#13: The Real Future of Work - Straight from Microsoft’s AI Lead: Nitin Aggarwal

Speaker A: There is this fear of like AI is going to take over all of these jobs.

Speaker B: A lot of roles will be disrupted for sure. That problem remains the same. The face of the problem has changed.

Speaker A: Innovation does require humans. It does require people that imagination or like something that you think it's hard to replace with just AI. What's the best way to prepare oneself over the next like 5, 10 years as an individual?

Speaker B: I will strongly recommend three things. First thing is Nitin Agrawal is the current leader of generative AI at Microsoft. He is responsible for transforming marketing using AI and developing a large scale platform with cutting edge features for marketers. He also played a pivotal role in developing and executing executive strategies for Google's Fortune 500 customer. My journey started as an accident. My first project, when I got it, I got it because nobody wanted to work on that project.

Speaker A: You started working on something that nobody wanted to. There are so many nowadays also videos being generated by AI. And you see this a lot in like the short form content videos and sometimes it's hard to trust a video. And I think that's where it's also important from us as people who have been in tech or people who develop tech to figure out that, you know, how do we make sure that the trust factor is not eroded? Hello, welcome to the Priyanka Shinde podcast where we dive into the art and science of strategy, execution and leadership. I'm uh, your host Priyanka Shinde, author of the Art of Strategic Execution and a consultant for, for tech leaders. In this podcast I bring you candid conversations with top executives, startup founders and industry experts who are driving innovation, delivering results and staying ahead in today's competitive landscape. We will break down, um, what it takes to tap into founder mode and turn vision into action. If you are serious about elevating your leadership and execution skills, make sure to subscribe on your favorite podcast platform or YouTube. It helps the podcast and ensures you never miss an episode. Hi everyone. Welcome to another episode of the Priyanka Shinde podcast. Today I have a guest who is an expert in the world of artificial intelligence. Welcome Nitin Agarwal. Nitin is a seasoned leader with over a, uh, decade of experience spearheading transformative initiatives in the field field of artificial intelligence. Nitin is a fervent advocate for harnessing the power of AI to solve real world problems and drive positive change on a global scale. In his current role as the leader of generative AI at Microsoft, Nitin is responsible for transforming marketing using AI and developing large scale platform with cutting edge features. For marketers. In his previous roles, Nitin was overseeing the design and development of large scale AI solutions and products with a particular focus on Gen AI. He also played a pivotal role in developing and executing strategies for Google's enterprise customers. Nitin's deep expertise in the field is further evidenced by his published papers in top academic conferences such as KDD and journals, as well as frequent speaking engagements at industry events like cii. Welcome to the show Nitin. Uh, very excited to have you.

Speaker B: Thank you so much Priyanka for having me over here.

Speaker A: Well I would love to learn a little bit more about you. So tell me and tell the audience a little bit more about your journey and how you came about to be in your current role.

Speaker B: Absolutely Priyanka. So I'm in this AIML domain for quite a long time. Um, my journey started as an accident, uh, for sure I never planned to be in AI. Uh, my first project almost like 12, 13 years ago when I got it, I got it because nobody wanted to work on that project. And eventually it came up that okay, hey AI data science is becoming a big thing. Eventually I moved uh, for my master's and then joined uh, exl, uh services company, um, leading a team of data uh, scientists over there as well building some new solutions for our customers and then eventually joined Google. So I feel that okay, I never planned for my journey end to an end. I just keep uh, working on some really impactful work that I loved working with it. I fell in love with this data science as a domain, started my journey learning R, uh, then eventually moved into Python, then moved into TensorFlow, PyTorch and now we are going through all those prompt engineering things and so on and so forth. So till this point of a time it's very enriching thing, very fast moving, fast paced thing. Never expected that, okay, this is going to be this big. It used to be like a small, sweet, cute, small field where a lot of people are trying to build some good models. But right now it's exciting to see uh, the way we are evolving.

Speaker A: Yeah, that's wonderful. I mean there's a point you made about you started working on something that nobody wanted to and you know, sometimes it's some of those things that nobody wants to work on, nobody wants to solve these things and we think oh you know, maybe it's because it's not something like a shiny object or something and maybe those are the things that down the line become something really big. It may not be the shiny object right now, but it might become something later on. So you really make an important point that it's not necessarily something that really looks like bright and shiny need now.

