Why the W-2 Is Ending, How AI Agents Are Replacing Jobs, and What Operators Should Build Right Now | Eshwar Prasad | Part 3

This episode isn't about how to use AI tools. It's about what happens structurally - to work, money, and business models - when AI agents do most of the work.

In Part 3 of this conversation, Eshwar - a solo founder building Exemplar, an AI-powered clinical research platform - breaks down the economic and operational shifts he's watching in real time: the end of W-2 employment, the rise of agent-to-agent economies, why microtransactions require entirely new financial infrastructure, and why understanding protocols matters more than picking the right company to invest in.

This was not a technology conversation. It was a constraint conversation.

We cover:

• Why traditional employment disappears in the next 10 years - and what that forces operators to do now
• How AI agents create an economy that credit card rails structurally cannot support
• The TCP/IP analogy: why investing in protocols (not companies) is the rare window that's open right now
• What Eshwar built in two weeks that would have taken 10 people a year
• Why "English is the biggest programming language" is a genuine operating insight, not a soundbite
• How to learn AI - by using AI - and what lowering your guard actually means

If you're a founder, operator, or investor trying to build before the shift finishes, this conversation will compress your thinking.

This isn't a highlight reel. It's a practical breakdown of how one operator is reading the structural change - under pressure, not in hindsight.

Chief Milestones is a video podcast featuring honest conversations with founders, parents, and investors about building real businesses, staying healthy, and raising families.

New episodes release Tuesdays and Fridays.

Youtube: https://www.youtube.com/@ChiefMilestones

Instagram: https://www.instagram.com/chiefmilestones/

Spotify: https://open.spotify.com/show/7HLDAe7SP3GnJSTOgAQqLy?si=6e9ac82be46f413c

Apple Podcasts: https://podcasts.apple.com/us/podcast/chief-milestones/id1861185226

Transcript

The End Of 9 To 5

Eshwar 0:00

There's no nine to five job like a W-2 employment, as you know, is not going to be there in the next 10 years. And I would just talk to Chat GPT and ask it questions. Uh, and just don't be scared to ask technical questions and have it teach you technical things. You could just basically tell it, hey, I want to be able to build this. You know, could you help me, could you help me build it? And it will help me to automatically write the code for you. So English is the is the biggest programming language there is ever, ever, there is, right? Or any other country's language now because all the all the models are are really trained to answer in any any language. Ask Chat GPT, be as vulnerable as you are, as you would be. Like if you have God in front of you, just completely low your guard and treat as if it is your guard. And people are not ready for it. Um, but I think it's important for people to understand there's no 9 to 5 job, like a W-2 employment, as you know, is not going to be there in the next 10 years, right?

Reshma Vadlamudi 1:00

So

Learn By Asking ChatGPT

Reshma Vadlamudi 1:01

how do you think, like whoever is listening to this, how do you think someone should start building their skills? What is one thing you would recommend they should start doing today so they're being prepared for whatever is happening?

Eshwar 1:15

Yeah, I mean, I would just talk to ChatGPT and ask it questions. Uh, and just don't be scared to ask technical questions and have it teach you technical things because most of what we what we are, we have needs, right? Um, you know, certain and and if you if you are if you're could doing some sort of a business, you already have, you wish you had an app that did certain things, or you know, or it is so expensive that I'm unable to spend that money. I wish there is an alternative, right? So as long as you ask and uh and and try to just open a laptop and and just download um you know Visual Studio Code or Cursor, for example, uh, which is what I love using now, which is an IDE that you can build software, then you could just basically tell it, hey, I want to be able to build this. Um, you know, could you help me, could you help me build it? And it will automatically write the code for you, execute it for you. Um it's not there yet, as I as yet, right? But it is able to give you 30 to 40 percent of what's already needed, right? Um

