{ "episodeId": "ep4", "episodeTitle": "AI and the Future of Software Engineering: Hype, Reality, and What Actually Changes", "speakers": [ "Vikas", "Srikanth" ], "speakerMap": { "speaker-0": "Vikas", "speaker-1": "Srikanth", "Speaker 1": "Vikas", "Speaker 2": "Srikanth" }, "duration": 2897, "durationFormatted": "48:17", "wordCount": 9181, "segmentCount": 75, "segments": [ { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 52, "timestampFormatted": "0:52", "text": "Hi, welcome to a new episode of the AI Valley podcast. My name is Vikas and today we are going to talk about software development and AI. This is something that clearly has been the center of debate in many circles. Given it's having a lot of impact in the way software is written, developed, one of the places where it's disrupting more than other places. So we thought we should have an objective discussion on the topic. And today we have a very special guest, Srikant, who works in a big tech in US. He's been coding for very many years. I've known him as one of the best software developers over the years. So I thought, who better than somebody who writes code for living to get him on the pod and have a sort of a open-ended discussion on how he see as a software writer or software developer, this field evolve and how it's impacting. Welcome to the podcast, Srikant." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 110, "timestampFormatted": "1:50", "text": "Thanks because good to be here." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 112, "timestampFormatted": "1:52", "text": "Subab, so let's get into the center of the discussion. How long have you been coding and do you think the software development as a field is changing, especially in last couple of years with the advent of AI? Do you think there is a way it has been impacted?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 129, "timestampFormatted": "2:09", "text": "Yeah, so I have been working for 18 and a half years in this as a developer architect. And yeah, last two years also have been pretty interesting in the way things are progressing and the speed of change at the lack of visibility as to where it's going. I think it's been very, yeah, the change is constantly there. We are right in the middle of it. So it's hard to know exactly where we're going to end up. but it's still certainly very interesting." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 154, "timestampFormatted": "2:34", "text": "Okay, so now I clearly remember I've met you around the same time for last couple of years and I've asked you same question and your answer has been definitely not the same every year because every year when we meet after there is a lot goes into the model capability, there are new models, new tool around it, but somebody who writes code Do you think there is a lot of fear in lot of software developers in the way that this AI may be coming for their jobs and there are people with mixed opinions. I have known people who stand on either side of the debate. Now somebody who write code, do you think has any AI helped you personally write better code? Or do you think AI can write better code than you?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 206, "timestampFormatted": "3:26", "text": "So the last part is rather straightforward. To the extent I have used AI, it doesn't write better code than me. In fact, in any object you measure, it's not good. It's a lot of repetitive functions and a lot of copy pasted stuff. So in terms of quality of code, don't think it's particularly great right now. Where it has been useful for me is in completely greenfield areas. Like if there's a Something I want to simulate, like the other day I I wanted to have a solution to a particular problem. Basically how, like if you have a hundred thousand different items and there is a 80-20 rule, the Pareto principle is being used in an item being selected. So there's basically bias towards some set of items which are selected often. This is generally true for the e-commerce industry, right? There's a bunch of items which are ordered a lot more than other items, right? If let's say you have a corpus of items, 10,000, 100,000, some pool of items and of which there is a clear 80-20 pattern and recursively 80-20 pattern. the 20 % which is sold 80%, within that also there's a 20 % which is sold 80%. So there's generally that kind of pattern. Though the exact 20-80 ratio is debatable, but the pattern is pretty much there. So I wanted to see if I have a pool of 10,000 or 100,000 items and I have a thousand requests for... one item each and then the items follow this kind of a pattern like Pareto principle. Then how many different items or how many unique items will be there among a thousand requests? So if it's completely random, close to a thousand unique items, right? But it won't be completely random. So I had this thought and yeah, I asked AI, okay, how do I do this? And then what it said, I double checked. It turned out that there is no direct way to calculate it. You can simulate it. And then it took maybe another five minutes for it to write a simulation code. And it so happens that in that iteration, the code was accurate, like it ran correctly. And so in that sense, it's really great in the sense that, there's an idea. What would have taken me a lot more time to go? It's because primarily it's, I'm doing something else and this is a thought that comes in between. And it's a kind of distraction from the main goal, right? And the earlier version before AI, I would have had to take the detour. which is a kind of distraction from the main purpose and so faster iterations in those are are certainly welcome." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 337, "timestampFormatted": "5:37", "text": "I see. Yeah. So I see what you're saying. You know, I agree. It's clearly easy to get into these rabbit holes with AI and quickly experiment things around bounce ideas. And it's a great partner to write mundane stuff, something that would have taken a lot more time for somebody. Yeah, I think that makes sense. one, because obviously, for the first 10 years of my career, I was writing code every day. Probably, you know, I kept doing that. But these days I don't write code for living these days. But one of the things that I have seen, in fact, the AI based coding or AI coding has certainly impressed people who did not code, right? Think of your product managers, think non-technical co-founders or people who sort of for whatever reason, good or bad, were not coders or didn't code for living, right? I've seen those people amazed at this capability or newfound capability where they could quickly, you know, build website or, know, for example, there's a web UX designer who's great with designs, but they don't know the intricacies of building a web front end. Now they have a newfound sort of skill. that they can complement their designing skills. those people have certainly been on the beneficial side of it. But I understand people who are, say, what I call software purists, they are getting very impressed. It's akin to a chef who's spent all his life or his or her life, you know, creating dishes, crafting dishes. And suddenly there is this robot or machine that spins out, spits out. these dishes. It's obviously not going to or probably not going to be the same kind of quality as that of a chef who has done this for life. So I understand there is a similar tension that exists here. But" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 452, "timestampFormatted": "7:32", "text": "But let's see." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 453, "timestampFormatted": "7:33", "text": "If you are able to understand the repetitive stuff or the lack of perfect code in the language that you've coded for many years, for example, in your case, could be Java or Kotlin or something like that. But let's say, if I told you that if you have to write something in Rust or say C++ or Go or even some mundane language, don't you think having an AI which is not great but does maybe good enough would help. Right. So what do you think about that? Right. Because it's certainly giving you breadth if not depth. So and so don't you think that's generally a good stuff in the" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 492, "timestampFormatted": "8:12", "text": "So yeah, you spoke about a lot of different themes here. So people who are not professional coders, yes, they are very happy. And coders are not that happy. And there is a reason for that. As you said, like chefs, they know the art well. And so that also enables them to see the problems in the implementation done by AI, for example, and done by others also. because you have been doing it, see someone who hasn't coded, someone product management or UX designer. who doesn't have exposure to code, yes, they will feel like they have new wings because now they can take the idea and then actually see something that is there, right? But there is such a thing as software quality. So you can build something, you can quickly go through some iterations and get something which looks like it should work. And so when they look at it, they have now uncovered a new skill that they wouldn't have been able to do earlier, right? So that makes sense. In my experience, to the extent I've used AI, it's very easy to see that it doesn't work well also. So I gave the example of Pareto principle. So there was a one, two file Python thing, and it works well in those scenarios. I've written, and going to the example, you took a nice example of Rust and something which have a language or ecosystem that I've not worked in. I can use AI to build stuff. Yes, you are absolutely right. And I have done that, and I have seen both sides of it where it does something which it would have taken me a long time to do, and it does something obviously wrong as well. one example is that I wanted to write an Android application. I have zero experience with Android programming. In fact, I have zero experience in any sort of UI programming. So to get started, I told it this what I want. I give detailed instructions on what I wanted, exactly how I envisioned it. And it built something like that. The positives is that, this entire Android manifest file, the resources, the icons, I didn't have to think about any of it, where to place them, how to bootstrap an Android project, none of them. It just went ahead and did all of that well. And I know it's well because it actually worked. But let's take that as a proof that it worked correctly, right? But the business logic that I wanted, I could clearly see there are bugs and nuances. in Kotlin, there's something called coroutines, which is, I asked it to use that, but it did it in a blocking way, which is defies the whole point of coroutines. So what is the outcome? So you can make an app, it runs, but frequently it just hangs your UI. So now," }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 627, "timestampFormatted": "10:27", "text": "I see." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 635, "timestampFormatted": "10:35", "text": "as a UX guy, the fact that I can have an app out there without the need for a development team is great. At the same time, someone has to go figure this out. Right? Now you've gotten out some code. That's where the word slop is a perfect fit. So it's there. And, but someone has to now go through all that code and figure out what are the bugs in it. So it's, it's very nuanced. It is useful, but it's not a silver bullet at the same time." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 654, "timestampFormatted": "10:54", "text": "Yeah, yeah, yeah. I understand that makes sense and honestly I really see what you're saying and I've seen sometime some pieces of code impresses me in a way 0 to 1 is amazing but sometimes the obvious mistakes or the loops that it goes one of my biggest pet peeves is sometimes you're trying to solve even if you're told that you want to solve a certain thing in a certain way and if it if it is stuck somewhere, if you don't pay attention, it will take an alternate route, something that you would not have asked or clearly asked not to go." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 691, "timestampFormatted": "11:31", "text": "Oh, so I had so this is a frequent thing. I think a lot of people have noticed it in lot of podcasts as well. A lot of people point out that when you say, make this test work, it sometimes goes, deletes the test. So exactly. So it can be wow on both sides. If you if you approach it, saying, thinking that, dude, this is extremely useless. It's a negative utility and you go use it. It can wow you. Wow. It can like the same so much time." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 702, "timestampFormatted": "11:42", "text": "Wow." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 716, "timestampFormatted": "11:56", "text": "And now you are happy and you think, wow, it's going to be so useful. And then the next thing you ask, now I have frequently had this issue where it goes into a loop. It doesn't figure out what the problem is. It goes into a loop. And then I ask it to write a test case. It writes a test case and just comments everything out. But then if I go uncommitted and say, now make sure this test runs, it'll run a few times and then it says, let's fix it. And then it just comments out the whole test case. So it's wow on both sides. That's exactly right. It is wow on both ways." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 741, "timestampFormatted": "12:21", "text": "Talking about wow, let me ask you something. One of the reason why, you know, I studied computer science engineering and when I first time I built something worthwhile, wrote, you this was, I think it was in turbo sea or boardland sea. And I first time in my life when I built something worthwhile and when it worked, there was a... I don't know how to describe that. There was a certain kind of euphoria or a dopamine hit when something was, which was obviously not working, but I got it working. And it was one of the main driver or the kick that pulled me in the world of software engineering, because the joy of making something, building something gave you certain kind of happiness, right? Now, the question that I have for you is, I obviously don't get those that sort of a same sort of a kick anymore because I'm not writing code for living. if do you think that one gets a similar kind of dopamine hit when they build something with AI or debug something with AI? I don't, you know, what's your thought?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 807, "timestampFormatted": "13:27", "text": "In the current state of AI, as I said, it's a wow on both sides, right? So, in a way, you get a dopamine hit when AI gets it right." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 813, "timestampFormatted": "13:33", "text": "I see. But is it the same feeling of you working through the night and fixing a hairy bug? It's the same feeling?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 822, "timestampFormatted": "13:42", "text": "I think it's a very subjective thing. I know people who really like to debug stuff. I personally don't enjoy that. Literally going by your example. But the general point is that you're still building stuff. You're still progressing in the direction you want. There is a reason you're coding. So you have something in your mind that you want to make happen, make concrete in real life through software. so you're still getting that done. you if you use AI to solve some particular sub problem, at least as it stands today, you have to verify it anyway. So I don't think that that that that aspect of having created something has gone. And there is it has certainly added this new fun part, Ropamine hit where things actually work with AI. Especially when you that Android project, for example, the fact that it even loaded even though it had bugs and all that, the fact that it helped me remove Especially things that were not of interest to me. Understanding where Android manifest has to be and how icons have to be, where they have to be put and how they have to be referenced. Not having to understand all that nitty gritty. I don't think there was too much value of that, you know, from the get go anyway. What I wanted is the outcome and I achieved that outcome even sooner. In some ways it's even more addictive. It's more of dopamine." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 890, "timestampFormatted": "14:50", "text": "Okay, it's more instant. Yeah." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 892, "timestampFormatted": "14:52", "text": "The lesser effort more guaranteed, but not more guaranteed. But when you use it properly, it certainly reduces the effort that's needed in that direction." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 901, "timestampFormatted": "15:01", "text": "Yeah, the dropout rates reduces for sure, right? Because, you know, it's good to be to work through the night and get it working. But I'm sure a lot of people give up." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 911, "timestampFormatted": "15:11", "text": "working through the night if 80 % or let's say 40 % of it is just trivial really trivial boring stuff and that goes away it's it's I think it's for the win" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 916, "timestampFormatted": "15:16", "text": "Yeah. Okay, so shifting gears, Shikant, I think now let's talk about some nuanced stuff, Something that, you know, a lot of has been on lot of people's mind, I'm certainly sure people who are listening to this podcast will must be thinking about it is around how should a software engineer of today, right? There a lot of people who might be studying to be a software engineer. Now as we speak, right? How should they develop or think about all of these things, right? For example, say you worked in, say, maybe retail domain or e-commerce or any specific domain, right? Now you've learned through multiple iteration, gone through, the hard way and became a domain expert. Now imagine if somebody, so that you are domain expert enough to point out the problems. in the way of AI implementation. But let's say somebody who's just coming out of college, they'll never probably get an opportunity to go at the depth that you had to or somebody else had to who was doing it the hard way. they ever get to a point or will software developer lose the skill to differentiate between good and bad?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 987, "timestampFormatted": "16:27", "text": "It really depends on whether that skill has a value in the future. So, okay, if I don't go into the weeds of it, I certainly lose the ability to go into the weeds of it or understand the language at depth. Yes, that's a rather obvious yes. The question is, does AI become so good that you really don't have to? It's akin to saying, hey, we have all these heavy equipments to do house construction. Did we lose the ability to dig for example or lay out bricks? Yes. We are not practicing those kind of motions. Our bodies are kind of... We don't use our bodies that way. We go to gyms instead. We create an artificial kind of environment where we have to put our muscles to work, right? So it's something very similar. The question is, will it be needed? So in the current state of affairs, it is really... Even if you, let's say in that Android example, I know Kotlin, so I know what it did wrong. Really right now in the state as it is right now, even though it's improving, it gets things so often wrong that if you don't know that field or if you don't know the language or how to code or how to debug or how to understand, then you really cannot get far with it anyway. If you take Android as an example, I want to build this. If it doesn't work, I anyway have to understand Android at this point. And now I have to do two things. I have to understand Android and I have to go through all the code that the AI has to go And I have seen AI go completely off the hook and write a lot of code. I ask it to do something and it goes and writes a lot of code like more five files and some 3000, 5000 lines in the next 30 seconds. Wow. Okay. Now how do I know it's right?" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1077, "timestampFormatted": "17:57", "text": "Hmm" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1078, "timestampFormatted": "17:58", "text": "How do I know what it did is even right or is it optimum? it over designed? Did it just do it because I asked it to do it instead of telling me, hey, that's not a good idea. This problem shouldn't even exist because you're thinking about a problem which should not exist. It doesn't push back. And because of course it didn't occur to me. I don't know that either. And it went in the 30 seconds, it wrote 5,000 lines. And now I have to go read that for the next, I don't know how long. Hmm. So what I'm saying is that if you have to understand, so you ask AI to do something and you anyway have to have the expertise to understand what it's doing. So to answer your question, yes, the kids that pass out, well, let me preface it by saying that if AI stays where it is, which is prone to hallucinations, prone to a of mistakes at every level, then you still need the ability to know the basics. And you have, which means you have to build experience with it. And in fact, you probably have more work now because you have to anyway learn the basics and now you also have to go through AIS code too." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1130, "timestampFormatted": "18:50", "text": "Okay, yeah, that that makes sense. Now I know why we're not sitting here with a you know, globe or we're not making any predictions here. But what do you think I think there is one school of thought says that we are going to need a lot less software developers or software engineers in future given a lot of these work will be done by really smart AI's, assuming the rate of development if we see last three years, every year the capabilities have generally tended to get better and assuming if it goes, follows the same trajectory, maybe in a couple of years we may have a capable AI which will probably get over some of the things that you're pointing out, right? So there's one school of thought says that they're going to be lesser software engineers. lesser need of software developers. think these will be, you know, there'll be folks like you who's done this for very many years and can understand good from bad. And you'd be able to direct your AI to build these high quality software and review it for them, right? Now, another school of thought is going to be, is about, no, we are going to have explosion in software. where there are going to be millions of more websites and apps because the entry bar is so low. And given this whole field is going to explode, they're going to still... It's around... We're generally going to have more software and need for people who maintain those softwares, right? So where do you stand or at least based on what you've seen, what do you think... is going to happen or could happen." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1234, "timestampFormatted": "20:34", "text": "Yeah, mean your guess is as good as mine. I have no clue where it's heading. Most people don't know where it's heading exactly, right? Firstly, we know that whatever the highest end LLMs are doing, which is to spend millions into training a model which lasts only for a few months. And I don't think they're cutting even, right? Not even remotely, to the extent they spend and how much they reap out of it. Yeah, we don't know where it's going really and we don't know how much better it's going to get going forward. Yeah, is... Yeah, no one knows, right? Okay." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1266, "timestampFormatted": "21:06", "text": "But I certainly, if you ask me, I definitely believe in Jayvon's paradox. It's an economic phenomenon where it says that something that was meant to solve a problem intensifies it. For example, when there were horse carriages or bad roads, people thought that good roads is going to lead to, let's say, faster travel. and reduce the need for so much traveling. But what in turn what it did was, you know, it did reduce the travel time. So, you know, one single car or a carriage spend less time on the road, but there were too many of them. Same and better you could say that with, you know, electricity and something that, you know, is" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1293, "timestampFormatted": "21:33", "text": "Opened up new avenues." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1308, "timestampFormatted": "21:48", "text": "supposedly meant to put a check on it. So if we look at it from a lens of J1's paradox, while on surface it seems like eliminating the need for software development, given the explosion of software that we're going to have or we're going to see, right? Because now everybody's going to have a website or a web app and stuff like that. So the concept of what's called as a personal software, may... You know want to write software that you want always wanted and stuff like that So, you know which which basically mean that we'll have lot many software development the developers now again What could happen right and again putting a little bit of crystal ball thing here But what could happen is it could like you said right that if we go Take the gym analogy people lot many people might be writing software, but maybe getting paid in a different way, maybe to sell it to to service it to to maintain it or something else, right? You just start writing and getting paid, right? Maybe some but different aspect of that software is is getting you money, right? Or for maybe for performance tuning or for scaling." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1376, "timestampFormatted": "22:56", "text": "If it reaches a place where AI can write the app for you, whatever problem you're facing, let's say it can do a good job, good enough job that you don't actually need a software engineer, then you don't need a software engineer to do any of the things you said, performance monitoring or testing or bug fixing or maintaining. AI can do all of that. So it really depends on how good it gets. Let's assume the scenario where AI is so good and so cheap that it's like a very good developer in everyone's palms. I have an idea. want to, I want to, there's a, there's a itch I have. There's some problem I'm facing with my very peculiar way of life where one particular app can solve the problem. I asked this, this cheap LLM, cheap but smart LLM to read goes and makes it for me. And it installs the app for me. Everything just runs. Then what is the need for a software developer? Right. It really depends on if it is that good, then you probably don't need software developers. At the same time, if it is somewhat good, where it can do low-level coding, but it cannot think in large-term systems. It can solve a problem, but not see that the problem shouldn't even exist in the first place. That this problem existing is because there's a problem at a different layer. If it is not able to do that, then yes, you still need people who can think in terms of systems, in terms of user experience, in terms of are you really solving the domain's problem? Or are you just... creating a patchwork solution or a quick fix for the immediate pain that someone is having. So if you consider that also as part of someone designing software architecture to solve a problem, then I guess a software engineering job will remain because that work has to be done by someone thinking in terms of systems and architectures and right architectures and trade-offs. Yeah, you'll need someone to do it. But if AI is good enough to do that, so it all depends on how good it gets from you." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1476, "timestampFormatted": "24:36", "text": "So that also begs then look, probably if the trend continues, it is as a software development on for let's talk about one person here, right for an individual software development, it's going to save a lot of time, right? As in it's going to free free them up with lot of mundane stuff that they would would otherwise do but even half good AI" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1500, "timestampFormatted": "25:00", "text": "Seeing someone from mundane stuff is not the same as saving someone's time. You're assuming that when someone is freed from that mundane work, they have that time for themselves. But typically that's not what has happened. If you look at capitalism, if you look at how..." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1512, "timestampFormatted": "25:12", "text": "So which was my question then, let's say today if you're planning a feature, right? For example, a new feature in capability in your software that you paid to build. Now, are you factoring in these AI efficiencies in your planning the way you see people plan your work as opposed to how you're doing, let's say five years back?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1536, "timestampFormatted": "25:36", "text": "I didn't get the question. Can you read that?" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1537, "timestampFormatted": "25:37", "text": "Well, for example, like when you are building a new feature, let's say we used to do agile planning or Kanban planning and we're factoring in man hours and said, you know, if something were to take six months, five years back, would in the way of planning as a lead or somebody who's in charge of this development, Would you still factor in six months or do you factor in these efficiencies?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1568, "timestampFormatted": "26:08", "text": "I think these kinds of experiments are happening in lot of companies. I don't have first-hand knowledge of calculating it. Intuitively, it's not there at a level where it will have a significant impact to the six-month timeline. You can use the AI, but right now, as I said, you still have to go through all of it. It does produce poor quality code, at least to my experience till now. It is improving rapidly, but as of as it stands right now at the time of this recording, it is still something you get it out and you still have to handhold it. You still have to go through it, fix it. So it does save some time, but not a lot of time." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1599, "timestampFormatted": "26:39", "text": "Okay, yeah, for context today is 18 January 2026. Somebody is listening this in future. So yeah, a lot of our capabilities that we are discussing is as of today. And obviously, we have not tried everything that's out there obviously." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1612, "timestampFormatted": "26:52", "text": "And exactly, that's what I'm saying. my, everything I'm saying is in my experience, which is not necessarily the best here. Okay. To be fair, Claude code just released, think yesterday, I don't know what is the product, not yesterday, just last two, yeah, co-work, which is like an assistant that will manage your laptop for you. And they said they quoted it in two weeks using Claude code. So clearly what I'm saying is not the only answer here. Right. But at the same time, I think a lot of" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1624, "timestampFormatted": "27:04", "text": "Co-work." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1637, "timestampFormatted": "27:17", "text": "Enterprises are where I am right now so." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1640, "timestampFormatted": "27:20", "text": "Yeah, I mean, see, look, I personally feel that not all software is the same, mean, developing software for healthcare or bank or aerospace or something whose where people's lives are dependent on those software is not same as a website, for example." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1659, "timestampFormatted": "27:39", "text": "Still pretty impressive if you think about the kind of coding you would have to do if you had to do it manually. Teaching a software how to move a mouse and click something, drag it, drop it, delete a file, delete a folder, do all sort of stuff which you can do on a laptop as a human. Doing it in two weeks is very impressive. Doing all of that in two weeks is still very impressive. So I think we are getting there where the time it takes to get something out should reduce." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1680, "timestampFormatted": "28:00", "text": "Okay, sure. Shifting gears. Now let's talk about recruitment, hiring, given, you know, one of the one of the things that is also evolving is how people are getting hired, right? I've seen software interviews where people I remember my first interview where software where they asked me bunch of concept questions. and they also made me write some code on paper. Obviously, it was not the best code, but I felt it tested a lot of what I knew then. Now, obviously, industry has changed. A developer is supposed to use AI to do a lot of their work. Now, how do you... There are people who still interview the old way. The other school of thought is no, let's give them whatever tool they would anyways use. So what should be your recommendation in terms of if you're hiring for your team, right? How do you pick a good software developer, somebody who would add value to your team, right? What do you think should be a hiring sarji? given AI is doing all these heavy lifting." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1751, "timestampFormatted": "29:11", "text": "Yeah, that's a good question. you're right. think there are basically two major ways of interviewing that's being experimented right now in Big Tech. Personally, I think I probably incline, I'm still probably inclined towards the earlier way because the earlier way is to really see from a technical standpoint whether a person is able to think. I think that is the important thing here, right? The person you're talking to, they able to troubleshoot their own analysis of the problem? Like if I point out something wrong with that analysis, are they able to figure it out? Are they able to come over it? Look at the larger picture. If it's a senior role, are they thinking more holistically? Like when, if I give them a requirement, do they ask why that requirement is even there in the first place instead of just solving it? Are they able to say that this requirement doesn't actually make sense? That doing it this other way is probably going to not require this at all. Really coding is just one quote unquote small part of it. If you're hiring a junior engineer, equation changes there quite a bit because now AI can do some amount of that work. But you need senior engineers and those senior engineers are going to come from junior engineers, so you still need junior engineers anyway. But even then you need someone who is able to think. that fundamental requirement hasn't gone. Being able to think, being able to plan." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1823, "timestampFormatted": "30:23", "text": "Okay." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1826, "timestampFormatted": "30:26", "text": "Then the whole attitude aspect are they able to speak up are they able to? Take ownership of whatever they are doing all of it right so there is so much more to it than just the coding part" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1836, "timestampFormatted": "30:36", "text": "Okay, yeah, I think that makes sense. So that also makes me think and this is my thought now, which I believe that a lot of software developers in past has been quote unquote, software purists, which, know, in way of saying I'm a front end developer, I'm a backend developer, I'm a database guy, I'm a DevOps person, I'm a security engineer, I'm a perf. I'm an automation, I'm an SRE or so on and so forth, right? I'm a kernel engineer. I only do middlewares and so on. Do you think one person that sort of expertise boundary is going to go away in a way that given AI is going to touch a lot of what you do, now is everybody's going to be a full stack or the expectation is, you know, because AI is anyway is doing part of your work. Now you spend that time in probably understanding SRE or DevOps principles or maybe MLOps or why don't you do your databases? Why do we need? So everybody does everything just like how AI is doing." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1902, "timestampFormatted": "31:42", "text": "As humans, like to push boundaries, don't we? So we had, you go to store and you go buy stuff and then there used to be store guy coming and delivering stuff and then you had, you know, all the Amazon kind of model where everything, there's no physical store, but you you just sit and order and then comes. And then now there's quick commerce, like 10 minute delivery, 15 minute delivery. See, at the boundary, when you are at the limit of any system, whether it's the kernel, let's take kernel as an example, right? Schedulers. Schedulers is a very deep topic of its own. Okay? So let me ask you and let me answer this way. In the scenario where software engineering does exist despite AI, despite AI getting better, which means AI is not so good that they can remove the entire need for the industry. But they are still helpful and helpful in a predictable way that I can ask you to do something and it will do it reasonably well." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 1946, "timestampFormatted": "32:26", "text": "Yeah." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 1952, "timestampFormatted": "32:32", "text": "You still need that expertise to go deep into the scheduler code and see where there are more possibilities for optimization. Same with database. Yeah, so you have database and you can maybe leverage the AI to go manage it for you, take backups, make sure it's synchronized, database repairs, stuff like that. Is it going to be as good in depth of it as a human is? It all boils down to that. If it is as good in depth, then maybe what you're saying is true. Basically that means that if I want to build UI, AI is better at going into the depth of UI than I would be. And in that statement, I would be in what time frame? If I spend one month, will I be better than AI? Will I spend one year? Is when I'll get better than AI. If the answer is the second, then AI is pretty good. if the quality of AI is so good, that it'll take me a year to catch up to doing that level of UI work or its quality of code is or its depth of thinking is as good as a one year old experience, right? Now make it 10 years. Is it as good as a 10 year experienced guy? If it is that good, then what you're saying is possible." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2013, "timestampFormatted": "33:33", "text": "What I also meant is, let's say, if you're hiring for somebody in your team, Now, somebody showed up and said, look, I'm a great Java backend or Kotlin developer, right? And you give a bunch of problems and they do it for you, right? Now, let's say I come along and I demonstrate, 80 % of that capability that the previous person, but I said, look, but I also do UX. I also do build my own CI CD pipelines. And in fact, I also do my performance tuning. So would you hire the first person or the second person?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2045, "timestampFormatted": "34:05", "text": "I think I understand your question. Yeah, so the answer really depends on the industry. So in most service kind of industries, right, where you just have a problem you want to solve, you want to integrate system A and system B. Probably that 80 % guy is good enough. Okay. If you are hiring, even my team, if I want to hire, I'll probably be, I won't care about Kotlin, to be honest. I will care how good the person is as a software engineer. Language will going to be secondary. To that person, as it stands right now, the state of AI right now. I really don't see them as being able to just hand it off to you and just walking out of it. So I think today, as of today, at least in my team, if I were to hire, it's going to be expertise based or the ability to gain expertise. I need not find someone who is great at UI, but if he's a good software engineer, he or she is a good software engineer with the ability to dig deep and learn anything that I throw at that person. then I'll hire them. probably that person is going to become an expert in UI, UX, and maybe stick there for some time before moving to a different thing. I think that kind of stickiness will still be there because at least in my case, I like to push boundaries because wherever you are, you are going to have challenges which are really limiting that ecosystem. And you want to push that, right?" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2119, "timestampFormatted": "35:19", "text": "That's an interesting take." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2121, "timestampFormatted": "35:21", "text": "On the other hand, a service industry where all you want is to connect these two and that's it, right? It's perfectly fine too." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2126, "timestampFormatted": "35:26", "text": "Okay, now let's switching gears right. lot of our listeners are in college or in school who are planning to what advice would you give to say your younger self if let's say you were in college, second year of your uni, you know, studying say computer science and I'm sure you're looking at the industry and having a certain amount of anxiety on how this is all going to pan out the job prospect and so on, right? How do you think they should approach this knowing what you now know, especially around everything that has been going and the pace of improvements? How should one think about building their career in software?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2170, "timestampFormatted": "36:10", "text": "The first thing I would say is that if you're anxious, that's good because you're supposed to be anxious because it is indeed changing in a direction that no one knows, no one predicts. It could be risky. It could have a real impact in your job prospects. Yes, so that anxiety is very much warranted and that anxiety is telling you to be careful and that's good. That's a good thing to be. What can you do? As it stands right now, I don't see software engineering going anywhere. At the same time, I don't see it staying exactly how it is or it was two years ago. At the very least, means you have to learn and at least somewhat be on top of where AI is. Although as I said, I know it's extremely hard, the rate at which it's changing, the new protocols that are coming up every few months and some of them good, some of them bad. It is very tricky. But if you generally have the inclination to be on top of it and be reasonably on top of it, at least the trends that are sticking, rag for example, is a good idea that is likely to stick than not to. Some of the prompt engineering or context engineering, I think just being on top of the field and using it in your projects or in your own learning is going to be good idea." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2231, "timestampFormatted": "37:11", "text": "Okay, but don't you think I mean, I know, if you look at the curriculums in most of the universities, unless you're taking an AI as a course as a specialization, if you look at computer science, for example, when they teach operating system or algorithms and so on, right? These are still being taught with books that were probably written in 80s and 90s and probably before, right? The law of curriculum hasn't changed a whole lot, right? Still majorly it's the same. Now, while it is understood that yes, the out is changing. When I did my engineering about, you know, 15, 20 years back, I think even then there was a gap between academia and the industry. I believe that gap has further widened. So do... Do you suggest that, well, obviously you made the suggestion to be on top of the trends, but it's not systemic, which basically means that it's dependent on individual's capability to go out and figure things out and stay on top, right? How do we ensure that this message reaches everybody, every student who's out there who's even aspiring to come to the industry, right?" }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2309, "timestampFormatted": "38:29", "text": "Do you think universities are doing enough? You know, or should, so my point being is, do you think there is a solve, right? One is going to every union and building that awareness, which is not possible, but do you think they should contribute to open source, build better portfolio? Or is there a way where they could level up faster? Are there any specific suggestions would you give, right? For example, everybody should contribute in open source or should build personal software. I see a lot of flex on Twitter when people show their GitHub contribution last year to show how many greens they have got, right? So is that sort of a thing or going and talking about what they build on Twitter, building a brand for themselves, right? Is there a specific suggestion you'd like to give?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2359, "timestampFormatted": "39:19", "text": "I think these examples that are quoting is not going to scale well, right? mean, yeah, let's say everyone tries to have the most number of following. Only some of them are going to end up having a lot of following, right? Doing something, building your own project, having something to showcase in GitHub, they're all good suggestions, I think. And you are right. It's not a systemic thing. It'll take some time for... universities to adapt, curriculum to adapt and as we stand today, we don't even know where it's going. So how do you want to adapt? We don't know. You like, like this is a problem today, right? At every school level, we're writing essays, writing code, everything, writing anything. In fact, how do you make sure that the person wrote it not AI wrote it? Right. It's a, it's a problem that everyone is forced to handle, like universities, schools, professors, everyone is forced to handle all of it right now. It's going to take some struggle to figure out how to do. Maybe you'll go back to paper and pencil, going back to class and writing it in front instead of online and all that. Being generally good involves you face a problem and you try to find a solution. And in the path to find a solution, either code it by yourself, coding being a specific way to solve a problem, but you code yourself or use AI to code it. This general attitude towards solving problem is what you need overall." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2426, "timestampFormatted": "40:26", "text": "Okay, so thanks. And I think I'll ask you one last question and probably we'll wind this up. Now, this question is about specialization, right? Specialization. A lot of people, a of software developers today when they enter industry, they're given this advice on if they should pick up a specific specialization, right? For instance, retail is a specialization. or a certain industry specialization or a certain area in retail or banking or fintech or robotics or insurance or underwriting and so on. Now, do you think that, I mean, complimenting the, assuming they are software developers, right? the domain expertise going to be a mode? Should they invest more into learning the nuances of domain given they have some free time now or the efficiencies gained by AI doing some of their work. Should they invest in building that domain skill or do you think they should focus on the other area of say software development, right? Like I said, DevOps, SRE, ML Ops, data engineering, or should they spend that time in building domain expertise?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2497, "timestampFormatted": "41:37", "text": "Firstly, you're assuming that kids will have more time because there is some amount of automation. They may not have more time. Academy will catch up. The kinds of assignments will change. They may not have more time. Is more expertise or domain level expertise or things like... So domain specific knowledge, I don't know if it's a good idea to build it upfront. It's not clear to me. Like in finite." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2518, "timestampFormatted": "41:58", "text": "not in the... Yeah, but when they're in first or first few years of the job." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2523, "timestampFormatted": "42:03", "text": "think as things are evolving, think you will naturally be pushed to if you have more time because you are not spending all the time writing low level code, assuming of course that AI is good enough that you don't have to debug it, assuming it works, assuming that AI is good enough that you don't have to deal with all those nitty gritties, yeah, you will be free to actually solve larger problem. Absolutely. That's what will happen." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2545, "timestampFormatted": "42:25", "text": "Right. So do you think and again, is another I sorry, I lied. I'll ask you another question. Let's let's take a best case scenario. Right. Let's assume that in a year's time or maybe two, the the AI or the LMS gets so good that they come over not just deficiencies of today, but build new capabilities in terms of depth, speed, and so on, right? Do you think every software, and assuming if such a world exists, where LLMs have become so good, better than the 10x engineers of, let's say, the top of our industry, right? Do you think every software which has probably existed for 30 last 30 40 years will have to be rewritten with a new intelligence, right? New sort of retail software, new sort of search, new sort of, you know, airline software, new sort of traffic management, new sort of finance, trading, cons, mining, you know, like because some of these softwares have stayed and, you know, been created or but would we have a need to revisit every single software?" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2615, "timestampFormatted": "43:35", "text": "I think the industry forces will automatically push you there. Right? So you have a smarter way to do something. There will be someone who will implement it. The others will have to catch up. Take traffic for example, smart traffic management. Like today, at least in US, there are these strips under the road to detect if there's a car. But it cannot tell how many layers of car are there, like how far it goes. I think Bangalore has a better system. I think it uses cameras to determine the depth of traffic. And so, yeah, of course, I mean, it's going to happen. It's bound to happen. What we have now is a new kind of tool. So we have this tool called software. We have a tool called internet. And it took a long time. It proliferated everywhere. We have this tool called phone and very cheap internet and accessibility everywhere. It just proliferated everywhere. in fact, a lot of websites, a lot of services are now more optimized towards an app than they are. actual desktop websites, right? This is the natural evolution. Now you have new tool which is code level intelligence. mean, intelligence that can be modeled in a device, right? Of course, it's going to go everywhere like a refrigerator which can tell you that it's out of milk. Yeah, a stove which has, you know, which tells that it's on and you know, your smart home devices know that you have left the house and the stove says it's on. Somehow there is a system which says, hey, there is no one at home, but the stove is on." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2678, "timestampFormatted": "44:38", "text": "or something is going to go stale." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2692, "timestampFormatted": "44:52", "text": "I mean, we are really at a place where our imagination is the limit at this point." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2697, "timestampFormatted": "44:57", "text": "That makes me believe that while the number of software developers may reduce, but the need for software is only going to explode, right? Because knowing" }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2709, "timestampFormatted": "45:09", "text": "The real threat, if I want to call that word, is that a machine is now doing what it could never do before. We have had machines since the beginning of humans, right? We learned to make stones and spears out of stones and wood and so we have been building machines forever. We built wheel and bullock carts and we've been building. But till now there is one domain of expertise that machines could not reach which was The human ingenuity, creativity, ability to solve problems using intellect is something that machines have not been able to do. And so whatever has happened till now is that every time, quite literally, have... there was a... at my place, I saw the people had come to cut a tree. There were two men and a huge machine. That machine was able to hold the tree. It had a saw to cut the tree. and then lift the tree and then put it in its own wood chipper which was integrated to that machine and chip it down into small small pieces and two people and a machine that's all it took. Okay. So what so a hundred two hundred years ago how would maybe it will require 10 people to do that. So what are those other remaining eight people doing. They are doing some sort of mentally intellectually difficult work that machines could not do till now. The problem with AI as it stands maybe it's not a problem but where it's heading towards is It's able to do this one thing which is still unique to humans, which is the ability to think. And so we don't really know where it's going to go. even if you're right, we can automate these things, our imagination is the limit, right? But for everything we imagine, we employ AI to do it because it's cheap for us. Then we never really know what humans are going to do. And there's a whole stream of thought about universal basic income and that there's a possibility, think Musk suggested that a few weeks ago that in just a few, he suggested that we probably shouldn't be worrying about saving up for retirement because just in a few decades, there's going to be