How to Get Hired for Microsoft AI Roles Without Applying in 2026 using Recruiter Inbound Playbook - w/ Shrey

I did not get a call back because I did not apply for a job at Microsoft. It was inbound. Senior AI Engineer at Microsoft, he never applied for that role. Meta came calling too and he thinks the way your job hunting stopped being relevant a while ago. I just don't see a point in applying for the job statistically because of AI. I think AI has killed traditional job applications in my opinion. So by the end of this episode, I promise you'll know what he did instead and you can then steal that starting with the interview that had no lead code in it. I did not have the standard lead code interviews. They are actually much harder in my opinion. If you give them what they want, you won't have to look for a job because they're going to come look for you. It's not because AI will take over people's job. It's because the roles themselves will merge. Please join me in welcoming Shrey Shah to the Ready Set Do podcast. Without any further ado, let's jump in. Where I want to start off, especially because it feels fresh and new, is your role at Microsoft. So I know you somewhat recently-ish, right? I think it's been about two or three months. Started as a senior software engineer with the AI team. Can you talk us through kind of lead up to the days where you were first, you know, getting that breakthrough in terms of when you first heard back? Because as you know, I'm sure you will attest to this. It right now is probably the most brutal job market of all time, literally like ever. So talk us through kind of, if you would please, your experience of getting that first call back. Let's start there, right? What did you do to get your foot in the door? Yeah. So I did not get a call back because I did not apply for a job at Microsoft. It was inbound. So I was speaking at a conference about my AI coding workflows and building agents and how to, you know, bring in security and testing and AI and all of that. And all of the executives at Microsoft listened to the talk and they found it interesting. So that's how we started talking. And then I felt like the work that I was going to be doing at Microsoft would be really interesting. So I think then we started the interview process. I went through a couple of interview rounds and then, yeah, then I just joined. So it was that simple. I have not applied for a job, in fact, in the last, I guess, seven years. Wow. I mean, that is a flex if there was ever one. So that is awesome. That is so cool. And I know, so as I was sharing earlier of the various parts. So this was obviously one of them, but you're being ambassador for, I don't know what's the right way to say. Do you say V0? How do you say it? I guess. V0, right? Yeah. Because I've heard people like some people say V not. So I never know what to say. It's like, I don't know what's like the cool thing to say, but with V0 and I think cursor as well. So we'll be going there. Right. That is, you know, we will be setting that up in just a little bit. But as far as the interview experience itself, was it pretty straightforward from what one would expect, just like lead code and such? Were there any elements there that you can talk through for others that would be going through that process? Yeah. Yeah, for sure. So my interview was a little bit different because of the kind of role that I was taking on. It was more of like a role that would, you know, bring the cultural change at Microsoft. And also just like, you know, how do we do things differently? That's been my guru mantra forever. How do I do things differently? How do I do things more efficiently? So my interview process was that I think I had like four rounds of interviews. Two of them were technical, but not in the traditional sense. So I did not have the standard lead code style interviews. They're actually much harder in my opinion, because they tested the different set of skills that I would require to work for what's coming, not for something that's that will be outdated. Right. Because I personally don't feel the need to do lead code anymore. I'm not saying you should not learn problem solving. I think that's the core of everything, because if you can solve problems, I think it doesn't matter whether you know lead code or not. You're not going to be able to do your job. Right. So my interview process, like if I talk about the technical ground, I think I was maybe about like, I was given a challenge, a really hard issue to solve. That would have taken a normal day or two to work on. But I did, I was not given a specific time set to really work on it. So normally in lead code interviews, like you would have a set of like, you know, an hour or so to work on the problem. I did not have the time. I still had to solve the problem while talking to them. So it was the way I looked at it was like, they were kind of testing my context switching skills. How do you manage sub agents, right? When you're managing sub agents, you're talking to multiple different agents at the same time. It was not agents, but it was also the agents and the humans. Because I was also talking to my interviewer. At the same time, I was also solving the problems at the same time. Yeah. Right. So I think probably that's what they're looking for. To see like how I work differently. So it was more about like, how do I solve problems? And what is, what does my workflow look like? How do I, and whether I can actually deliver quality code or not. Right. It's not about like, white coding or producing stuff, but like creating workflows that actually gives you reliable results without creating. I think it was more about like, how do I develop agents and like, how do