Speaker B: That's absolutely right. Priyanka Statistics there for a long time. When talking about neural network, it came in before I, before my birth itself. So it's a pretty, pretty old field. Right. But it never made a mark because of definitely some certain limitations. We're talking about computation, we're talking about data collection, we're talking about capabilities at application level and so on and so forth. That point of a time, it's a boring field. Actuaries are there for a long time. Like insurance is using actuaries for a long time. That's a very heavy statistical based uh, problem solving, talking about weather forecasting. So a lot of these kind of fields, lots of these kind of sciences are available. But right now we're talking about, we have a lot of abstractions now you can start working and start consuming the coolness of that technology. So definitely right now there might be a lot of these kind of problems. A lot of these kind of fields are available. That doesn't look cool. That might be evolving into something larger, something bigger, something better moving forward, who knows. But definitely AI is going to be a field but I will put my money on.

Speaker A: Yeah, I mean you have been uh, doing some work in AI one way or the other even before generative AI became a thing in 2022, right?

Speaker B: Yes. Yeah. So we are working on this field for a quite long time. Even if you're talking about generative AI, that itself field is not new language models exist, uh, a lot. Whether we can start asking questions like encoder, decoder, duality, nature, so on and so forth. Adversarial networks are there for a long, long time. I still remember that back in 2018, 2019, we were solving a problem using Gans for images to make it better and so on and so forth. Definitely it's getting better, it's getting easier to consume it. Previously it was a headache and a big pain to use these kind of models. So it's there for a long time. But I'm really impressed with the way things are evolving and people started building up their full time careers and started pivoting so fast.

Speaker A: Yeah. How have you been using or leveraging generative AI both personally and professionally?

Speaker B: I'm using it a lot for sure. It's making me more productive. If you're talking about professionally, um, Microsoft is one of the speedheading companies in AI so we use Copilot as a tech stack without any Doubt it's making us more productive for email writing even to prioritize and sort our emails using it for the documentation summarization. In the marketing field itself, we have so many other use cases we are focusing on to empower, uh, marketer we call it as a modern marketing. Right. So it's already happening. If you're talking about like on a personal basis. I write a lot on LinkedIn, so definitely AI helps me out. English is not my native language, so I think in my native language convert that thing and translate it into English, convert that thing into some sort of a flow. So AI definitely helped me out to give it a good shape so that it would be easy to consume and consolidate some of those ideas. Along with that I use it to write stories for my son. So he loves giving me characters and some theme. So adding that kind of a creativity that helps me out a lot. So definitely feels like that. Okay. AI used to be feeling like you are using a particular tool that's dedicated for this technology. But right now if we talk about most of the tools, tools are coming with that kind of a technology. We're talking about maps, we're talking about spam filtering for the calls on day in, day out basis. I hardly know anybody who is not using AI knowingly. Annoyingly that's a different thing. But almost everybody's using AI in some form.

Speaker A: I love the idea of like creating stories. That's a really fun way to leverage it. And AI can sometimes be very funny. I mean my son does like they roast each other uh, with AI, so I don't know. He gets a kick out of roasting.

Speaker B: Um, I recently delivered like a one like not like a deliver but it's primarily about. I was at my son's school and there's like a career week, uh, where parents uh, go to go to their uh, their kids classes and share what other work they are doing on. So that point of a time it's very hard to tell. These are like five, four, five year old kids, right? Very hard to tell them what an AI is. So I give, try to give it a context. Hey, just think it about like a robo who can do whatever you want to and starting with a very dedicated use case, it can draw whatever you want to. So tell me what you want to see. Somebody wanted to see that. Okay, hey, dragon catching a fish. Somebody sees that. Okay, what about this? Like strawberry in the shape and so on and so forth. It gives them kick that. Okay, wow. Whatever I want, it's available into the market. It's a good, good hedgebee. And the way that I see the reaction from the kids, it's priceless.

Speaker A: Yeah. Ah, it's, it's so interesting that the next generation is going to grow up with something like this where they have a tool that can do almost anything that they can ask for and a lot around, especially with. I feel like kids have such a lot of imagination which we as adults can sometimes lose, which maybe now we might get back because we are now trying to play around with our imagination using ChatGPT or something like that, AI other AI tools. And they can actually maybe explore their imagination with such tools.

Speaker B: Absolutely. Definitely have some, some different opinions around it that might be a little bit uh, different uh, than imagination. So I wrote some piece as well. I got a mixed reaction around it or not. So definitely AI is a good enabler. But we always say that okay, hey, this tool is as good as their master. So whether it will be enabled.

Speaker A: Tell us a little bit more about that. I mean I would. Yeah. I mean there are definitely, you know uh, every technology comes with its pros and cons. So what do you think are ah, I mean there's obviously a lot of advice that we can give anybody young and old to who is using these tools. What are you thinking in terms of how these tools can help or hurt?