enough wealth that you don't have to worry about saving money for retirement. You'll be taken care of. But that's not how it's been. Capitalism has not spread that wealth to everyone, right? It has still always concentrated higher up. That's how it is. So we really don't know where it's going." }, { "speaker": "Vikas", "speakerRaw": "Speaker 1", "timestamp": 2836, "timestampFormatted": "47:16", "text": "Okay, yeah, thanks. That's some interesting take on what's happening. I think we've pretty long episode and I think we should wind it as much as I enjoy this conversation. think, you know, we should keep keep this to do it for today. And thank you, Srikanth. It has been a wonderful discussion. Thank you for coming to the pod. And I assure you that we will do, you know, maybe another one of this in a year's time and see how You know things evolved. how some stuff that we spoken about today holds versus change, right? And to our listeners, I hope you found this discussion engaging and if you have your thought, feel free to comment right to us. We also have a new website which you feel free to engage and discuss. And thank you for being a great listener and as always, everything that we spoke here on this episode represents our own opinion and does not represent that of our employees and organizations that we work. Thank you and thank you Srikant. Thank you. Yeah, bye." }, { "speaker": "Srikanth", "speakerRaw": "Speaker 2", "timestamp": 2897, "timestampFormatted": "48:17", "text": "Thanks for having me." } ], "chapters": [ { "title": "Hi, welcome to a new episode of the AI Valley podc...", "timestamp": 52, "timestampFormatted": "0:52" }, { "title": "So the last part is rather straightforward", "timestamp": 206, "timestampFormatted": "3:26" }, { "title": "I see", "timestamp": 337, "timestampFormatted": "5:37" }, { "title": "So yeah, you spoke about a lot of different themes...", "timestamp": 492, "timestampFormatted": "8:12" }, { "title": "I see", "timestamp": 627, "timestampFormatted": "10:27" }, { "title": "Talking about wow, let me ask you something", "timestamp": 741, "timestampFormatted": "12:21" }, { "title": "Okay, it's more instant", "timestamp": 890, "timestampFormatted": "14:50" }, { "title": "Hmm How do I know what it did is even right or is ...", "timestamp": 1077, "timestampFormatted": "17:57" }, { "title": "Yeah, mean your guess is as good as mine", "timestamp": 1234, "timestampFormatted": "20:34" }, { "title": "If it reaches a place where AI can write the app f...", "timestamp": 1376, "timestampFormatted": "22:56" }, { "title": "Seeing someone from mundane stuff is not the same ...", "timestamp": 1500, "timestampFormatted": "25:00" }, { "title": "Co-work", "timestamp": 1624, "timestampFormatted": "27:04" }, { "title": "Okay", "timestamp": 1823, "timestampFormatted": "30:23" }, { "title": "Yeah", "timestamp": 1946, "timestampFormatted": "32:26" }, { "title": "That's an interesting take", "timestamp": 2119, "timestampFormatted": "35:19" }, { "title": "Okay, but don't you think I mean, I know, if you l...", "timestamp": 2231, "timestampFormatted": "37:11" }, { "title": "Okay, so thanks", "timestamp": 2426, "timestampFormatted": "40:26" }, { "title": "I think the industry forces will automatically pus...", "timestamp": 2615, "timestampFormatted": "43:35" }, { "title": "Okay, yeah, thanks", "timestamp": 2836, "timestampFormatted": "47:16" } ], "quotes": [ { "text": "And yeah, last two years also have been pretty interesting in the way things are progressing and the speed of change at the lack of visibility as to where it's going.", "speaker": "Srikanth", "timestamp": 129, "timestampFormatted": "2:09" }, { "text": "So you can build something, you can quickly go through some iterations and get something which looks like it should work.", "speaker": "Srikanth", "timestamp": 492, "timestampFormatted": "8:12" }, { "text": "One of the reason why, you know, I studied computer science engineering and when I first time I built something worthwhile, wrote, you this was, I think it was in turbo sea or boardland sea.", "speaker": "Vikas", "timestamp": 741, "timestampFormatted": "12:21" }, { "text": "working through the night if 80 % or let's say 40 % of it is just trivial really trivial boring stuff and that goes away it's it's I think it's for the win", "speaker": "Srikanth", "timestamp": 911, "timestampFormatted": "15:11" }, { "text": "Okay, so shifting gears, Shikant, I think now let's talk about some nuanced stuff, Something that, you know, a lot of has been on lot of people's mind, I'm certainly sure people who are listening to this podcast will must be thinking about it is around how should a software engineer of today, right?", "speaker": "Vikas", "timestamp": 916, "timestampFormatted": "15:16" } ], "fullText": "[0:52] Vikas: Hi, welcome to a new episode of the AI Valley podcast. My name is Vikas and today we are going to talk about software development and AI. This is something that clearly has been the center of debate in many circles. Given it's having a lot of impact in the way software is written, developed, one of the places where it's disrupting more than other places. So we thought we should have an objective discussion on the topic. And today we have a very special guest, Srikant, who works in a big tech in US. He's been coding for very many years. I've known him as one of the best software developers over the years. So I thought, who better than somebody who writes code for living to get him on the pod and have a sort of a open-ended discussion on how he see as a software writer or software developer, this field evolve and how it's impacting. Welcome to the podcast, Srikant.\n\n[1:50] Srikanth: Thanks because good to be here.\n\n[1:52] Vikas: Subab, so let's get into the center of the discussion. How long have you been coding and do you think the software development as a field is changing, especially in last couple of years with the advent of AI? Do you think there is a way it has been impacted?\n\n[2:09] Srikanth: Yeah, so I have been working for 18 and a half years in this as a developer architect. And yeah, last two years also have been pretty interesting in the way things are progressing and the speed of change at the lack of visibility as to where it's going. I think it's been very, yeah, the change is constantly there. We are right in the middle of it. So it's hard to know exactly where we're going to end up. but it's still certainly very interesting.\n\n[2:34] Vikas: Okay, so now I clearly remember I've met you around the same time for last couple of years and I've asked you same question and your answer has been definitely not the same every year because every year when we meet after there is a lot goes into the model capability, there are new models, new tool around it, but somebody who writes code Do you think there is a lot of fear in lot of software developers in the way that this AI may be coming for their jobs and there are people with mixed opinions. I have known people who stand on either side of the debate. Now somebody who write code, do you think has any AI helped you personally write better code? Or do you think AI can write better code than you?\n\n[3:26] Srikanth: So the last part is rather straightforward. To the extent I have used AI, it doesn't write better code than me. In fact, in any object you measure, it's not good. It's a lot of repetitive functions and a lot of copy pasted stuff. So in terms of quality of code, don't think it's particularly great right now. Where it has been useful for me is in completely greenfield areas. Like if there's a Something I want to simulate, like the other day I I wanted to have a solution to a particular problem. Basically how, like if you have a hundred thousand different items and there is a 80-20 rule, the Pareto principle is being used in an item being selected. So there's basically bias towards some set of items which are selected often. This is generally true for the e-commerce industry, right? There's a bunch of items which are ordered a lot more than other items, right? If let's say you have a corpus of items, 10,000, 100,000, some pool of items and of which there is a clear 80-20 pattern and recursively 80-20 pattern. the 20 % which is sold 80%, within that also there's a 20 % which is sold 80%. So there's generally that kind of pattern. Though the exact 20-80 ratio is debatable, but the pattern is pretty much there. So I wanted to see if I have a pool of 10,000 or 100,000 items and I have a thousand requests for... one item each and then the items follow this kind of a pattern like Pareto principle. Then how many different items or how many unique items will be there among a thousand requests? So if it's completely random, close to a thousand unique items, right? But it won't be completely random. So I had this thought and yeah, I asked AI, okay, how do I do this? And then what it said, I double checked. It turned out that there is no direct way to calculate it. You can simulate it. And then it took maybe another five minutes for it to write a simulation code. And it so happens that in that iteration, the code was accurate, like it ran correctly. And so in that sense, it's really great in the sense that, there's an idea. What would have taken me a lot more time to go? It's because primarily it's, I'm doing something else and this is a thought that comes in between. And it's a kind of distraction from the main goal, right? And the earlier version before AI, I would have had to take the detour. which is a kind of distraction from the main purpose and so faster iterations in those are are certainly welcome.\n\n[5:37] Vikas: I see. Yeah. So I see what you're saying. You know, I agree. It's clearly easy to get into these rabbit holes with AI and quickly experiment things around bounce ideas. And it's a great partner to write mundane stuff, something that would have taken a lot more time for somebody. Yeah, I think that makes sense. one, because obviously, for the first 10 years of my career, I was writing code every day. Probably, you know, I kept doing that. But these days I don't write code for living these days. But one of the things that I have seen, in fact, the AI based coding or AI coding has certainly impressed people who did not code, right? Think of your product managers, think non-technical co-founders or people who sort of for whatever reason, good or bad, were not coders or didn't code for living, right? I've seen those people amazed at this capability or newfound capability where they could quickly, you know, build website or, know, for example, there's a web UX designer who's great with designs, but they don't know the intricacies of building a web front end. Now they have a newfound sort of skill. that they can complement their designing skills. those people have certainly been on the beneficial side of it. But I understand people who are, say, what I call software purists, they are getting very impressed. It's akin to a chef who's spent all his life or his or her life, you know, creating dishes, crafting dishes. And suddenly there is this robot or machine that spins out, spits out. these dishes. It's obviously not going to or probably not going to be the same kind of quality as that of a chef who has done this for life. So I understand there is a similar tension that exists here. But\n\n[7:32] Srikanth: But let's see.\n\n[7:33] Vikas: If you are able to understand the repetitive stuff or the lack of perfect code in the language that you've coded for many years, for example, in your case, could be Java or Kotlin or something like that. But let's say, if I told you that if you have to write something in Rust or say C++ or Go or even some mundane language, don't you think having an AI which is not great but does maybe good enough would help. Right. So what do you think about that? Right. Because it's certainly giving you breadth if not depth. So and so don't you think that's generally a good stuff in the\n\n[8:12] Srikanth: So yeah, you spoke about a lot of different themes here. So people who are not professional coders, yes, they are very happy. And coders are not that happy. And there is a reason for that. As you said, like chefs, they know the art well. And so that also enables them to see the problems in the implementation done by AI, for example, and done by others also. because you have been doing it, see someone who hasn't coded, someone product management or UX designer. who doesn't have exposure to code, yes, they will feel like they have new wings because now they can take the idea and then actually see something that is there, right? But there is such a thing as software quality. So you can build something, you can quickly go through some iterations and get something which looks like it should work. And so when they look at it, they have now uncovered a new skill that they wouldn't have been able to do earlier, right? So that makes sense. In my experience, to the extent I've used AI, it's very easy to see that it doesn't work well also. So I gave the example of Pareto principle. So there was a one, two file Python thing, and it works well in those scenarios. I've written, and going to the example, you took a nice example of Rust and something which have a language or ecosystem that I've not worked in. I can use AI to build stuff. Yes, you are absolutely right. And I have done that, and I have seen both sides of it where it does something which it would have taken me a long time to do, and it does something obviously wrong as well. one example is that I wanted to write an Android application. I have zero experience with Android programming. In fact, I have zero experience in any sort of UI programming. So to get started, I told it this what I want. I give detailed instructions on what I wanted, exactly how I envisioned it. And it built something like that. The positives is that, this entire Android manifest file, the resources, the icons, I didn't have to think about any of it, where to place them, how to bootstrap an Android project, none of them. It just went ahead and did all of that well. And I know it's well because it actually worked. But let's take that as a proof that it worked correctly, right? But the business logic that I wanted, I could clearly see there are bugs and nuances. in Kotlin, there's something called coroutines, which is, I asked it to use that, but it did it in a blocking way, which is defies the whole point of coroutines. So what is the outcome? So you can make an app, it runs, but frequently it just hangs your UI. So now,\n\n[10:27] Vikas: I see.