I develop agents using agents? Right. I think, so I think, I mean, that's what I've been doing since last three years, building agents with agents. So yeah. And I think that's been working really, really good. Yeah. That's very interesting. So you mentioned that this role was like a different type of role, right? It wasn't your traditional SD role that, that you have at big tech companies. So can you share more about what made this different? Like, are you almost like an embedded culture changer when it comes to writing like agentic code? And like, just tell us more about that piece. Yeah. I mean, the role is not too different from a traditional software engineer. It's just that I think, the value that I bring is not in the traditional software, but it's in like, how you do things differently. Right. So I think, I'm still building agents at Microsoft. Right. So traditional software engineer, it's not traditional software engineer, but at the end of the day, I'm still building software. Right. Right. To give you an example, like I read a post of, I think a lot of people did post about using HTML. Yep. I saw that too. Yeah. But if you ask how many people have actually tried it, very few people would have actually tried in practice, like for days. When you do that, you understand the different pros and cons of different approaches. And then basically this is just one simple, very, very simple example. The point is like trying something new and then figuring out what works and how it's mapped to the future. That brings, that I think I would say like that makes the software forwards compatible. That's what I call it. Yeah. I've always been talking about backwards compatible software. I always think about the forwards compatible software now because the models change every three to four months. And I always am thinking, Hey, okay, I'm building this agents now. Will this be obsolete in like three to four months? And how do I make sure that it doesn't go obsolete in three to four months? In order to do so, you need to have a different mindset. It's not the skillset that matters here anymore. It's the mindset that matters. Right. I think that's because all the rules are merging. Like I'm doing, sometimes I'm doing a bit of research. Sometimes I'm just hands down coding. Sometimes I am, you know, acting like a product manager to figure out what really should I ship. The shipping is easy. What to ship is hard. Right. So I'm just deciding that. So all those rules are merging. I think it's just the mindset shift that really matters. And I think that's been my main mantra on like how I do things or how I drive things. I think that is the value that I'm bringing at Microsoft. It is not in the traditional, you know. Yeah, that's very interesting. Would you say that this is also developing at other big tech companies outside of big tech? Do you think like the role of an SDE as it exists today is just changing right now in front of us? That's almost like what it sounds like from what you're describing. Yeah. Yeah. I think I also had someone from Meta reached out to me for a similar thing as well. And they were also kind of asking me to do similar things at Meta as well. So I think it's at all big techs, they need someone like that who will kind of bring the fresh perspective to make things, to keep everyone informed. So I think the problem is that all the big tech companies, it is really hard to change someone's behavior. And not just at big tech, in software in general, like changing someone's behavior is extremely difficult compared to like asking them to adapt any technology. And I think, and this is especially hard because it's changing every three to four months. For example, like Jeff Dean from Google said, like, and I think in a podcast with Dwarthas Patel, he said that new models are coming. That are going to like, what is 10,000 tokens per second? Right. So my mind goes like, Hey, if that's coming, what do I need to do now? So that I don't know, rebuild my software after four to five months. Right. And the reason I know that is because I'm always in the weeds, because I truly care about AI. If you're just a software engineer or an AI engineer building agents, you're not going to have that perspective. Because you know, we always on Twitter, you need to always kind of, I think social media is super weird now, but I think it is because of Twitter that I know what I know. Right. And that helps me build that perspective. Right. You go and read about it. And then you can, if you keep doing it every single day, you're going to start to connect all the dots and put them together. And then you understand what's going to happen three months for now or six months for now. I think I still can fully comprehend what's going to happen when more of that past will come. But I'm trying to prepare myself for trying to think differently on how I do things now because even that is changing. Yeah, it's so interesting. So I guess what I'm sure a lot of the audience right now is wondering about is how do you become like such a, you know, magnet for recruiters from big tech companies? Right. Like it sounds like you've just been fending them away, keeping them at arm's length. And as they keep, you know, coming to you, you had reach outs from obviously Microsoft. You mentioned that you joined, you mentioned Meta. What is, what are some things that folks can start doing that will allow themselves to start featuring on top of recruiter reach outs? Because I think that literally sounds like a cheat code and it feels just impossible to do in this day and age. I think it's really about supply and demand. It's nothing more than that. I'll give you an example. If you're looking for, I don't know, let's