Speaker B: No, that's a good question, Priyanka. So when, when, when, when we see these kind of tools with AI itself definitely it started with a very basic technology of next token prediction with some sort of uh, a smart innovative ideas and thoughts. We started giving it a feeling that okay, this algorithm is thinking it's ideally it's again it's a language model auto encoders or a transformers running in the background for the next token prediction. It alludes to that. It gives a feeling that it's innovation. So when talk about it, it's a very common phenomenon. These tools are as good as their masters. They are using it. If you know your job and you know how to use these tools, you will be super productive. You will be able to do 10, 10 things in your lives. But you should and you must know those things. The question is how you learned those things by doing those things by yourself. Right? If I say that okay. Hey Priyanka, can you write a document or can you write a proposal? For this use case you have written the proposal from scratch. Sometimes you might have started with a blank slate, blank document. That's where your mind runs with all the thoughts. Some few people call it as a writer's Block, few people call it as like that. Okay, hey, just uh, limited envisioning or whatever, it can be different terms. But you started with that and now you can say that, okay, you know, you have some draft in your mind, you can ask AI, uh, I want to write this proposal. Can you write these, these kind of pieces for me? Now give the same tool to somebody that has never written a proposal in their life. The way they will be using that tool is going to be very different. Now they will be asking rather than telling the algorithm or telling the tool what to do. And in that particular case, the responses are going to be very different. And if that's going to happen, if I'm going to give this tool to 10 people, there's, there's a very high probability that those 10 people will be coming up with some similar outputs. So is it actually enabling innovation? I doubt it. If you know how to use the tool, where to use this tool, it can surely augment the innovation. It can just generate a 10x results. But if you do not use it, and if you want to use this kind of a tool as an augmentation for that particular skill, innovation is hard to get.

Speaker A: So it's almost like it. If the person who's using it is skilled enough, then it can leverage the tool to maybe get faster at what it's doing.

Speaker B: Absolutely.

Speaker A: Or further their skill.

Speaker B: Yes.

Speaker A: So now we are getting into more like how we get better at prompt engineering here in terms of how we are using the tool.

Speaker B: Prompt engineering is a one part of it, Priyanka. Uh, so definitely biggest question comes in what to prompt. Right. So to give you an example, if I say, and um, this one example actually went very famous in last couple of weeks I was reading uh, through LinkedIn as well. Somebody posted a picture that these algorithms cannot create a full glass of wine. Okay. When you say, hey, can you create a picture of glass of wine? It creates a generic glass of wine that's half full. Then says that it can make it a full glass. It creates the same picture. Then it says that okay, remove the air, make it full. It never fills up the glass with wine. It always consider that half full glass as a full glass of wine. Right. Prompts might change. Now if you will be a better prompt engineer, you, you can use that, okay, full the wine till the brim of this glass. It may work. You might change the model. It may work. Or there are multiple other mechanisms. So prompt is just a one way of doing this. My view is your critical thinking and how you're Approaching a particular problem solving, that's going to be the differentiating factor. Prompt will come in the way that we were prompting last year. It's very different than the way that we are prompting it today. Agents are coming so they can do a lot of your prompting itself. Few people started using LLMs to do the prompting for them. So they tell them what to do and then ask a question. Can you convert it into a detailed prompt, ask me a follow up question and then send that output to another algorithm that's already happened. So prompt engineering, we offer it, but asking the right questions, what are you doing? Why are you doing, what do you want? And how you can solve a problem. Those questions are going to be more important, right? Yeah.

Speaker A: And where do we go from here? Like what happens if, when we think about innovation, because when we think about we are training this data, uh, the algorithms, we are training the models. And a lot of these models have been trained on all of the data in the world. And if people are not coming up with new things, say for example, then there is no new data being fed. All of the data that is now being produced might be produced by models or ChatGPT or something equivalent. What happens with innovation? Like what, what are, uh, in your opinion?