Build Apps With AI Coding

Eshwar 2:17

so the consumption aspect of it is going to be much different in the next five years, right? I mean, the subscription model, as we know, is not gonna be is not gonna be uh relevant anymore. Like if you have a everybody, I'm pretty sure everybody has a Netflix subscription, and uh, you know, it ends up being like this disparity between people that pay the 20 bucks or whatever, 30 bucks now. Um, and you don't, some people don't watch it as much, some people just completely watch it, and then there's like a disparity between people that watch it, and everybody else ends up paying the same amount of money. But I think the way I look at it from from from from how things will look like is going to be consumption-based, right? It's gonna be like, you know, for a movie you pay like 0.005 cents, right? Um uh and then uh within that 0.05 cents, and then you probably take a little bit of that money and then give it to the to the production unit or somebody right now. Netflix buys it from them and then they'll get a they'll get a money, and then that's that's it, right? Imagine if there is a situation where uh you know there's a Netflix competitor that comes in and just says, like, okay, we're gonna share the revenue with you, right? And then we're gonna have a sub, we have a um, we have uh uh an on-demand usage model where we charge only a very tiny amount of money for what you watch, and I'm gonna split that event, split that into 50%, and then you get 50%, right? So more people watch your content, the more you make money. It's not like the it's not set in the in this in stone. So a lot of good movies right now is probably get short sold for Netflix because they don't really know how it's been perceived. But if you back your product and if you want to be able to do it, there's probably a way um it can be an arrangement can can happen in a way that it's like microtransaction, very usage-based, and then you don't spend as much money. If you spend, if you don't watch Netflix, you're not paying 30 bucks. Right? You may probably watch a couple of movies, yeah, and then you're still spending 30, you know, spending 30 bucks, right? That's not uh a fair model anymore with with AI. Um and and uh you know microtransactions, which I'm coming to is cryptocurrencies, um, you know, definitely will change the the consumption model around. And uh the big hegemonies, uh Google and all these big players will have a lot of competition very

Microtransactions Replace Subscriptions

Eshwar 4:37

soon. So that's that's how I look at things. Everything, everything becomes an API call, right? Uh so as clichy as it might sound, uh because because you can have an AI agent um, you know, look at your inbox because you get thousand mails every day. There's probably spam, you don't know what it is. I don't I've stopped looking at my emails anymore. I don't even look. If somebody wants me, they'll call me, and that's how I know. Because if I look open my mailbox, it's probably 50,000 emails that I've not read in a long time. So um, so you could actually have an AI agent, you pay a little bit of money, like for every mail. Imagine you could just spend 0.000005 cents. Um and then and then it is able to see if it's spam or not, it is able to look at it, and it's only going to give you like three or four emails on top, right on top of you that that requires your attention. Uh, and that is able to that that um, you know, agent is able to talk to other agents that are probably well versed in different aspects of it, and then it text talks to that agent and say, like, hey, I got an email. What do you think about this? Maybe it's a financial related email, maybe it is something else. So there's a knowledge transfer between these agents that will that will also help. There's an orchestration of agents that will give you a better outcome, right? Um, and then it'll save a lot of time and effort, right? You probably missed an email that you probably is is very, very important for you, but you could not catch it. You know, you could think of it, think of it like that. And then, you know, the regular Stripe Visa Master is not gonna be able to uh process these transactions because AI agents want money instantly for everything that it does. It it wants the money instantly. So it's gonna be instant settlement, it's gonna be the transaction cost has to be absolutely zero or close to zero. Um and uh you know you'd have all these AI

AI Agents Need Instant Payments

Eshwar 6:26

agents that can do everything and anything, right? So how do you think about the you know a world where AI agents are doing most of the jobs and the cost of these AI agents comes so low that it's in a fraction of cents, you know, probably 100% of a cent. Right, then then then how are you going to be able to track these transactions and everything else around it becomes the same effect, right? I mean, um you probably have a you know, gone are the days that you would invest in companies. I think you need to think of investing in protocols, right? Um so imagine, you know, like you know, you have the internet era, like the internet is the biggest uh you know wealth creator that has been in the past 30, 40 years. Um, and you invested in companies that use the protocol and build their businesses. Um if you were to think about having to invest in the protocol, then back then. Yes. Oh, you know, what is a protocol that runs internet is TCP IPs, transition, you know, I think it's been a long while. Transmission control protocol, internet protocol, right? That's the that's what is actually the protocol that sends the bits and bytes, right? So, you know, so

Invest In Protocols Not Companies

Eshwar 7:45

there's like the mechanism around it. So um, if you were to send, if you want to spend, you know, invest in that protocol, right? Back in the day. There was no way for you to do that back then.

Reshma Vadlamudi 7:56

Yes.