\n\n[10:35] Srikanth: as a UX guy, the fact that I can have an app out there without the need for a development team is great. At the same time, someone has to go figure this out. Right? Now you've gotten out some code. That's where the word slop is a perfect fit. So it's there. And, but someone has to now go through all that code and figure out what are the bugs in it. So it's, it's very nuanced. It is useful, but it's not a silver bullet at the same time.\n\n[10:54] Vikas: Yeah, yeah, yeah. I understand that makes sense and honestly I really see what you're saying and I've seen sometime some pieces of code impresses me in a way 0 to 1 is amazing but sometimes the obvious mistakes or the loops that it goes one of my biggest pet peeves is sometimes you're trying to solve even if you're told that you want to solve a certain thing in a certain way and if it if it is stuck somewhere, if you don't pay attention, it will take an alternate route, something that you would not have asked or clearly asked not to go.\n\n[11:31] Srikanth: Oh, so I had so this is a frequent thing. I think a lot of people have noticed it in lot of podcasts as well. A lot of people point out that when you say, make this test work, it sometimes goes, deletes the test. So exactly. So it can be wow on both sides. If you if you approach it, saying, thinking that, dude, this is extremely useless. It's a negative utility and you go use it. It can wow you. Wow. It can like the same so much time.\n\n[11:42] Vikas: Wow.\n\n[11:56] Srikanth: And now you are happy and you think, wow, it's going to be so useful. And then the next thing you ask, now I have frequently had this issue where it goes into a loop. It doesn't figure out what the problem is. It goes into a loop. And then I ask it to write a test case. It writes a test case and just comments everything out. But then if I go uncommitted and say, now make sure this test runs, it'll run a few times and then it says, let's fix it. And then it just comments out the whole test case. So it's wow on both sides. That's exactly right. It is wow on both ways.\n\n[12:21] Vikas: Talking about wow, let me ask you something. One of the reason why, you know, I studied computer science engineering and when I first time I built something worthwhile, wrote, you this was, I think it was in turbo sea or boardland sea. And I first time in my life when I built something worthwhile and when it worked, there was a... I don't know how to describe that. There was a certain kind of euphoria or a dopamine hit when something was, which was obviously not working, but I got it working. And it was one of the main driver or the kick that pulled me in the world of software engineering, because the joy of making something, building something gave you certain kind of happiness, right? Now, the question that I have for you is, I obviously don't get those that sort of a same sort of a kick anymore because I'm not writing code for living. if do you think that one gets a similar kind of dopamine hit when they build something with AI or debug something with AI? I don't, you know, what's your thought?\n\n[13:27] Srikanth: In the current state of AI, as I said, it's a wow on both sides, right? So, in a way, you get a dopamine hit when AI gets it right.\n\n[13:33] Vikas: I see. But is it the same feeling of you working through the night and fixing a hairy bug? It's the same feeling?\n\n[13:42] Srikanth: I think it's a very subjective thing. I know people who really like to debug stuff. I personally don't enjoy that. Literally going by your example. But the general point is that you're still building stuff. You're still progressing in the direction you want. There is a reason you're coding. So you have something in your mind that you want to make happen, make concrete in real life through software. so you're still getting that done. you if you use AI to solve some particular sub problem, at least as it stands today, you have to verify it anyway. So I don't think that that that that aspect of having created something has gone. And there is it has certainly added this new fun part, Ropamine hit where things actually work with AI. Especially when you that Android project, for example, the fact that it even loaded even though it had bugs and all that, the fact that it helped me remove Especially things that were not of interest to me. Understanding where Android manifest has to be and how icons have to be, where they have to be put and how they have to be referenced. Not having to understand all that nitty gritty. I don't think there was too much value of that, you know, from the get go anyway. What I wanted is the outcome and I achieved that outcome even sooner. In some ways it's even more addictive. It's more of dopamine.\n\n[14:50] Vikas: Okay, it's more instant. Yeah.\n\n[14:52] Srikanth: The lesser effort more guaranteed, but not more guaranteed. But when you use it properly, it certainly reduces the effort that's needed in that direction.\n\n[15:01] Vikas: Yeah, the dropout rates reduces for sure, right? Because, you know, it's good to be to work through the night and get it working. But I'm sure a lot of people give up.\n\n[15:11] Srikanth: working through the night if 80 % or let's say 40 % of it is just trivial really trivial boring stuff and that goes away it's it's I think it's for the win\n\n[15:16] Vikas: Yeah. Okay, so shifting gears, Shikant, I think now let's talk about some nuanced stuff, Something that, you know, a lot of has been on lot of people's mind, I'm certainly sure people who are listening to this podcast will must be thinking about it is around how should a software engineer of today, right? There a lot of people who might be studying to be a software engineer. Now as we speak, right? How should they develop or think about all of these things, right? For example, say you worked in, say, maybe retail domain or e-commerce or any specific domain, right? Now you've learned through multiple iteration, gone through, the hard way and became a domain expert. Now imagine if somebody, so that you are domain expert enough to point out the problems. in the way of AI implementation. But let's say somebody who's just coming out of college, they'll never probably get an opportunity to go at the depth that you had to or somebody else had to who was doing it the hard way. they ever get to a point or will software developer lose the skill to differentiate between good and bad?\n\n[16:27] Srikanth: It really depends on whether that skill has a value in the future. So, okay, if I don't go into the weeds of it, I certainly lose the ability to go into the weeds of it or understand the language at depth. Yes, that's a rather obvious yes. The question is, does AI become so good that you really don't have to? It's akin to saying, hey, we have all these heavy equipments to do house construction. Did we lose the ability to dig for example or lay out bricks? Yes. We are not practicing those kind of motions. Our bodies are kind of... We don't use our bodies that way. We go to gyms instead. We create an artificial kind of environment where we have to put our muscles to work, right? So it's something very similar. The question is, will it be needed? So in the current state of affairs, it is really... Even if you, let's say in that Android example, I know Kotlin, so I know what it did wrong. Really right now in the state as it is right now, even though it's improving, it gets things so often wrong that if you don't know that field or if you don't know the language or how to code or how to debug or how to understand, then you really cannot get far with it anyway. If you take Android as an example, I want to build this. If it doesn't work, I anyway have to understand Android at this point. And now I have to do two things. I have to understand Android and I have to go through all the code that the AI has to go And I have seen AI go completely off the hook and write a lot of code. I ask it to do something and it goes and writes a lot of code like more five files and some 3000, 5000 lines in the next 30 seconds. Wow. Okay. Now how do I know it's right?\n\n[17:57] Vikas: Hmm\n\n[17:58] Srikanth: How do I know what it did is even right or is it optimum? it over designed? Did it just do it because I asked it to do it instead of telling me, hey, that's not a good idea. This problem shouldn't even exist because you're thinking about a problem which should not exist. It doesn't push back. And because of course it didn't occur to me. I don't know that either. And it went in the 30 seconds, it wrote 5,000 lines. And now I have to go read that for the next, I don't know how long. Hmm. So what I'm saying is that if you have to understand, so you ask AI to do something and you anyway have to have the expertise to understand what it's doing. So to answer your question, yes, the kids that pass out, well, let me preface it by saying that if AI stays where it is, which is prone to hallucinations, prone to a of mistakes at every level, then you still need the ability to know the basics. And you have, which means you have to build experience with it. And in fact, you probably have more work now because you have to anyway learn the basics and now you also have to go through AIS code too.\n\n[18:50] Vikas: Okay, yeah, that that makes sense. Now I know why we're not sitting here with a you know, globe or we're not making any predictions here. But what do you think I think there is one school of thought says that we are going to need a lot less software developers or software engineers in future given a lot of these work will be done by really smart AI's, assuming the rate of development if we see last three years, every year the capabilities have generally tended to get better and assuming if it goes, follows the same trajectory, maybe in a couple of years we may have a capable AI which will probably get over some of the things that you're pointing out, right? So there's one school of thought says that they're going to be lesser software engineers. lesser need of software developers. think these will be, you know, there'll be folks like you who's done this for very many years and can understand good from bad. And you'd be able to direct your AI to build these high quality software and review it for them, right? Now, another school of thought is going to be, is about, no, we are going to have explosion in software. where there are going to be millions of more websites and apps because the entry bar is so low. And given this whole field is going to explode, they're going to still... It's around... We're generally going to have more software and need for people who maintain those softwares, right? So where do you stand or at least based on what you've seen, what do you think... is going to happen or could happen.\n\n[20:34] Srikanth: Yeah, mean your guess is as good as mine. I have no clue where it's heading. Most people don't know where it's heading exactly, right? Firstly, we know that whatever the highest end LLMs are doing, which is to spend millions into training a model which lasts only for a few months. And I don't think they're cutting even, right? Not even remotely, to the extent they spend and how much they reap out of it. Yeah, we don't know where it's going really and we don't know how much better it's going to get going forward. Yeah, is... Yeah, no one knows, right? Okay.\n\n[21:06] Vikas: But I certainly, if you ask me, I definitely believe in Jayvon's paradox. It's an economic phenomenon where it says that something that was meant to solve a problem intensifies it. For example, when there were horse carriages or bad roads, people thought that good roads is going to lead to, let's say, faster travel. and reduce the need for so much traveling. But what in turn what it did was, you know, it did reduce the travel time. So, you know, one single car or a carriage spend less time on the road, but there were too many of them. Same and better you could say that with, you know, electricity and something that, you know, is\n\n[21:33] Srikanth: Opened up new avenues.\n\n[21:48] Vikas: supposedly meant to put a check on it. So if we look at it from a lens of J1's paradox, while on surface it seems like eliminating the need for software development, given the explosion of software that we're going to have or we're going to see, right? Because now everybody's going to have a website or a web app and stuff like that. So the concept of what's called as a personal software, may... You know want to write software that you want always wanted and stuff like that So, you know which which basically mean that we'll have lot many software development the developers now again What could happen right and again putting a little bit of crystal ball thing here But what could happen is it could like you said right that if we go Take the gym analogy people lot many people might be writing software, but maybe getting paid in a different way, maybe to sell it to to service it to to maintain it or something else, right? You just start writing and getting paid, right? Maybe some but different aspect of that software is is getting you money, right? Or for maybe for performance tuning or for scaling.