Speaker B: That's a good question, Priyanka. So I feel like that when we are generating um, a lot of content using this technology, there is always a loss factor. So for an example, if you're writing something on your own and then you feed this thing to some algorithm and then keep asking question, can you write some more pieces that it will get distilled. The information, knowledge, the emotions, the sentiments, the relatability of that content starts going down in the next wave. You will see that this new data will be acting as an input to the model. So I will say that, okay, actually those models will start going bad. We saw some of those uh, examples in the past. For certain of those models, anthropic Claudius model face the same feat. Um, Sam mentioned about, uh, ChatGPT models. The latest versions are not performing as good as that of the older ones. Typically didn't happen. Ideally by an organic means, the next version should be superior than the before. But these kind of NA algorithms are as good as the data. Like garbage in, garbage out. Right. If you're talking about the innovation spring, just like I mentioned, innovation cannot happen without humans into the loop. Definitely where we are evolving, it has to be human plus AI augmentation. It is going to be that kind of an interface where humans will be utilizing, generating some New innovation and then augmenting AI to make that thing happen faster. Using execution agent rather than a thinking agent. Definitely at certain places, a lot of thinking can be augmented, can be, uh, implemented using these kind of algorithms. But right now, if you talk about the innovation, it cannot stop. And that's why a lot of people started going back to their, to their old school thing, hey, I want to write something by myself, I want to build something by myself. And then I write a lot of that content with my hashtag written by human. Somebody tagged me. A lot of people started using it, that, okay, I'm writing it. So that might be there. Innovation will not be stopped. But it might get sparse because there's such a burst of AI generated content coming up. So determining that, okay, what humans are writing and what AI is writing is going to be very tough.

Speaker A: That's so true. And it kind of gives you that hope as well, that because there's sometimes can be fear that, oh, what happens now? All of these jobs are going on. And I did want to bring up this whole question because there's a lot of news coming up lately with all of these CEOs talking about AI first. Um, so I'll come back to in a second. But you know, there's this fear of like, AI is going to take over all of these jobs. And at the same time now we're also talking about like innovation does require humans, it does require people to actually think because there is this, that imagination or like something that you think it's hard to replace with just AI. And so when you hear all this news, with all the, a lot of these CEOs we have heard Shopify, now dual lingo, think Salesforce box, just as of like the last couple of weeks, uh, about hiring or like thinking about AI first before hiring. What's your take on it?

Speaker B: It's an interesting point, Franka. Um, I was reading a book, Nexus, interesting book from, uh, Evil Harari. He mentioned about how the information flow changed. So for an example, when printing press came into the picture, people started saying, is that, okay, hey, now anybody can share their thoughts out, but who will generate the thoughts? Then after that newspaper came in and the information flow changes, Internet came in. But one thing never changed. That was like humans were generating the content and we were using the technology as a medium to relay that information. Right. When search came in, like whether it's a Yahoo search, Google search, we can, we can definitely debate who came first. The point is when the search came in, it provided us a technical tool to Parse the information, but it never generated the information. Now we are going into a phase where we are generating the new information, right? It's going to change, it's going to disrupt and it's going to evolve the businesses and the business model. Without any doubt, till this point of our time, I have hardly seen any technology that displays the jobs. It evolved the jobs industrial revolution. We say that, okay, the jobs changed, but it generated a lot number of new jobs. How the things will happen in future, how the roles will be evolving, we are still seeing that kind of a pragmatic version over a period of a time. There are some, some speculations. I, uh, will not go into that one. Everybody has their own forecasting, predictions and opinions how the future will look like. I think it will be better to go step by step. Like do not think about five and ten years down the line. If you're thinking about, hey, what will be that job while planning for the career, think about the learnings, thinking about evolving. Think about going back to the first principles, critical thinking, problem solving, domain expertise, and then using these technologies to enable the actual impact you can bring on the table. So I feel that's going to be uh, the future. It's very, very hard to predict that, okay, what will happen? What kind of a jobs will exist or not. That's going to be a tough part. I keep getting those questions from a lot of students that, okay, hey Nathan, like, shall we learn coding? Yeah, you should, like, do you feel that coding is going away? No, it's not going away. The way that you're writing code is going to change for sure. Stack Overflow changed it. Now it's going to be GitHub, copilot and cursor and all those tools, they are going to, in future things will change. But if you say that, okay, whether this field will not exist in future, that's going to be hard.

Speaker A: Yeah, I'm sure you definitely get a lot of questions, uh, because I think you also teach uh, uh, around the topic of AI as well. And so there are a lot of people probably also coming, you know, uh, not just uh, parents and students, but also asking about like, what do we do with respect to our jobs and is this, you know, is our role going to be secure or not? And so what do you tell these people?

Speaker B: That's an interesting point, Priyanka. I personally feel security is emotional. So if you talk about that, okay, if my role will be existing in five years down the line, I don't know, like the way that we are working I'm trying to automate my role as well. That's going to be the future for all of us. Right? So if I think about I'm this AIML domain for more than a decade. If you think about the way that we used to build models in the past, the kind of problems that we were solving in the past, is it the same? It's not at all the same but the problem remains the same. That's an interesting point. We used to do that kind of customer segmentation, customer profiling. We're talking about forecasting, we're talking about lifetime value predictions, we're talking about like a uh, better patient care and giver, uh, so a lot of those problems, that problem remains the same. The face of the problem has changed and the way that we are solving is, has changed for sure. So you need to keep evolving with that one. A lot of roles will be disrupted for sure. If we talk about for example program management or a technical program management, it will be evolving A lot of the backend things, a lot of the very basic, for example knowledge consolidation and everything will go away. But the human touch to handle a lot of escalations, firefighting, that will remain as is talking about software engineers, very basic software engineering jobs where we are, where you are just taking a boilerplate code, that role will change. But how to debug that code and how to make it ah, useful for a larger application that will remain as is. So it will be very, very hard to say what roles will not exist. But definitely some roles that are very rule based, very uh, grunt work. We're primarily talking about knowledge extractions, retrieval, summarization that might be evolving faster than the other of them.