say, Apple's magic mouse, right? I come to your doorstep. There is probably very few magic mouses available in the market and you want to get one. You're probably going to pay a high price and still get it. Right. But let's say like you own a Windows machine and you can't really use this magic mouse. It doesn't matter whether someone is giving it to you for free or not. You don't want it because it's no use to, right? So that's how I've been driving my career. Like I, I've always been thinking, Hey, what does the market need right now? Or what will it need in future? And then I try to learn those skills myself rather than waiting for my employer to give me a chance to learn those skills. So when the right time comes, I know that I will not have to look for a job. Like back in 2019, I was learning about Web3 because at that time, since I was still at testing, but at the same time, like I just wanted to be in the area where the demand is. Right. So I was learning Web3. I was also learning coding using AI. So in July, I would say, I don't know if I can call it AI or not, but I was using GPT-2 in a tab nine back then to auto-complete my code. And that was very, very early. I started using GPT-3 back in 2020. I understood one thing that this is going to be the future back in 2020. It was June, 2020 when I think it was around June or July, when I first got access to GitHub Copilot in VS Code. And I think I went all in. I was like, this is the future. Like seniors around me asked not to do it. They were like, don't do this. It's going to hurt you. But I just could not unsee. I kind of am always in the weeds and I know what's, I try to know what will be the future thing two years or three years from now. And then I start to learn those skills to prepare myself for the market in future. Right. And that has really helped. So back in 2020, end of 2022, 2023, I started learning about RAG and PR and a lot of that. Like change. I think it was very, very early. I was using cursor in, I think, May, 2024, May 2023, not 2024, May 2023. I was using cursor in May 2023, very, very early, just when they launched. Right. So because I was doing all those things, they all compounded. And the reason I was doing those was because I understood one thing that, hey, like I've been using GPT-3 for three years and now it's inside of a chatbot. I need to learn to build applications with this, because that's going to be the future. At that time, there were no jobs in AI. In this sense, like building agents and stuff. There was nothing, almost none. Right. And even in 2024, I think like there were no jobs about building agents. Like there were very few jobs because people were still skeptical about, you know. Yeah. And they didn't even work all that well at the time. So people weren't sure. That is true. But I mean, it's like if you started programming computers in 1969, and if you are still relevant in 2026, you will have a perspective. You will bring the perspective that no one else has because they have not gone through the journey. They have not had time to absorb the changes. Right. If you want to stand out in this market, you have to understand what the companies want, what the industries want. If you give them what they want, you won't have to look for a job because they're going to come look for you. Yeah. Right. Versus if you have a fixed mindset that says like, Hey, I know react and I know springboard. I'm going to go to the market with that because that's what I know. Because everyone else is the same there. Right. Like there's no differentiation. I guess my follow up to what you're saying, Shrey is, I think a lot of people maybe now have realized, right, that you need to, you know, be ahead of the curve, figure out these trends as a drop. Maybe they weren't seeing them as long ago as you were, right. As in like, it's like obvious now that it's here. So, you know, just go all in on this. I think what I'm getting at is, I don't think just doing that automatically gets recruiters in your inbox. You know, I also feel like there's like a big element of, I think maybe you can call it positioning. I don't know what's the right term. So can you, and I think that is actually the part that trips most people up, right? Because I'm sure from the population that's listening, a lot of them are AI forward. They can develop agents. They can, you know, do all of these really cool things. Maybe not to the really great extent that, you know, someone like yourself can, but maybe like 70% of the way there. They're not getting any inbound recruiter DMs. So how do they change that? Yeah, I think you're right. Like it's about, so I think I don't believe in resumes anymore. Um, so it doesn't matter what you put on your resume. And the reason is almost everyone is now a lot of people are using bot. Everyone is at this point. I have to assume at, you know, at this point. Yeah. So they're like telling the CVs about the job, sometimes telling the resumes, everything is being tailored to a job. Right? So what differentiation can you really bring in from your resume? Yeah. It doesn't matter. So, I mean, I saw this long time back and that's why I started like posting on LinkedIn about a year, two years ago. Like I was like, it doesn't matter how good my resume is if I'm not visible, because I think it's like, what can you really do? Well, if there are thousands of jobs, thousands of, sorry. I meant to say like, if there are thousands of applicants applying for a job and you are just depending on the ATS to scan your resume and pick one out of those thousand. So really it just becomes a numbers game and it's really hard to compete in that. Right? So I don't see a point in applying for the jobs traditionally because of AI. I