\n\n[22:56] Srikanth: If it reaches a place where AI can write the app for you, whatever problem you're facing, let's say it can do a good job, good enough job that you don't actually need a software engineer, then you don't need a software engineer to do any of the things you said, performance monitoring or testing or bug fixing or maintaining. AI can do all of that. So it really depends on how good it gets. Let's assume the scenario where AI is so good and so cheap that it's like a very good developer in everyone's palms. I have an idea. want to, I want to, there's a, there's a itch I have. There's some problem I'm facing with my very peculiar way of life where one particular app can solve the problem. I asked this, this cheap LLM, cheap but smart LLM to read goes and makes it for me. And it installs the app for me. Everything just runs. Then what is the need for a software developer? Right. It really depends on if it is that good, then you probably don't need software developers. At the same time, if it is somewhat good, where it can do low-level coding, but it cannot think in large-term systems. It can solve a problem, but not see that the problem shouldn't even exist in the first place. That this problem existing is because there's a problem at a different layer. If it is not able to do that, then yes, you still need people who can think in terms of systems, in terms of user experience, in terms of are you really solving the domain's problem? Or are you just... creating a patchwork solution or a quick fix for the immediate pain that someone is having. So if you consider that also as part of someone designing software architecture to solve a problem, then I guess a software engineering job will remain because that work has to be done by someone thinking in terms of systems and architectures and right architectures and trade-offs. Yeah, you'll need someone to do it. But if AI is good enough to do that, so it all depends on how good it gets from you.\n\n[24:36] Vikas: So that also begs then look, probably if the trend continues, it is as a software development on for let's talk about one person here, right for an individual software development, it's going to save a lot of time, right? As in it's going to free free them up with lot of mundane stuff that they would would otherwise do but even half good AI\n\n[25:00] Srikanth: Seeing someone from mundane stuff is not the same as saving someone's time. You're assuming that when someone is freed from that mundane work, they have that time for themselves. But typically that's not what has happened. If you look at capitalism, if you look at how...\n\n[25:12] Vikas: So which was my question then, let's say today if you're planning a feature, right? For example, a new feature in capability in your software that you paid to build. Now, are you factoring in these AI efficiencies in your planning the way you see people plan your work as opposed to how you're doing, let's say five years back?\n\n[25:36] Srikanth: I didn't get the question. Can you read that?\n\n[25:37] Vikas: Well, for example, like when you are building a new feature, let's say we used to do agile planning or Kanban planning and we're factoring in man hours and said, you know, if something were to take six months, five years back, would in the way of planning as a lead or somebody who's in charge of this development, Would you still factor in six months or do you factor in these efficiencies?\n\n[26:08] Srikanth: I think these kinds of experiments are happening in lot of companies. I don't have first-hand knowledge of calculating it. Intuitively, it's not there at a level where it will have a significant impact to the six-month timeline. You can use the AI, but right now, as I said, you still have to go through all of it. It does produce poor quality code, at least to my experience till now. It is improving rapidly, but as of as it stands right now at the time of this recording, it is still something you get it out and you still have to handhold it. You still have to go through it, fix it. So it does save some time, but not a lot of time.\n\n[26:39] Vikas: Okay, yeah, for context today is 18 January 2026. Somebody is listening this in future. So yeah, a lot of our capabilities that we are discussing is as of today. And obviously, we have not tried everything that's out there obviously.\n\n[26:52] Srikanth: And exactly, that's what I'm saying. my, everything I'm saying is in my experience, which is not necessarily the best here. Okay. To be fair, Claude code just released, think yesterday, I don't know what is the product, not yesterday, just last two, yeah, co-work, which is like an assistant that will manage your laptop for you. And they said they quoted it in two weeks using Claude code. So clearly what I'm saying is not the only answer here. Right. But at the same time, I think a lot of\n\n[27:04] Vikas: Co-work.\n\n[27:17] Srikanth: Enterprises are where I am right now so.\n\n[27:20] Vikas: Yeah, I mean, see, look, I personally feel that not all software is the same, mean, developing software for healthcare or bank or aerospace or something whose where people's lives are dependent on those software is not same as a website, for example.\n\n[27:39] Srikanth: Still pretty impressive if you think about the kind of coding you would have to do if you had to do it manually. Teaching a software how to move a mouse and click something, drag it, drop it, delete a file, delete a folder, do all sort of stuff which you can do on a laptop as a human. Doing it in two weeks is very impressive. Doing all of that in two weeks is still very impressive. So I think we are getting there where the time it takes to get something out should reduce.\n\n[28:00] Vikas: Okay, sure. Shifting gears. Now let's talk about recruitment, hiring, given, you know, one of the one of the things that is also evolving is how people are getting hired, right? I've seen software interviews where people I remember my first interview where software where they asked me bunch of concept questions. and they also made me write some code on paper. Obviously, it was not the best code, but I felt it tested a lot of what I knew then. Now, obviously, industry has changed. A developer is supposed to use AI to do a lot of their work. Now, how do you... There are people who still interview the old way. The other school of thought is no, let's give them whatever tool they would anyways use. So what should be your recommendation in terms of if you're hiring for your team, right? How do you pick a good software developer, somebody who would add value to your team, right? What do you think should be a hiring sarji? given AI is doing all these heavy lifting.\n\n[29:11] Srikanth: Yeah, that's a good question. you're right. think there are basically two major ways of interviewing that's being experimented right now in Big Tech. Personally, I think I probably incline, I'm still probably inclined towards the earlier way because the earlier way is to really see from a technical standpoint whether a person is able to think. I think that is the important thing here, right? The person you're talking to, they able to troubleshoot their own analysis of the problem? Like if I point out something wrong with that analysis, are they able to figure it out? Are they able to come over it? Look at the larger picture. If it's a senior role, are they thinking more holistically? Like when, if I give them a requirement, do they ask why that requirement is even there in the first place instead of just solving it? Are they able to say that this requirement doesn't actually make sense? That doing it this other way is probably going to not require this at all. Really coding is just one quote unquote small part of it. If you're hiring a junior engineer, equation changes there quite a bit because now AI can do some amount of that work. But you need senior engineers and those senior engineers are going to come from junior engineers, so you still need junior engineers anyway. But even then you need someone who is able to think. that fundamental requirement hasn't gone. Being able to think, being able to plan.\n\n[30:23] Vikas: Okay.\n\n[30:26] Srikanth: Then the whole attitude aspect are they able to speak up are they able to? Take ownership of whatever they are doing all of it right so there is so much more to it than just the coding part\n\n[30:36] Vikas: Okay, yeah, I think that makes sense. So that also makes me think and this is my thought now, which I believe that a lot of software developers in past has been quote unquote, software purists, which, know, in way of saying I'm a front end developer, I'm a backend developer, I'm a database guy, I'm a DevOps person, I'm a security engineer, I'm a perf. I'm an automation, I'm an SRE or so on and so forth, right? I'm a kernel engineer. I only do middlewares and so on. Do you think one person that sort of expertise boundary is going to go away in a way that given AI is going to touch a lot of what you do, now is everybody's going to be a full stack or the expectation is, you know, because AI is anyway is doing part of your work. Now you spend that time in probably understanding SRE or DevOps principles or maybe MLOps or why don't you do your databases? Why do we need? So everybody does everything just like how AI is doing.\n\n[31:42] Srikanth: As humans, like to push boundaries, don't we? So we had, you go to store and you go buy stuff and then there used to be store guy coming and delivering stuff and then you had, you know, all the Amazon kind of model where everything, there's no physical store, but you you just sit and order and then comes. And then now there's quick commerce, like 10 minute delivery, 15 minute delivery. See, at the boundary, when you are at the limit of any system, whether it's the kernel, let's take kernel as an example, right? Schedulers. Schedulers is a very deep topic of its own. Okay? So let me ask you and let me answer this way. In the scenario where software engineering does exist despite AI, despite AI getting better, which means AI is not so good that they can remove the entire need for the industry. But they are still helpful and helpful in a predictable way that I can ask you to do something and it will do it reasonably well.\n\n[32:26] Vikas: Yeah.\n\n[32:32] Srikanth: You still need that expertise to go deep into the scheduler code and see where there are more possibilities for optimization. Same with database. Yeah, so you have database and you can maybe leverage the AI to go manage it for you, take backups, make sure it's synchronized, database repairs, stuff like that. Is it going to be as good in depth of it as a human is? It all boils down to that. If it is as good in depth, then maybe what you're saying is true. Basically that means that if I want to build UI, AI is better at going into the depth of UI than I would be. And in that statement, I would be in what time frame? If I spend one month, will I be better than AI? Will I spend one year? Is when I'll get better than AI. If the answer is the second, then AI is pretty good. if the quality of AI is so good, that it'll take me a year to catch up to doing that level of UI work or its quality of code is or its depth of thinking is as good as a one year old experience, right? Now make it 10 years. Is it as good as a 10 year experienced guy? If it is that good, then what you're saying is possible.\n\n[33:33] Vikas: What I also meant is, let's say, if you're hiring for somebody in your team, Now, somebody showed up and said, look, I'm a great Java backend or Kotlin developer, right? And you give a bunch of problems and they do it for you, right? Now, let's say I come along and I demonstrate, 80 % of that capability that the previous person, but I said, look, but I also do UX. I also do build my own CI CD pipelines. And in fact, I also do my performance tuning. So would you hire the first person or the second person?\n\n[34:05] Srikanth: I think I understand your question. Yeah, so the answer really depends on the industry. So in most service kind of industries, right, where you just have a problem you want to solve, you want to integrate system A and system B. Probably that 80 % guy is good enough. Okay. If you are hiring, even my team, if I want to hire, I'll probably be, I won't care about Kotlin, to be honest. I will care how good the person is as a software engineer. Language will going to be secondary. To that person, as it stands right now, the state of AI right now. I really don't see them as being able to just hand it off to you and just walking out of it. So I think today, as of today, at least in my team, if I were to hire, it's going to be expertise based or the ability to gain expertise. I need not find someone who is great at UI, but if he's a good software engineer, he or she is a good software engineer with the ability to dig deep and learn anything that I throw at that person. then I'll hire them. probably that person is going to become an expert in UI, UX, and maybe stick there for some time before moving to a different thing. I think that kind of stickiness will still be there because at least in my case, I like to push boundaries because wherever you are, you are going to have challenges which are really limiting that ecosystem. And you want to push that, right?