Speaker A: What's the best way to uh, prepare oneself? I mean be it um, from an individual perspective or from an organizational perspective over the next like five, 10 years.

Speaker B: That's a good question Priyanka. I'm not sure about like the. For the next five, 10 years. For the next couple of years the way things are evolving, it's like six months or one year or two years is a feels uh, like a long time. As an individual I will strongly recommend uh three things. First thing is about keep evolving and keep learning. Don't get bogged down into all the launches. Go back to the first principle. Learn about this technology and connect it with your job that how can you do it better? How can you evolve your thought process? Second thing is don't just focus on doing a ah, one part of a task, think about a bigger picture, be a part of that kind of a larger discussions that hey, uh, why are you solving this problem? Why you should be solving this particular problem? What's the larger impact on the organization? How did you prioritize it? And then thinking about a larger ecosystem that what are the pieces that are connected to each other? What are the different teams uh, working around on these ones that's going to be really important to think it through. And the third thing is about collaborate. Build your network without right now, like I got this uh, advice few years down the line. A network is your network. It's a very common theme. I uh, will not call it as a net worth but very strong network of even 5, 10, 20 people to whom you can rely on. That's really important. It's very hard to build a network of thousands of hundred thousands of people. I will not say your LinkedIn connections is your network, it's just your audience, but some really good people to whom you can, you can rely on getting the right mentors who can guide you. That's going to be really important. So that's on the individual perspective. If you talk about uh, the organizational perspective, it's a larger discussion. So definitely again three things. One is what will be the strategy and what will be the vision for an organization? How are you evolving? AI is just a way of doing things, but the problem definition and a formulation is really important. The second thing is about the implementation strategy. Whether you're going for a top down versus bottoms up, how are you making these things happen? None of the approach is foolproof. Every approach comes up with these pros and cons. But what suits the best for your uh, maturity, where you are, how you're evolving, what are the problems you're solving, at what speed you want to move on. So there's a, there's a number of dimensions that you need to focus on. And the third thing is about focus more on execution. So definitely we can see a lot of strategies evolving it through. But execution is very hard in AI. Even when you say like building an agent is easy, but taking that agent into production, make it available to the business and make it in such a way that it can generate a business impact. It's not that easy. It takes time, not just at the technology level, but at the uh, processes level. Change management, adoption, a lot of things like a mindset shift is a big thing. So those are my, what I should say three quick points. On the organization side, can you uh,

Speaker A: elaborate a little bit more on the mindset shift? Especially when you talk about the challenges of um, AI execution, uh, for organizations.

Speaker B: Absolutely. Priyanka. So we're talking about um, the mindset shift. It's primarily about why should I adopt this AI? There's a very interesting research happened recently. I forgot the name. Whether it's a McKinsey or Deloitte or somebody like that, they shared a very interesting insight and they say that a lot of employees, like 40 or 50% of the employees are using AI, but they are not telling this thing to their bosses or their companies that I'm using AI, ah, for problem solving and so on and so forth. So that acceptance is low. A lot of the times these systems are heavily dependent on humans to be successful. So it's almost like they're okay, for example, if I want to replace you, you need to train these models with all the inherent knowledge that you get, support these models and so on and so forth. The question comes in, why would I do that? With my full energy and potential as a leader, definitely I would want to, but as an employee, why would I do that if it comes to that particular level? Right. Just like you mentioned, there are some discomfort, uh, or potential fear, um, among people that okay, it will take away my job. That also uh, impacts the adoption. So there are multiple dimensions around when talking about change management, about the uh, mindset, uh, shift along with that. When you're talking about the AI, it's not the core job for the teams, right? It's a parallel job. So people are learning it, they are reading it. But making and using AI for their day to day activities takes time and there are risks for the failures as well. So the question comes in if the organizations are supportive enough to take those risks for failures. So it cannot run as a parallel where you say that, okay, hey, you can keep doing your job as is what's happening in today, but also get trained on AI as well and then start adopting and utilizing it. So it will slows down the adoption. That okay, how to prioritize what task you have a limited hours in a day, right? How to learn that additional tool and then keep running the business as is, and then keep running another parallel train that you feel like it's going for a transformation. So there are a lot of those kind of shifts will be coming through. Either you're talking about a cultural shift, talking about human mindset shift, you're talking about like the process shift, or you're talking about just the dynamic shift. Who will own these systems, right? Who will be accountable for the decision made by those systems? So a lot of those uh, questions are not getting answered right now. We are only at the technical level of the problems.