think AI has killed traditional job applications in my opinion. Um, um, so the only way in my opinion is to go do a lot of networking. Right? Um, that's one way to do it. Um, second way to do it is obviously like being more visible on LinkedIn, on X. Right? Um, visibility is everything. Right? Um, I think that earlier I was getting a lot of calls because of my skill sets. Right? Uh, because that was rare. Uh, these days, however, I think I'm getting the calls not because of my skill set, but they're not getting the same. Um, and also mainly because of the brand that I'm trying to create around myself. The networking, cause I meet a lot of people, uh, from a lot of companies and I have made really good connections with, uh, uh, leaders at different companies, different, uh, even like, you know, uh, different big tech as well. Um, I engage with them on LinkedIn as well for that. Like they kind of know, Hey, this person exists and this is their point of view on a topic. Right? Um, so when they want to hire someone, they know that there's someone else that they can reach out to. Maybe that's a good. Yeah. Um, I never needed to do that specific thing cause I never need to optimize anything for, uh, jobs cause I was never looking for a job. Um, but, uh, if I had to, right, I think that's the approach I would take. Uh, uh, I would figure out and I think I would just like try to organically connect with people. When you try to just look for jobs, I think you come as desperate and that actually hurts me more and it helps. And I always see it as a longterm game. A lot of people will only apply for jobs when, or only start networking and all these things when they're looking for a job. I feel like it has to be a consistent effort. I think it pays off eventually. Um, I mean, I feel like there is no other way to look, to stand out in this world because everyone, everyone's resume looks the same. Everyone's scene looks the same. Even to some extent, the room also looks the same because it's all been written by elements. Right? Um, so how do you differentiate yourself? How do you prove that what you say is true? How do you prove before someone, uh, calls you, they have to know that it's worth their time to talk to you. Right? Unless like you're constantly posting on LinkedIn or on X, showing what you've done, showing the progress you've made. I think that is the only thing in my opinion that will make you stand out, uh, in this, uh, new world where everything can be fake and everything just looks exactly the same because everyone is using the same tools. Right. And so in terms of frameworks, right? So say somebody has realized all these things that like, okay, this is something I need to do and where they're stuck. And I think this is where most people get stuck is what to post, right? So I know you mentioned that a little bit. You touched on that. First of all, go build helpful slash cool things. That's step one. Step two post about said cool things. So is that, would you, would you agree roughly that, you know, just find something that interests you, solves a specific problem. It's not like a Kaggle data set that is just anybody can tell that this was done purely to serve as a bullet on your resume under your project section. Is that roughly the mindset to think through with this in terms of what to post? Yeah, I think I'm glad that you asked this question. So I think you asked two questions, like first, what to build. I strongly believe that you should only build the things that you care about. That's all. Yep. I agree with that. Yeah. Right. Because you learn the best when you do that. The reason is that you are the product manager. When you do that, you know what product you want to build. You know what goes inside of it. But since like, if someone asked you to build, I don't know, some random project on the internet, you're probably going to build it, but you're not going to have that motivation to go and do it. If it's a real problem that you're trying to solve for your servers or some of your company members, you're going to go deep into it and keep going at it until you're successful. I think that matters a lot. But that's the first thing. And the second question you... Yeah. Yeah. Yeah. The second question was, what should you post? Right? Like in terms of outside of projects, what else are we doing or should it just be projects? Yeah. So you can either like post about, you know, your learning journey that, Hey, like I'm learning about building agents. Today I read about LangChain. This is how you can do xxx on LinkedIn. This is how we can do XYZ on LangRap. So you can post what you learn on LinkedIn and that's helpful. That's how you can get started. And then if you want to do it consistently, and that's what I was saying, like, I think my analogy is that you should always post more than you consume. So create more than you consume. But at the end of the day, you still need to have ideas, right? Because you can like, it's really hard to come up with like actual ideas to post. The way I go over is like, Hey, like I'm following, you know, AI on Twitter. I read, let's say like 30 to 40 posts in a day. Out of those 30 to 40 posts, I would have an opinion on something. If I don't, then I have a problem. Right. But I would have my own point of views on those topics. Why not just pick that, do more research on that and create a post with your own point of view. Because I think people wants to, people really want to hear humans opinions, not the AI agents opinions. Like AI can create something, but it doesn't really have a point of view on anything. So I think at the end of the day, I think it's you, it's your opinion that matters. So I think if you, if you want to post