortunately for me, the person I was working for that time, and I had expressed some interest to do some work in this space, and he was kind enough to create a role for me. So for about a year and a half, I was just by myself trying to figure these things out. So the interest was basically, I think it was a mix of, I, I was at a stage of a career where I wanna do something different. And maybe some luck that these technologies were coming to the fore at that time. So it was a natural marriage,

ben parker:

okay. And then, so looking back, has there been a defining moment when you've realized that this industry, was the right choice for you?

dushyanth:

Absolutely. I think when I moved the first made the move, a lot of the people I spoke to was, the technology was still very nascent at that point, and they felt I'm taking a very big chance and I should just stick to what I'm doing. But yeah, so there was a little bit of pressure on me that I, for a initially couple of years, I was like, okay, did I do the right choice? But I think. The defining movement for me, if I look back and say, oh, it's difficult to pick one. But, the first supervised system we built, this was the first one we ever got into production with. This was, I think, way back in 2017 or 18. I think once I started seeing the results from what we had built, I think I, some of those apprehensions were less, and I thought this is the future. So I think probably that first production run of our, we built a supervised information retrieval system. It was small, it was not a big platform, but I think the success there probably was the defining moment.

ben parker:

Okay. Amazing. And then obviously you've obviously come from India to the us. How was that for yourself? Was that sort of a big challenge for yourself?

dushyanth:

So I have been I have worked in the UK in the past, and I have been around, but yeah, I think that was the right time to the move to the US because the technology was on that gradient that was moving up. And I think with all the innovation happening here at that point of time, I think it was the right move. And also from, from a mindset perspective, I think I, I think coming here helped me a lot, right? I was much closer to the customers. I was interacting with a lot of smaller companies, which were in the space. So I think all of that kind of came together when I moved here in 2018. So I think moving here in hindsight has been beneficial. Of course, I miss home, but professionally, it, it has been very enriching for me.

ben parker:

Yeah. No, I agree. I think obviously moving does add to your experience in life. So what with yourself there, obviously you've had a great career worked with some great companies or so you, have you, is there been a period or like a failure or learning that's turned into an important lesson for you in disguise?

dushyanth:

Absolutely. Failures are a part of, part and parcel of AI and transformation. I think the key here is to learn fast from your failures. And so that's been very critical to this journey that, you will fail. I think in the AI world, the way I see this is you'll have to make many bets and maybe a lot of those bets will fail, but the ones. Which succeed, have much larger payoffs. So unlike traditional software development cycles where, success is fairly guaranteed in most cases when say you're just building an application. But in the AI world, it's not like that, right? The problems are all linear in that sense. So sometimes we'll start with a problem, take an approach, it doesn't work. And you'll fail, but you have to pivot fast and learn fast. So I think that's the kind of the cultural side, which has to also shift in organizations where such experimentation and failure is encouraged and tolerated. And I think s and p has given us that freedom to go and go for the. The big price. And as a part of that, you do fail at times. So that's, you just have to take it in the stride.

ben parker:

And is it, can you, can you pinpoint a specific example which has really been critical to your learning?

dushyanth:

So when I initially started on this journey especially for the, on the information retrieval side, so we were, grabbing a lot of information from a. From big financial documents, right? These are three, 400 pages long. So my initial thought was I could just build a system which will just do everything and everything will be fully automated. But like within three or four months I realized that, the best way to work with these systems is to. Cohabitate these with the human intelligence and not supplanted. So I think that was a big learning from me, that taking a more human plus machine approach rather than looking at it as a choice between a human and a machine. So that is a big learning for me.'Cause initially I, you are always little bit, these technologies are high. So you feel you can just code your way through it. But I think that reality dawned upon me pretty soon and that was really helpful, right? Once we realized that the better architecture was to be able to invoke the human as needed, and then things became more clearer.

ben parker:

Yeah, so obviously, yeah, we obviously AI is gonna help us save time and then, so it's not gonna completely wipe out the human nature with taking all our jobs.

dushyanth:

Yeah, that's a very interesting point you make, right? Because do you see a job as a series of tasks, right? That's the key question. Like tasks will get automated, but your job is just not a series of tasks, right? Once some, those of those tasks get automated, it also frees up you to do other stuff. So I think that's the way to think about it, that giving that time back to the humans and AI helps doing that, and the humans can then use that time and do more creative stuff.

ben parker:

Yep. No I completely agree. I think obviously it's getting more, obviously, I guess getting humans whilst more strategic thinking to do the thinking, which is obviously what we should be doing.