Speaker A: Yeah, I think that is such a important point to make. And I want to reiterate that because it requires normalization and based on what you're saying, there isn't a complete normalization that AI can be used by people to augment their jobs without the fear of like, if I tell somebody that I'm using AI, ah, they're going to think I'm m not capable enough of doing my job by myself, which is not necessarily the case. And it kind of for me, because I'm also a coach. It kind of draws parallels to it because a lot of times as a coach a lot of people need like even the best sports people have coaches and they are open about it. But in the industry, like uh, especially in tech and professional world, not a lot of people talk about it because I think there's a little bit of stigma associated with it. That that means I'm not good, but is not necessarily that even the highest performers have coaches or like executives have coaches and just because you want to amplify your high performance even to the next level. So I think that, yeah, I don't think this kind of, um, point has been discussed before, at least, I mean, not so much. Um, and so I just wanted to reiterate that. So thank you for sharing that because. And it goes back to something I think that was mentioned in that Shopify memo as well, around the AI thing that if you like, we're going to talk about use AI, use it every day, make it part of your job and things like that. And I think that was that normalization of do it. It's not just about, oh, we are going to not hire anybody because we want to use AI or have AI agents. It was more about we want to normalize this usage and think about how we can have a multiplier effect of the people that we already have. Which probably makes sense if you think about it that way and not necessarily the way about like, oh, we just going to replace everybody with an AI agent or something like that.

Speaker B: That's absolutely right, Frank. I'm not sure about the intent of um, that Shopify memo, but definitely one thing is consistent. Most of the tech CEOs we're talking about Mark Zuckerberg talking about Satya Nadella talking about Sundar Pichai, almost everybody is saying that, okay, this is the year or the next few years are uh, the years of optimization. Right. So definitely one intent can be that, okay, hey, how can we use this technology to do some of the tasks without getting some additional help, like to be more productive, to be optimizing it. Definitely we are lacking the causality. That's why a lot of these reports or uh, uh, researches that are coming up that hey, how people are comfortable accepting that if this is um, the right way of using AI or not. One interesting report published by Adobe as well. They say that okay, when they are using and publishing an image with a tag like built by AI or created by AI, they are not getting that much traction on versus the images that are not being tagged. So the question is yes, we want to use this technology, but are we ready to accept that okay, this has been done by AI or not, will it be having a very similar kind of a mindset? Right. So if I ask you that, okay, hey Priyanka, like this is your speech for, for a keynote tomorrow. You have two options. First thing is it's written by Nitin. Second one is it's written by AI. Which one will you, will you be more comfortable with to accept? Right. So even you don't even know me, you don't even know if I'm a good writer or not. Right? So uh, but in that case at least you know that if some, something goes wrong, you have uh, somebody to talk to, somebody to blame, somebody to actually put accountable around it. I think strong word but if it's AI, whom to blame, you to blame.

Speaker A: Yeah, I mean, yeah. And then, then I guess the, the follow up to that question is like so if you're giving a speech, is it you who's giving it, who has written it and who's giving it? Right. And, and then I think uh, the, the word that comes to mind when you say that is like the trust factor. And I was, I was hearing about this as well is like with AI coming in, there's this aspect of trust now that's coming in as well. And so I think putting that label like generated by AI or written by AI or something, does it increase the trust or does it actually decrease the trust? Because if you also think about it, there are so many nowadays also videos being generated by AI and you see this a lot in like the short form content videos. And sometimes it's hard to nowadays uh, trust a video because you don't know if it has been generated by AI or not, uh, or if it's fake, uh, or not. And I think that's where uh, where it's also important from us as people who have been in tech or people who develop tech to figure out that how do we make sure that the trust factor is not eroded? Because that's where as technology companies we have to make sure that when we develop tech it is also for the good and that there's not a lot of mistrust or misuse of that new technology.