consistently, I think like just having your own point of views on the content you consume already. I think that brings the most value. And that's how we can keep creating. I think nowadays it's even more easier to just create content using the front of me and stuff. So I have like my Hermes agent that sends me really cool posts or demos and videos on my telegram. And then I talk to her and say, Hey, this is pretty cool. I think let's keep this on our watch list. And then next time when something cool happens in that area, it sends that to me. And then if I find it cool, I create the content on it right from my tenant. It's much more easier to like stay updated and also create content. Yeah. And you can even post about how you built your Hermes agent, right? That's like an entire content folder. But the only reason you got to do that was because you were interested enough to solve a problem that you were facing that was specific to you that you then went out and solved. And then, yeah, you can, you get the point, right? You just apply this analogy to everything. I want to bring us almost full circle and go back to the talk you were at where you were first accosted by the Microsoft person that ended up, you know, being the pathway to you getting hired there. How does one do that? How do I or folks in my audience start getting invited to do talks on these things? I think it's all just compounding, right? I think years of work. It's not like something that happened all of a sudden, right? I think it was like me being a visionary back in 2020, saying that, Hey, I want to use AI for coding. It doesn't matter what anyone says, I'm still going to do it. I know it might hurt me in the short term, but in the long term, it's going to be more helpful. So I kept at it. I think the reason I was invited was first of all, because I did a lot of local workshops and meetups above cursor and posted about those. Some of them also posted on YouTube, right? So people are the audience already had some idea of what kind of workshops and how I talk. So that's how they approached me. And then they're like given a talk. And then I ended up giving the talk conference, multiple conferences actually. So, yeah, I think I would just say like either be the best at what you're doing. Like it doesn't matter what you're doing. Just try to give it a hundred percent. Even if you're still just a react dev or Angular dev, I think it's great career as long as you know what you're doing and you're bringing AI into the mix. It doesn't matter what you're doing. If you're the best at it, you're going to get all the opportunities. You just have to be with you, right? I think just turning out and being the best in your career or in your skill set, it's not as valuable anymore. Interesting. It is only valuable for your job. But I think if you really want to get more value out of basically your skill set, you have to be visible. But that's it. I think there is no other option. Yeah, no, I think you make a very timely point, right? Because as of recording this, we have just learned about the meta layoffs. Almost 8000 people laid off and yeah, it's so unfortunate and it's so sad and heartbreaking. But it just really feels like this is going to be the norm. And I feel like for anybody that's in, you know, employed in tech, especially in corporate America, I really like that you're making this point so strongly that it is not even an option anymore to just be good at your job and call it right. Like you almost have to be, you're almost forced to be, you know, that visible presence in terms of what you're doing, what you're building in order to almost ensure yourself against events like these that there's, you know, there's really no protection around that. While researching for this episode, I was going through a lot of your writing on LinkedIn and some and a theme that keeps coming up with a lot of your writing is that a lot of AI engineers are still using vibe coding tools incorrectly, right? Like there's a lot of pitfalls that you see people make and obviously in your own experience as well, maybe you've learned from your own failures. So maybe can you help paint the picture in terms of maybe two or three of the biggest pitfalls that you see, even like intermediate expert level developers that are they still make these errors with AI tools or vibe coding tools? Yeah. I'll talk about the general principle because I like a lot of AI tools that for sure. So I think the big, yeah. Yeah. So in my opinion, the biggest bang for buck would actually be understanding the differences between a skill, a sub agent, a hook, a rule. Rules are also like three or four different types and like project based rules. And then even put those like different patterns, like glob pattern rules or, you know, always up to rules and then module level rules. A lot of devs, when I talk to them, they don't understand the differences between those and like when to use what. If I ask someone, hey, should you use a skill here or a sub agent? You'll almost never have an option because they cannot differentiate because a lot of people, it's really hard to understand as well because there's a very big time between all of these things, right? Because at the end of the day, everything goes to the front. It's just the markdown file, right? That's what most people think. In my opinion, that is the biggest bang for buck because you're just understanding how different markdown files work and what goes into what. If I go and explain all the differences. Yeah, no, that I would, I agree. That's outside the scope. Although I can't help but ask, and this is just for my own understanding. Yeah, like now that you say it, what is the difference between a skill and a sub