dushyanth:

Yeah.

ben parker:

So what quality or skills have made the biggest difference for you stepping into leadership?

dushyanth:

I feel leadership has, I think, and I think at a more strategic level, you have to set the vision. So that's important. Execution is important that the vision has to be ex executed and. What underpins these two is the culture, right? That, so at a strategic level, those are the way I look at it. But on a more day-to-day level, I feel leaders in the future, or even today, have only three jobs, right? They set the vision, they remove roadblocks, and they stay out of the way. That's the last part is always slightly difficult because I, I had to not. Maybe relinquish some of the stuff, which I said I can't micromanage everything. So that's been a journey. I've not yet perfected it, but try to get better at it. But yeah, I think those are the big lessons. And also, create a culture where people are willing to take risks, right? So I think that's very important in today's day and age that you know if people are not, are risk averse. Then innovation is not going to happen. And a culture where people will be willing to take risks are when they feel okay there now no punitive action if they fail. So I think that's very important in today's day and age. And which means also empowering our their, your teams to make those decisions, right? So I think those are the few things which probably shaped my leadership.

ben parker:

Yeah, no, I agree. I think you sh, I mean you as a leader, you should be pushing your team to push the boundaries and not necessarily ex promote failure, but obviously us just promoting the learning, isn't it?'cause if you're gonna keep, you need to push the boundaries basically.'cause things do change. You need to evolve as everything else evolves around you.

dushyanth:

Absolutely. I think I'm not a big fan of fail fast. I think the key is to learn fast. If you fail fast and not learn anything it's not, it doesn't help much.

ben parker:

Yeah, you don't ground dog day. So why, what's that? Who, or has there been a person that's been the biggest influence on your career or has it been technology? And why?

dushyanth:

I think of course professionally, there have been many people I've worked with. Some of my, some of the leaders I've worked with have had a lot of influence in me, on me, but I think it's an interesting story, right? I used to play a lot of chess. I still play. And I think it was 90. I never thought about it at that time, but one of the things which really got me interested into machine learning per se, was. I was a big fan of Gary Kasparov. And I think Deep Blue was the IBM engine, which beat him in 94, I think, or 95. I don't exactly remember, and I was really upset. Because still then, just the, today it's pretty much par for the course that engines beat humans very easily in chess. But at that time it was like a watershed moment, I just couldn't believe how. How a machine could be the grandma. And then I got into kind of learning, I, that got me interested into this whole machine learning stuff, right? Of course that was a very old technology, which I think IBM never I think it that deep blue got killed after that match. But I think that probably it, that always was on my mind, right? How this was even possible, right? And. Now when I look back, it makes more sense. But I think that was probably something which it still is back of my mind that it's unbelievable to, 30 years back this happened. And I think that did have some influence on me choosing this path. Maybe I didn't realize that time, but now if I think back, I think that did play a part in at least getting me interested in this space.

ben parker:

Yeah. That's amazing that, and obviously yeah, it'd been obviously at the time it had been, yeah, like crazy to think

dushyanth:

Yeah, it was very crazy, right?

ben parker:

like now it be around, it's crazy to be a human could beat. Robot or

dushyanth:

Yeah. Yeah.

ben parker:

So obviously, so you've progressed into leadership then, and say if you were mentoring someone who wants to step into become a leader, there's many that wanna move from the technical ex expert into like leadership. What what's your, what advice would you give.

dushyanth:

I think if somebody wants to take more leadership roles in, in the AI world, right? Or kind of work on AI platforms, I think one thing which is important is the fundamentals of the business don't change, right? You still have to. Look at what customer problems you're solving. So AI is not working in a vacuum, right? So one example, I think I was speaking to one, one professor who was pioneered some of the research in the seventies and eighties around ai, which was more conceptual. And he was telling me this, that let's assume that you are, you have a business which sells hot dogs, right? And one day you get a AI chef or a robot, which makes the hot dog. Customers would come, they would be excited probably to see a robotic hot dog vendor, but eventually they'll only buy your hot dog. If it tastes good, it's hygienic. And so those things don't change in the AI world, right? What the customers expect of the product. So customer are not going to accept a hot dog that doesn't taste s well, just because it is made by ai, right? So that first principle has to be. Etched in your mind as a leader that you still have to satiate that customer need. Now, you could do it better with ai, but customer is not just going to pay for a product just because you use ai. So that's important. Secondly, I think you ought to keep an open mind. AI only works if you can. The full value of AI only comes if you're reimagining the workflows. So as a leader, you will have to have some system design skills and understand the overall system rather than just looking at the technology side of this, right? So that's second And thirdly, as I mentioned, more on the softer side. I think you will have to relinquish more decision making to the people who are closer to the problem. Your job is to set the vision and remove roadblocks. And lastly. You, not literally, but metaphorically, you'll have to go to school many times in your life, right? So learning has to be continuous, which probably in the non-AI world, was not very pivotal once you read certain leadership positions. But today it's not like that, right? You, the space is evolving so fast that you have to keep yourself abreast. So I think those are a few things which I. Probably would be my inputs to anybody foreign into this space.