Speaker B: Absolutely. So that's going to be a very interesting problem to see that. Okay, hey, if till the time you reach to a point where you say, hey, it's being created by or generated by AI and you trust that thing more, you need to feel around it. So if you are going by buying some cloth or buying some accessory, for example, say a bag, if it says that, okay, built by a machine or built by a human, which one you trust more? If I say that this bag is built by a human versus machine, which one? You'll say that, okay, this is going to be sturdy. Forget about the trend factor, forget about the designing, forget about the price. But if I say you need to go and travel with this suitcase, build by machine, built by human, which one you trust more? That's a game. Right now a lot of people say I will trust like machine because it's a set standard processes. I trust it. It's going to be strong, tough, in detail than humans, human generated things. It will be primarily about short, small things, hand painted things in a very limited capacities. Not primarily for the sturdiness, but primarily for their trendiness and that touch and that emotions and the sentiments, right? But if you convert that thing into written by AI versus written by humans, which one will you trust more? You don't have that trust with written by AI right now versus written by human. So till the time we, we pass that chasm, it's going to be very, very hard for AI to transform.

Speaker A: It's so interesting. Um, that also brings me to this question. I was like there is this trust factor and then at the same time a lot of these companies are also trying to scale, right? You probably have seen that. We talked about the execution factor and trying to just process and things like that. And there's also challenges in scaling, AI, uh, implementation and innovation. What have you seen, uh, in terms of those challenges?

Speaker B: Scale is always a challenge irrespective of the technology. So it was always a challenge. Even if we're talking about websites, even a very basic scaling of our kubernetes clusters and making all those apps available. The question is, what are we calling it as a scale? We saw that okay here a lot of even famous uh, platforms, they got crashed during some of the high Fidelity, high frequency matches, right. With number of people who started watching it at a simultaneous um, like in a concurrent concept. We're talking about the AI the same thing right now we are not at a place definitely these LLMs gives a chance and gives a flavor that it can do everything. But it's not at a place where a single model can be consumed for multiple enterprise grade problems. From one problem to another problem, their behavior changes, the requirement changes, the data changes, the formatting changes, foundations changes. You might need to tune it. So it becomes really hard to consume the same model and the same process for a range of different problems or even similar problems across the teams. And that makes scaling very very tough. That's why where you will see the scaling where the use case is not very much specific summarization, writing, open ended concept, right? Parsing and generating something. But can you use the same model, can you use the same interface for your medical documents, for your logistical documents, for your financial documents? Maybe not. That's why we say that keep adding all those tokens, keep making those models larger and larger and larger. The idea is that okay, can we provide all those information and details into the same model or not? Whether this is going to succeed or not we are going to see in near future. We had a mixed uh, bag of reactions, sometimes really amazing successes. Recently one uh, use case came in where ChatGPT just looked at the mole on the hand and say that okay, it can be a skin cancer and become really wild because uh, that was a good diagnostic. So those kind of things that we have seen in the past with the dedicated healthcare based models but now we are seeing that things are actually evolving. So if we're coming back to it on the scaling of these kind of problems that will always be there. And that's exactly these organizations are going through. If you're spending some good amount of money and time to solve a problem, how scalable is that? Can it scale across the organization, can it scale across the teams, can it even scale for hundreds of people? If it's not, is it worth solving that particular problem or not? MIT recently uh published like um some, some concepts around it how to define what should be your evaluation metric to prioritize your use case. So McKinsey published their agentic structure. So where they started providing some sort of a uh concepts that how to evaluate, evaluate a particular problem if it's worth solving or not. So we'll seeing a lot of frameworks coming up in near future but right now you're absolutely right. It's hard to scale up uh, these kind of AI solutions.

Speaker A: Yeah, I mean there have been so many startups that have just come up right because of this whole AI ah, revolution. And I'm sure a lot of startups are finding it difficult based on some of these challenges that you have mentioned, from everything from execution to process to the scaling. And so what according to you can startups do figure out uh, how they can sustainably make it work for them where they can actually create something that can be a product or find product market fit?

Speaker B: That's an interesting uh, question, uh, so I may not have a direct answer uh, for that thing. It's very, very hard to determine that if this is the, this is the right problem for AI or not. Whether a part of a problem that can be solved with AI or not, that's a different question. But getting that product market fit, you need to try and test it out if this technology is working at what level. That's why it's becoming really, really important to define what problem you will be solving. And even more importantly, why are you solving that particular problem? Is it a blocker, is it requiring a process changes? Is it because we are spending more of times because no technology is available? Why is becoming a really important. If Y is really strong enough then you say that okay, hey, right now there's hundreds and thousands of tools, startups available even for a one problem. What's your mode, what's your differentiating factor that's not copyable and so on. So there is like a multiple multi step framework when you're solving a particular problem to evaluate it. If this is the right problem to solve, are we moving in the right direction with the right product or not? How to evaluate what kind of a business processes must be attached to it, getting some initial customers to bring you a success and so on and so forth and going for that product market fit. That's the kind of a journey that I have seen with a startup. Priyanka. I'm not coming up with a direct startup experience so.