agent? And it's funny because I thought I understood both these things pretty well, but apparently I don't because I actually can't think of the difference. So can you just share this one piece? Because I'll just be thinking about it nonstop unless you do. Yeah. So I think the skill, in my opinion right now, is an agent. He is the main agent. That's how I look at it, right? Because if you look at, at least in terms of coding, right? If you look at like Copilot, Cursor, Cloud Code, all of them can pretty much do everything. Those are all general propositions, right? And you to understand like what, like earlier, back in 24, when I was doing agents, I used to enclose all of our business logic in tools. And those tools were very, very rigid. It's all of our business logic and problems would live inside that sort of tools. And those had to be programmatically bound to the agent. Right. To solve that problem, we brought on MCP servers. Because if you have two agents and both of them had to share the tools, you are now duplicating the code. And it was also like, you could not add more tools easily. That's the reason why we have MCP servers. If you ask someone, why do we have MCP servers? They just know that we have MCP servers. They don't know why. Understand the why behind it, right? That's what helps you understand why do you have skills, some agents, MCP and all of that. So that's why we have MCP servers. But then the MCP servers were like, if you added like 50 MCP servers, that added a lot of load. And probably a lot of people already know that, right? So to solve that problem, we have a skill. I see. Now, skill is not just a background file. Skill can be, if you really want to be complex, it can be... Yeah, it can be multiple files. It can have routing. It can have a directory to navigate where to go for what case, what to do. Yeah, yeah, exactly. It doesn't have to be one file, correct? Right, yeah. So skill is basically how you do something, right? As a whole. For example, I am a software engineer, right? So I can write a skill about myself to replace myself, right? Hey, this is how Shrey does things. This is my skill set, right? This is how I do things. Step one, step two, step three. Use this code to execute this and that. But what a lot of people don't do when they build skills is that they don't know that some skills are user invocable. Some skills are agent invocable only. You can also form the context in the skill. So the skill can run its own context, not polydentimed context. Right? There are probably like 20, 30 different front meter options depending on the coding agent you are using. So knowing those, you can actually control the output of the skills. Right? But skill is not like as part of my job, right? As a dev. Sometimes I would be doing a lot of research, right? So think about it. When I'm doing the research or when you're doing the research for this episode, for example, you have a skill set of building the podcast, right? But you executed a sub-agent that was just doing the research. You did the research. You came back to the set of points that were necessary to continue building this podcast episode. You did not need to keep the log of everything inside the podcast episode. You only came back with the research and that's how you build the episode. Right? That's the difference between a sub-agent. Sub-agent is basically how you do a piece of work. And then you get a piece of work. Either in a dedicated context or in a shared context. Right? And the sub-agents can run in parallel with skills you don't have as much flexibility. So when you need to run things in parallel to make things faster, and sub-agents now can also spin up their own sub-agents, right? If you want a tree of execution, if it's a really complex task, right? Doing that really will help you save the context. The problem is... Exactly. I was just going to say because you can even have a skill that has, you know, guardrails around steps one, two, and four. And on step three, you can let it go loose and do research, like figure out, be autonomous, basically. So you're right. Like these things are super complicated. And I think the biggest takeaway, at least for me, is figure out what you want it to spit out as output. And then do that, like almost reverse engineer to get to that very specific output that you want. Right? And it doesn't matter what you use, as long as you reliably get to your output success criteria. So yeah, really appreciate that deep dive. Yeah, yeah. Go ahead. Yeah, yeah, yeah. One more thing. Podcast. Podcast. So the main thing that the sub-agent is doing is actually parallelism and context isolation. Because if you have like 10 different skills working together, and if you don't have contact isolation, everything just happens in step one, step two, step three, you will hit your tone of tokens contacts of one million. Yeah. Yeah. Really fast. So if you're only coming up with probably like tone of tokens of research, it's not going to cost you more. It's also like about the cost and also about the speed and the quality of the output. Definitely. No, yeah. The great call out, right? And especially with how expensive at least Claude Code is getting and I'm sure the others will follow very soon. Yeah, that is not something that's optional anymore. Yeah. So I guess final segment here before I let you go. Because you've throughout your career, right? You've had this uncanny knack of predicting what's going to be the winning horse like two or three years from now. Now, where do you stand in terms of all of the AI doomsaying that happens online? You know, like you can call it rage baiting, whatever term you want to use. Where do you think we will be? Like what is your