ben parker:

Yeah. No, I agree. I mean it's, yeah, constant learning. And how do you go about, as a leader, how do you stay afloat with the changes because it is moving rather quick.

dushyanth:

It's getting hard, right? I'll be very honest with you, the amount of information out there. Now to just to consume it. Just to, there's also a lot of noise, right? So you gotta make a choice of what you want to read and spend your time. I do more micro reading now than something I learned from someone I was talking to that, I don't read the entire stuff. Of course some of the LLM tools help summarization and things like that. And then the topics which I find which are more relevant or interesting to the work I do, I dwell deeper into. So that's important. The other thing is important is also to reduce bias, right? Because you wanna know both sides, both views, right? And not get subsumed by the hype. So that's been helpful, but to be very honest, that can I keep up with everything which is going on? No. So that's where the teams come in and they fill in the vacuum, but yeah, so I, I try to spend. At least a few hours every day just after work, just looking at the news or reading some of the research papers. But yes it's, the information is much more than time, so that's always a challenge.

ben parker:

Yeah. No, and I think it's obviously it's so much to take, obviously our brains have, you can't take in, you can't remember everything. It's like that's not changed. So you need to, I think it's where you specify your team to. Everyone's gotta be their expert in their own rights. Like you can't know it all. And I think this is we, our leaders, you need to have faith in your junior members, in your team that have got the more specialist knowledge to get the right answer because you just can't, there's too much going on.

dushyanth:

Exactly right. Maybe say two or three years back, I would more in the weeds with all the projects and initiatives, which is going on now. It is just not possible, right? It's I think if I try to do that, I just become a roadblock. So sometimes you are absolutely right. Sometimes you just have to step back and trust the people you hired to do the right thing.

ben parker:

Cool. And then so what's ex, what excites you about the future, I guess the next chapter of your career?

dushyanth:

I am an optimist, so I'm very optimistic of what the language model the agents bring to the table. Of course these are uncharted territories and there are always counter views, but I feel we are in a similar situation like we were in say, late nineties with the internet, right? I was watching one clip of Bill Gates, I think it was on a David Letterman show. It was 95 or 96, where, bill Gates was asked, what is this internet? And he just said, oh, it's a page where you can go and see some information. So I think we are probably in the AI world at a similar space, but I think the future is, I think the opportunities and what we can do with it, not just in the work I'm doing, but you know how we can do good in the world overall, I think the opportunities are limitless, so I'm pretty optimistic and I think that's exciting. The other thing which excites me is. There are like, just the excitement in the teams, right? Every day we have this kind of a teams channel where I, every day people are like putting the stuff they're experimenting on or sharing ideas. So that's, that the whole buzz around that is very invigorating. So it's so I don't feel stressed, even if you have a long day, you are doing so much exciting stuff that. It feels gratifying. So that's, those are a few things which I'm excited about. How this will play out, I think we can all have our crystal balls. But I think in the long term the, I think humans have a tendency to overestimate risk and underestimate opportunity. But, I think in three or four years back, when we look back at this conversation, I think the world will be very different.

ben parker:

Yep. No, and I think it's exciting times and I think, yeah, it's, yeah, it's just interesting, right? Things are evolving so quickly and it's, I think it's inter, it's good. Great to see. And it's obviously just gotta take it step by step, haven't it really?

dushyanth:

Exactly. Yeah. Just let it play out. Yeah.

ben parker:

Okay, brilliant. So obviously LMS Genis has taken all the hype over the last couple of years and probably will do for the next for foreseeable future. So obviously you've been involved in this space. So from the hype to like utility, how is s and p moving beyond, like just the buzzwords with LMS and Gen ai?

dushyanth:

So actually one of the things s and p has been on this journey even before these buzzwords came into play, right? We had a language model in production somewhere around 2018 or 2019. Of course that time the word LLM and the transformer architecture are not come in. We were just calling it a model, right? It was more library based training, but it was a small language model for one of our, for data sets, which we had trained. So I think, so we have seen the journey from very rule based systems like RPA to supervised systems. So we were al already on that curve. So a lot of this has not come as a big surprise unlike other industries where. Data is more a byproduct. In our case, data and analytics and workflows are the product, right? So we were been dealing with big data even before the term big data was coined. So the ecosystem was already there, right? So it was easy to see where the opportunities are and cut through some of the hype. There's always hype. I think these cycles. Or have to play out. So that's something at s and p, we have been and we have been very focused on also upskilling our workforce on these technologies. Even seven or eight years back, right? So we have these essential tech programs where we bring these technologies to the masses, to all employees. So we have been on that journey. So that's the kind of the people and the strategy side. On the actual work side, I think. The, these technologies are blend very well with the work we do. We are in the data world. We have a lot of data to provide these models, and that's definitely a competitive advantage. And I think the other ingredient is that which is becoming more and more important now. The subject matter expertise around those workflows and ecosystems, right? So we have one like world leading experts on data analytics ratings index. So when you bring all these things together and then also unleash the power of the LLMs the value proposition to our customers and investors is very strong. So I think that's how I see it. Of course we can get into specific areas on where we are implementing these technologies, but a high level I think s and p is front and center of this revolution.

ben parker:

Been. Doing the project, obviously for a while now. So you've got, it's an understanding. It's like not just hype, is it? It's just you need to just, it's, there's a lot of work that has to go into it, isn't it? If you said you started up 2018 and still working on it, obviously it's a constant evolution and especially when you got companies out there, just wanting to just put L lens in straight away without doing the backgrounds is and you've got, there's a lot of work goes into it, isn't it?

dushyanth:

absolutely. As you said, it's an evolution, right? Probably I should not have said revolution. More of an E. You are right, that see the AI. You could also always parachute in an agent or an LLM and you'll get some value, right? It's not that. But if you truly have to, make this as a competitive differentiator it has to be embedded in the fabric of what you do, right? It can't be the standalone thing, which you just have an API and everybody's just calling it, right? So I think that system design has been core to how we look at these things. So I'll give you a couple of examples, right? Like. When you look at, very labor intensive processes, those processes are fundamentally based on the principle of division of labor, right? And it's, and it's understandable. So 20 years back, we are take a very complex task. You would break it down into chunks, right? And say, okay let Duchen do this and Ben do this, and somebody else do this. Now when you try to morph that process in an AI world and you don't fundamentally design the process for a machine you're not gonna get the benefits, right? So you might have to do some system to redesign or lean the process out before you cut a parachute in the technology. So I think we had a very strong culture of lean even before, these technologies came up. So I think now with that and the technology, we can, we are able to blend it, especially in this workflow we had, in, in a human workflow this complex workflow I was talking about, it had 20 components, right? But when we moved it to an AI world, even before we did anything with the technology, we re-engineered that process to only have three workflows. And then the AI came in and gave us a lot of benefit. But if you had left the process as is, we would've still been successful, but the return would have been lower. It would not have been as much as when we re-engineered the workflow.

ben parker:

Yeah, and I'll ask, I'll speak to a lot of leaders and they say, obviously the LLM is the easy bit, or it's not it's still hard, but it's like the hard bit is actually the, all the stuff behind the scenes that has to be done.

dushyanth:

Yeah, all the stuff, which is not cool. So that's where the Yeah the value is.

ben parker:

it's always the ways of it, even like in sport, like you have to do the training behind to get to the. To the level of excellence.

dushyanth:

Exactly. I think that's a great point, right? If you're watching the premiership, rather, you only watch the 90 minutes or what but all the stuff which goes beyond. Behind that before the match. I think Adobe sees that, right? Yeah,

ben parker:

Yeah, I think it's like Usain Bolt used to say I trained four years to run Nine seconds. Like it, it shows. So it's it's similar to that.

dushyanth:

Where you bring that up. I think the first bot we put into production in 2016, something, we called it Ussein Bot, actually. Yeah, a little bit of

ben parker:

so we'll see SP Global, we'll see a massive company, obviously massive data at scale. How are the, how are LLMs changing the game across the platforms?

dushyanth:

There are variety of initiatives. I'll, but I'll give you a broader view on this, right? So as a part of the EDO, I think one of the, kind of implicit assumptions of bringing the IDO together was to leverage the power of data technology and subject matter expertise, which we have within the divisions, right? Bringing the that to the center has been very powerful. So that's where while I start Now, on a strategic side, we are focusing on three areas, right? One is growth. So when I talk about growth, you can look at LLMs on how it can enhance existing products. So we have things like if you are a subscriber to our Capital IQ Pro platform, we have now a chat based interface where you can do some very complex queries just through natural language, right? These were workflows where probably users had to click many times. Now they can just interface with our. Huge database through a chat interface and get a lot of information out. So that's an example of kind of, enhancing an existing product. Then we also have our ability to bring new products to the market as significantly improved with LLMs, right? For example, the. General life cycle of going to market with a new data product has come down. Our ability to mine information at scale has significantly gone up. Like one of the areas we do private for private market is a big area of investment for us. And one of the initiatives we worked on was, how do you do probability of defaults for private companies where there are no financials, right? So you basically go with semantic footprint of these companies on the web and triangulate that data with other publicly available sources, and then come up with some quantitative model to do a pd. But in the, in a non LLM world, these products would've just not been possible. We are talking about 30, 40, 50 million private companies. For humans to manually mine and triangulate that data would have just been impossible. So on the growth side, we are enhancing new products and we are also building new products which are powered by these technology. The second area is where, what we call speed and optimize, right? We have to optimize the back backroom engine, which powers all these products. So where I get into information retrieval, linking we have a lot of classification type of problems, but our approach is to build platforms with scale across these use cases. So that's on speed and optimize, right? So one is grow the business ops, optimize what's already there. And the last part is the interesting part, right? What we call future proofing, right? We, how we future proof our products. And people both. So those are the three areas which are basically targeting and then some, there are some interesting work happening across all three.

ben parker:

No, I love it. And then it is good that you actually focus on our business terms, if you get what I mean, like growth.

dushyanth:

Yep.

ben parker:

Efficiencies because a lot of businesses are just implementing, they want LLM just to be part of the crowd. And this is feedback from a lot of businesses. So it's end day, I think you still need to remain to your core. Whatever situation you are in as a business whatever plan you have is you still need to focus on business, don't you?

dushyanth:

Absolutely right. As I said that if you're a hot dog seller, eventually you have to make good hot dogs, right? So that's the key that, and that's how you could give value to your customers, investors, and stakeholders, right? So that fundamental principle doesn't change. So when you apply ai, keeping those outcomes in mind, the success is much better. But you are right If you. Just do it for the sake of doing it. And because it's a new shiny tie, actually in technology, I don't remember the exact name. It's some kind of an anti-patent, right? When you're just using some technology, which just because it's available, it's actually an AntiPattern.

ben parker:

So what what does a like reimagined workflow look like in the age of ai?

dushyanth:

Yeah I don't know. I may not have a definitive answer, but what I could say is that I think it'll transform a lot of things we do as humans. And people who use AI will definitely replace people who don't. So that's the fundamental principle, right? Irrespective of what you do. These technologies will touch your work or world, right? It's, you're not, you cannot go to be insulated from it, alright? You could always argue that there are some stuff which AI cannot do, but I think the, keeping those anomalies aside, I think anybody in any shot of profession today is already being touched by ai, right? Like basic things like copilot, right? Everybody who's. He's a user of certain technologies, will have a copilot. So those things, so that's, it's more, it's not a thing which few data scientists are doing in some back room. So everybody's touched by it. So that's, something which gonna come, right? And now how it'll transform the businesses. I think the jury is out. What new businesses will be created? We don't know, but we definitely know that's going to be the state of play, right? That I don't see a business in the future, which doesn't embed AI in some shape or form into their workflow. And the customers are demanding that as well. So that's it's from both sides. So that's

ben parker:

how would, yeah how would be the best way to embed it then?'cause obviously you've got this new, fantastic technology. Businesses have got their old, the old ways of working. Then they've got this new technology that they need to implement. How's best to embed it?'cause there's a, obviously there's a lot of knowledge transfer and learning involved isn't there?

dushyanth:

Yes. So I think that's an interesting point, right? Like in effectively you, you changing tires on a running truck, right? So you can't just stop the truck and say, okay, I'm gonna pause my business for a couple of years and then reincarnate in the AI world, I think that's not gonna happen. So you have to keep running what you're running and. At the same time, bring these technology, which is easier said and done, because there are, the reality of the world is there are still a lot of legacy applications which you have to run in, especially in large organizations. So I think the fundamental principle is to be redesigned, the systems itself and the workflows, right? So that's where it has to start. And. And that has to be driven by, what your customer is needing, right? So once you blend those things together, then you see where AI will help you, right? But if you just start with the AI first, then it's going to be a problem, right? Then you might have a lot of solutions, which you deploy. But will you deliver the full value to your customer in investors? That's not gonna happen if you just do point solutions. So think platforms, think system redesign, and always be ready that something will change, right? So especially with Al Lambs, you see every three months, just the progress they make is pretty significant. So you have to keep your options open so that if tomorrow you have to pivot or change some components in your ecosystem, that process is fast.'cause Oles is much faster route unlike say 10 years back. So you'll have to be prepared to be able to pivot very quickly when the technology makes very smic shifts, because those shifts are not linear anymore, right? The. You know what an LLM could do in a year's time. It's very difficult to predict at this point of time.

ben parker:

Yeah, no, I think it's key for people now to be adaptable because, and even in the past, people used to be set in their ways. They only want. This way, but now you, you constantly gotta keep learning and evolving, haven't you?

dushyanth:

Yeah. So I think that's a cultural part of that as well, that. How do you have a large workforce and how do you create a ecosystem where that learning is continuous and it's not seen as a standalone thing, right?

ben parker:

Okay, cool. And obviously there's a lot of talk around agents obviously the future of automation. What's your thinking around this?

dushyanth:

Yeah, I think the agents are the next kind of frontier of the LLMs, right? When the LLMs cannot take actions on their own, right? So in the real world you know what humans do is take actions, right? They take some information and they have to act on it. So I think the agents come in to in that last mile space, right? You say, okay, there is. So think of this like an agent could be interacting with some tools, some resources, which is some data, and then some prompts and things like that. And then also can be able to take some action. So the agents are gonna be very critical to this equation. But again, the principles apply, right? That do you just want to parachute an agent? Where it does something on the side, or do you see agent as a part of your larger ecosystem, which is like in working with all these tools, resources, and prompts. So I think the latter is the way to go. Of course it's there there's a lot of stuff happening. We have about. 20, 25 agents in production right now. And the last part of the equation is also the, how do you pivot your workforce? On the technical side, right? Developers were always, their output was code, right? So today that equation also changes a bit, right? That I feel today the outputs for developers are more structured communication. So they are solving problems, right? A code is a very. Like a small part of that equation. So I think these are the areas where, you know, if you look at traditional development life cycles and you expect these people to pivot and build agents, you also have to start thinking differently about how their work is, right. They have to be more interactive with the customers. And every time you put something into production, the measure of success is not just how the code runs. But when the code runs, is it solving a customer problem? Which becomes the next frontier to capture?

ben parker:

And these, do you feel they should, it's more important to build these from scratch than like out the box because you've got, there's obviously the benefits of out the box, but I think. So important, isn't it, to build from scratch in view.

dushyanth:

Yeah, in some of the things, yes, you just have to completely rewire but in some things you might still get some benefit having some point solutions, but you are right. The general approach will be to build platforms, which are not just individual agents, which are running all over the place.

ben parker:

Cool.

dushyanth:

The agent has to be a part of an ecosystem, right? It's not some standalone thing.

ben parker:

Yeah. No, definitely. So what lessons have you learned from integrating AI into your environment?

dushyanth:

One lesson is the last mile integration is much more complex than it appears to be because sometimes you cannot like rewrite all your legacy. Stuff, which is in the periphery, right? Sometimes you'll have to take some tactical call and kind of evolve. So that's a big lesson. And the other thing is, eventually models however they good are, they are stochastic in nature, right? The outcomes are, the outputs are probabilistic, right? So you will have to be able to invoke humans in where. In some workflows, that probabilistic output has to be more deterministic, so that human has to come into the picture. Now, how much, where and when could be very workflow specific, but it's a gradient, right? It's not a binary kind of condition that, okay, you take a workflow and totally move it to an agent, or you don't. So on that gradient, there could be some workflows, which could be require very little human touch. But then there could be some workflows where that might, that human touch might be more so you have to look at it as a gradient rather than a binomial thing.

ben parker:

You've been fantastic. You've shared some great insights. I'm glad you've yeah, done some great things around the sort of generative AI space, and I really appreciate your time on today's podcast.

dushyanth:

Thank you, Ben. It was great chatting with you and some great questions. Appreciate it. Have a great weekend.