\n\n[35:19] Vikas: That's an interesting take.\n\n[35:21] Srikanth: On the other hand, a service industry where all you want is to connect these two and that's it, right? It's perfectly fine too.\n\n[35:26] Vikas: Okay, now let's switching gears right. lot of our listeners are in college or in school who are planning to what advice would you give to say your younger self if let's say you were in college, second year of your uni, you know, studying say computer science and I'm sure you're looking at the industry and having a certain amount of anxiety on how this is all going to pan out the job prospect and so on, right? How do you think they should approach this knowing what you now know, especially around everything that has been going and the pace of improvements? How should one think about building their career in software?\n\n[36:10] Srikanth: The first thing I would say is that if you're anxious, that's good because you're supposed to be anxious because it is indeed changing in a direction that no one knows, no one predicts. It could be risky. It could have a real impact in your job prospects. Yes, so that anxiety is very much warranted and that anxiety is telling you to be careful and that's good. That's a good thing to be. What can you do? As it stands right now, I don't see software engineering going anywhere. At the same time, I don't see it staying exactly how it is or it was two years ago. At the very least, means you have to learn and at least somewhat be on top of where AI is. Although as I said, I know it's extremely hard, the rate at which it's changing, the new protocols that are coming up every few months and some of them good, some of them bad. It is very tricky. But if you generally have the inclination to be on top of it and be reasonably on top of it, at least the trends that are sticking, rag for example, is a good idea that is likely to stick than not to. Some of the prompt engineering or context engineering, I think just being on top of the field and using it in your projects or in your own learning is going to be good idea.\n\n[37:11] Vikas: Okay, but don't you think I mean, I know, if you look at the curriculums in most of the universities, unless you're taking an AI as a course as a specialization, if you look at computer science, for example, when they teach operating system or algorithms and so on, right? These are still being taught with books that were probably written in 80s and 90s and probably before, right? The law of curriculum hasn't changed a whole lot, right? Still majorly it's the same. Now, while it is understood that yes, the out is changing. When I did my engineering about, you know, 15, 20 years back, I think even then there was a gap between academia and the industry. I believe that gap has further widened. So do... Do you suggest that, well, obviously you made the suggestion to be on top of the trends, but it's not systemic, which basically means that it's dependent on individual's capability to go out and figure things out and stay on top, right? How do we ensure that this message reaches everybody, every student who's out there who's even aspiring to come to the industry, right?\n\n[38:29] Vikas: Do you think universities are doing enough? You know, or should, so my point being is, do you think there is a solve, right? One is going to every union and building that awareness, which is not possible, but do you think they should contribute to open source, build better portfolio? Or is there a way where they could level up faster? Are there any specific suggestions would you give, right? For example, everybody should contribute in open source or should build personal software. I see a lot of flex on Twitter when people show their GitHub contribution last year to show how many greens they have got, right? So is that sort of a thing or going and talking about what they build on Twitter, building a brand for themselves, right? Is there a specific suggestion you'd like to give?\n\n[39:19] Srikanth: I think these examples that are quoting is not going to scale well, right? mean, yeah, let's say everyone tries to have the most number of following. Only some of them are going to end up having a lot of following, right? Doing something, building your own project, having something to showcase in GitHub, they're all good suggestions, I think. And you are right. It's not a systemic thing. It'll take some time for... universities to adapt, curriculum to adapt and as we stand today, we don't even know where it's going. So how do you want to adapt? We don't know. You like, like this is a problem today, right? At every school level, we're writing essays, writing code, everything, writing anything. In fact, how do you make sure that the person wrote it not AI wrote it? Right. It's a, it's a problem that everyone is forced to handle, like universities, schools, professors, everyone is forced to handle all of it right now. It's going to take some struggle to figure out how to do. Maybe you'll go back to paper and pencil, going back to class and writing it in front instead of online and all that. Being generally good involves you face a problem and you try to find a solution. And in the path to find a solution, either code it by yourself, coding being a specific way to solve a problem, but you code yourself or use AI to code it. This general attitude towards solving problem is what you need overall.\n\n[40:26] Vikas: Okay, so thanks. And I think I'll ask you one last question and probably we'll wind this up. Now, this question is about specialization, right? Specialization. A lot of people, a of software developers today when they enter industry, they're given this advice on if they should pick up a specific specialization, right? For instance, retail is a specialization. or a certain industry specialization or a certain area in retail or banking or fintech or robotics or insurance or underwriting and so on. Now, do you think that, I mean, complimenting the, assuming they are software developers, right? the domain expertise going to be a mode? Should they invest more into learning the nuances of domain given they have some free time now or the efficiencies gained by AI doing some of their work. Should they invest in building that domain skill or do you think they should focus on the other area of say software development, right? Like I said, DevOps, SRE, ML Ops, data engineering, or should they spend that time in building domain expertise?\n\n[41:37] Srikanth: Firstly, you're assuming that kids will have more time because there is some amount of automation. They may not have more time. Academy will catch up. The kinds of assignments will change. They may not have more time. Is more expertise or domain level expertise or things like... So domain specific knowledge, I don't know if it's a good idea to build it upfront. It's not clear to me. Like in finite.\n\n[41:58] Vikas: not in the... Yeah, but when they're in first or first few years of the job.\n\n[42:03] Srikanth: think as things are evolving, think you will naturally be pushed to if you have more time because you are not spending all the time writing low level code, assuming of course that AI is good enough that you don't have to debug it, assuming it works, assuming that AI is good enough that you don't have to deal with all those nitty gritties, yeah, you will be free to actually solve larger problem. Absolutely. That's what will happen.\n\n[42:25] Vikas: Right. So do you think and again, is another I sorry, I lied. I'll ask you another question. Let's let's take a best case scenario. Right. Let's assume that in a year's time or maybe two, the the AI or the LMS gets so good that they come over not just deficiencies of today, but build new capabilities in terms of depth, speed, and so on, right? Do you think every software, and assuming if such a world exists, where LLMs have become so good, better than the 10x engineers of, let's say, the top of our industry, right? Do you think every software which has probably existed for 30 last 30 40 years will have to be rewritten with a new intelligence, right? New sort of retail software, new sort of search, new sort of, you know, airline software, new sort of traffic management, new sort of finance, trading, cons, mining, you know, like because some of these softwares have stayed and, you know, been created or but would we have a need to revisit every single software?\n\n[43:35] Srikanth: I think the industry forces will automatically push you there. Right? So you have a smarter way to do something. There will be someone who will implement it. The others will have to catch up. Take traffic for example, smart traffic management. Like today, at least in US, there are these strips under the road to detect if there's a car. But it cannot tell how many layers of car are there, like how far it goes. I think Bangalore has a better system. I think it uses cameras to determine the depth of traffic. And so, yeah, of course, I mean, it's going to happen. It's bound to happen. What we have now is a new kind of tool. So we have this tool called software. We have a tool called internet. And it took a long time. It proliferated everywhere. We have this tool called phone and very cheap internet and accessibility everywhere. It just proliferated everywhere. in fact, a lot of websites, a lot of services are now more optimized towards an app than they are. actual desktop websites, right? This is the natural evolution. Now you have new tool which is code level intelligence. mean, intelligence that can be modeled in a device, right? Of course, it's going to go everywhere like a refrigerator which can tell you that it's out of milk. Yeah, a stove which has, you know, which tells that it's on and you know, your smart home devices know that you have left the house and the stove says it's on. Somehow there is a system which says, hey, there is no one at home, but the stove is on.\n\n[44:38] Vikas: or something is going to go stale.\n\n[44:52] Srikanth: I mean, we are really at a place where our imagination is the limit at this point.\n\n[44:57] Vikas: That makes me believe that while the number of software developers may reduce, but the need for software is only going to explode, right? Because knowing\n\n[45:09] Srikanth: The real threat, if I want to call that word, is that a machine is now doing what it could never do before. We have had machines since the beginning of humans, right? We learned to make stones and spears out of stones and wood and so we have been building machines forever. We built wheel and bullock carts and we've been building. But till now there is one domain of expertise that machines could not reach which was The human ingenuity, creativity, ability to solve problems using intellect is something that machines have not been able to do. And so whatever has happened till now is that every time, quite literally, have... there was a... at my place, I saw the people had come to cut a tree. There were two men and a huge machine. That machine was able to hold the tree. It had a saw to cut the tree. and then lift the tree and then put it in its own wood chipper which was integrated to that machine and chip it down into small small pieces and two people and a machine that's all it took. Okay. So what so a hundred two hundred years ago how would maybe it will require 10 people to do that. So what are those other remaining eight people doing. They are doing some sort of mentally intellectually difficult work that machines could not do till now. The problem with AI as it stands maybe it's not a problem but where it's heading towards is It's able to do this one thing which is still unique to humans, which is the ability to think. And so we don't really know where it's going to go. even if you're right, we can automate these things, our imagination is the limit, right? But for everything we imagine, we employ AI to do it because it's cheap for us. Then we never really know what humans are going to do. And there's a whole stream of thought about universal basic income and that there's a possibility, think Musk suggested that a few weeks ago that in just a few, he suggested that we probably shouldn't be worrying about saving up for retirement because just in a few decades, there's going to be enough wealth that you don't have to worry about saving money for retirement. You'll be taken care of. But that's not how it's been. Capitalism has not spread that wealth to everyone, right? It has still always concentrated higher up. That's how it is. So we really don't know where it's going.\n\n[47:16] Vikas: Okay, yeah, thanks. That's some interesting take on what's happening. I think we've pretty long episode and I think we should wind it as much as I enjoy this conversation. think, you know, we should keep keep this to do it for today. And thank you, Srikanth. It has been a wonderful discussion. Thank you for coming to the pod. And I assure you that we will do, you know, maybe another one of this in a year's time and see how You know things evolved. how some stuff that we spoken about today holds versus change, right? And to our listeners, I hope you found this discussion engaging and if you have your thought, feel free to comment right to us. We also have a new website which you feel free to engage and discuss. And thank you for being a great listener and as always, everything that we spoke here on this episode represents our own opinion and does not represent that of our employees and organizations that we work. Thank you and thank you Srikant. Thank you. Yeah, bye.\n\n[48:17] Srikanth: Thanks for having me.", "generatedAt": "2026-01-18T19:16:00Z" }