official on the record prediction for where we will be? Like maybe in 20, we'll say 2028, maybe. I think based on what I know now, I think it's really hard to predict. I wish I had a really golden answer because AI is so unpredictable. And I believe that our human minds are not capable of estimating or understanding this exponential growth that comes with it. But I think what I'm thinking is going to happen is that it is likely going to be true that a lot of roles are going to merge. And the job loss and all of that is going to come is mainly going to be because of that. It is not because AI will take over. In my opinion, it's not because AI will take over people's job. It is because the roles themselves will merge. Everyone is doing more. I think because of AI, I'm doing a lot more than I was doing before. I'm working a lot more than I was doing before, right? And the main reason is because I'm doing a lot more. I'm not just, you know, developing now. I'm doing a lot more now, right? Everyone is doing a lot more. PMs are, for example, like PMs at Google are now coding, right? I think that's what Sundar Pichai said in my interview. I think all the rules are merging. If all the rules are merging, some people will have to go, right? Some new jobs will be created, but the number of jobs that will be created will not be equivalent to the jobs that will be lost. So, but, you know, we'll see how the AI costs go up as well. Because it's absolutely why some people are spending millions of dollars a month. Yeah. Steinberger. Yeah, yeah, yeah. I saw his spend recently. I even saw this funny meme that somebody posted that they've hired a fresh grad out of college to manually review code. Because having Claude do that was more expensive, actually, than, you know, the salary of an entry-level grad. So it's so funny that we've come full circle somehow already, which is like, I don't know. I didn't fact check this. I don't know if this was somebody just joking. But it was like, it's believable to me that this might even start happening. But yeah, this was... Go ahead. If you, like, look at VHC, V4, and some other Chinese models, those are just as good as like OV4.6. A fraction of cost, right? So I hope that probably by 2.0, either, like, we are able to run our own models locally, so that we don't have to depend on, you know, this model for orders. Or we'll just go back to, like, you know, hiring people in India and stuff, because it is cheaper to hire talent in India compared to, like, doing the models here. If there are more expensive. And I think it's like, as the models become more capable, they're going to do more. Because if you look at the token cost, the overall cost per intelligence or cost per token, that is going down, in fact. Right, right. Real by year. But the cost is going up because we are just relying on agents to do more and more. Yeah. Yeah, no, it's a good point. Yeah, definitely something to keep in mind as we, you know, go and find, build this plane as we are flying. So, this has been so cool. Yeah. Sorry, go ahead. Yeah, yeah, yeah. I think it's coming two months. It's more, sorry, not two months. It's coming two years or three years. It's more going to be about knowledge of automation. I think that's what I see. So back in, I saw, like, you know, the next phase, the next wave will be about building agents. I'm seeing, I'm feeling right now that in the next two to three years, it's either going to be how do we build AI with quantum computers. I think Google or IBM, I forgot who, but someone is already trying to train lousin with models. Yeah, I think it's Google. Yeah, yeah, yeah. I think that was Google. Yeah. Yeah. So, it's either going to be that or it's either going to be, like, how do we automate most of the knowledge? Interesting. Yeah. I can see that. Yeah. And, I mean, it is already happening to a good extent. So, yeah. Right. Right. Right. What I'm saying is, in the two to three years, you will see, I'm seeing that probably there will be experts or whatever the project engineers whose entire job would be, to shadow someone for a month or so, understand how they work, what they do, and what are the things that live in their head and convert those into machine-readable documents or parameter skills or whatever you want to call it. But I think that's what's going to happen at scale in probably three years. Got it. Interesting. Wow. Yeah. I mean, that's wild. I don't even know what to make of that yet. But, yeah, Shrey, this has been so cool, man. I feel like I often think about all of this just noise that goes out around, you know, getting hired and such. And it is a concept that is starting to make sense to me now because I talk to people like yourself. But I guess, yeah, I think what I'm trying to get at is people haven't really realized when the ground under their feet completely shifted and like they just got rug pulled from being in this world where you like easy apply on LinkedIn and, you know, you get an interview and then you get a job. It doesn't happen anymore. And I just worry. And part of my mission with episodes like this is to help these people get unstuck in terms of realizing that the old model is broken forever. Like it's not coming back. Like it's over. Pivot fast. So I really appreciate, you know, you being so candid with your experiences, what has worked for you, what allowed you to get unstuck at a time when you were also figuring these things out. So, yeah, really appreciate you taking the time. This has been so great. And hopefully we can have you back for around two in a couple of years to see if your predictions turned out right. Yeah. Yeah, for sure. And it was really fun taking this conversation and yeah, I just love being here.