Speaker A: But that's what I. Yeah, yeah, more from a perspective like AI because they're just using. Because a lot of them are just saying okay, we'll just use AI to solve this problem, right? So whatever their problem they've say figured out this is the problem. But a lot of them are just saying okay, we're just going to use AI to solve this problem. Um, but like you said, it's not necessarily always easy to use AI to solve a problem. Um, and because also AI is evolving so fast A lot of startups may find that it's hard to scale or it's hard to actually implement it in reality or adopt it to the needs of the problem or maybe the problem is actually solvable in a different way. Um, so I just wanted to see if there was you know, if you had any take on that specific from an AI perspective.

Speaker B: I strongly believe Priyanka, in that case, uh, AI is an enabler to solve a problem. It cannot be your front end factor. So if I'm saying is that this is the right problem to solve, the question comes in again why you're solving that problem and then going with when you got an answer for why, how can we solve this problem? Easily, responsibly, sustainably. Right. So recently Jensen Huang also mentioned about Nvidia, uh CEO and co founder that go deep into a domain because getting that kind of a dedicated data for a one particular domain for AI to use, it's very, very tough. That can build a different mood, that can actually build your uh, point of differentiation. That's going to be the key important factor I will feel definitely right now. Startup ecosystem for AI is very hot. That technology is evolving very very fast. We have seen a lot of those changes of startups actually got uh, consolidated. Few actually really showed up very well for the use cases. One of the biggest challenges, ah, a lot of um, larger uh organizations are facing, it's about moving fast on the execution. And when these are these startups comes up with their use case they do not have any baggage of the processes. They can take risks and they can move fast and they can build something native from scratch. Um, for this technology just consider like that okay, you're building an EV electric vehicle from scratch versus uh, along with your IC engines or standard cars.

Speaker A: Right?

Speaker B: We have seen that thing in the past. We have seen uh Tesla coming up with that kind of a mindset because they came up with a very fresh concept, not with a baggage. And that's exactly the kind of themes that we started seeing it in AI space. Perplexity came in glean came in a lot of other uh, players that actually uh, just change uh the market that we are approaching a particular problem. So I think that's exactly what's going to happen. AI is going to be a big factor for it. But AI will not be uh, the problem statement. Problem statement must be attached to a business need. AI will be a way of solving the problem.

Speaker A: Yeah, that's very well said and you're absolutely right. I think startups have the speed advantage for sure. And you said something else which I wanted to reiterate is they still need to find their moat. I think they can only leverage that speed if they can find their moat. Because otherwise without that even if it's AI or whatever it is, it's not going to be able to sustain. And eventually some of these big companies uh, because they have some of like they have the capital I guess a lot and they have some of the largest language models, it will get consolidated with everything that's happening in AI and I'm sure we can go on and on. What are you most looking forward to from here onwards?

Speaker B: I'm really loving this journey Priyanka, without any doubt. It's sometimes overwhelming. Uh, I will accept it for sure. But things are going pretty well so I'm really excited to learn more. Definitely it put a lot of onus on us. Whatever we have learned, even the basics that we have learned in AI, a lot of wrappers, a lot of abstractions, the tools they are changing. The base technology might not be evolving that a lot but a lot of innovative concepts keep uh, coming it up. So I'm really looking forward to where this technology is evolving, what kind of a new problems that we can solve. It very, very excited for agents that how agents will be evolving. Definitely we started seeing some initial successes but I still feel a lot of uh, potential has yet to be tapped in. I feel os, uh, operating systems will start getting their agent flavors as well. Most of these softwares, most of the ecosystem has been generated with human experience or human ah mindset but right now the users are changing. Human may not be be the user computer or uh, agents might be the user. So definitely we have an old age concept of APIs but this time it's going to be way going way beyond APIs. So let's see how things that will be evolving. I will be very excited to see that kind of an integrations and see that uh, how our lives are going to be even better in future.

Speaker A: Definitely looking forward to it. Thank you so much for sharing your thoughts.

Speaker B: Pleasure. Thank you so much for inviting me. Thank you so much. I really enjoyed this discussion.

Speaker A: Same here. Thank you. Bye. If today's conversation sparked new ideas or challenged your perspective, I'd love to hear from you. I work with tech leaders and founders to tackle complex challenges whether it's scaling teams, making high stakes decisions or turning ambitious ideas into reality. If that's sounds like something you are navigating, let's connect. Reach out via my website thepriyankashinde.com or find me on LinkedIn.