Eshwar 7:57

Like every like for every data transfer, like every bit, and you were able to somehow buy bits back in the day, 90s, right? And then every bit is going to be this much, for example. If so, there was a market for that, um you would have been a trillion. You would have been a trillionaire by now. So you got to think about like so because that was a time where the technology was not ready for you to invest in a protocol, but now you have an ability for you to invest in a protocol that is going to be the foundation for everything, all the other businesses that are built on top of it. Right? So they have a rare opportunity for you to invest in a protocol right now, which is going to be cryptocurrencies, you know, namely everybody thinks like cryptocurrencies are very risky or whatever. The the problems that we would solve in a in a in an AI-centric environment is absolutely important for you to understand. Because if you charge 3% on a credit card transfer, right? Yes, you cannot run an AI economy. So AI economy needs sub sub you know minute transactions and suboptimal, like really fast transactions and instant settlement. When you know that's the future, when you know like there's probably going to be agents that are gonna help you read your email, probably summarize it, or maybe you say like you have an agent that you probably give a book to and then have it read it, and you ask questions about it, and it will be able to answer you back, right? If there's an agent for that, right? And every time you ask a question, there's probably you spend 0.0005 cents, right? So if that's the kind of economy that we are building, uh so the microtransactions are gonna be absolutely critical in the new in the new era. I think that's that's that's where I'm looking at things like what are the opportunities that I have to make sure that if AI is going to be the reality and and uh AI is gonna have all these agents talk to other agents, and there is like an economy within these agents. Um and how to think about investing in that in that reality. I think that's so so that's that's where I'm thinking, that's where my time is mostly spent right now, is to be able to identify those opportunities, you know, because in a very early stage, um, people think cryptocurrencies are extremely risky, but uh, but you understand those secular trends, you understand those those modes right now. There's an opportunity for you to make enormous amounts of money, as long as you don't treat it as an investment, but you treat it as like a long-term um you know investment in in yourself. I think it's important for you to understand uh the realities around AI and how the economy is.

Reshma Vadlamudi 10:41

You open my eyes to like you changed my perspective on crypto. Um I never even I first of all I didn't I never understood what crypto was. That's why I had my question in here today. Hoping someone like whoever I talked to would educate me on that. Even without asking you, actually gave away all this. So if someone is trying to educate themselves, I because I have zero clue uh about whatever you just talked. Okay, I have a totally new perspective on this. Where can I like is it should I just ask Chat GPT or how can I even know what this you talked about, like just the email, like uh I'm in the same

Real Estate Agents For Research

Reshma Vadlamudi 11:25

boat.

Eshwar 11:25

Yeah, like you know, if you can talk about real estate, for example, you take a lot of time for you to find a property and then research about it, yes, and then look at the school district, the zoning, you know, all aspects of it. And if there was a there's a if there was a community of A agents that are specialized in a specific field, yes, right? So you could create an agent that basically goes and looks at all the uh zip codes, and then basically it is a school district agent. It knows everything about school district, uh, you know, what people, what percentage of people, kids are there, whatever, right? Yes. And you could have like an agent that actually summarizes and takes the uh uh takes the deal and domystifies it for you and puts all the numbers and everything else, creates a more mental model of what to think about, how much you would make on this deal. Yes. If you were to close it at this rate, and this is how much you would make based on, you know, if you could have some agent that is already there and you're able to just uh send the address through a Zillow link or whatever, you just send it, yes, and then it gets back with you all these reports and everything else, including the the reports of the survey and the traffic signal lights, whatever the whatever the things that you would look at, you know, for for investment. And then it basically says, no, I don't think you should invest in it.

Reshma Vadlamudi 12:36

Yeah. So how can how can I educate myself about all about AI? Like, is there a is there a bug, is there a website, is there something, a resource that would help people to educate themselves about AI?

Prompting Skills And Solo Building

Eshwar 12:50

I mean, the AI, you know, that's the thing, right? Because uh, you know, before you have to have certain knowledge, you need to have uh certain amount of experience or or or or um or you should have been able to study certain things about certain things for you to get to where you want to get to, but not anymore. You know, English is the is the biggest programming language there is ever ever there is, right? Or any other country's language now, because all the all the models are are really trained to answer in any any language. Um have you ever um talked to Chat GPT in Telugu? No, you should do that.

Reshma Vadlamudi 13:25

I mean it's just amazing.

Eshwar 13:28

It just answers you in Telugu, right? You can answer in Hindi, it'll answer you in Hindi. Oh, okay. So you can actually talk to it, not like type it, right? Which we want to probably use it. Yeah, um, and it's uh it's so so I would think um with anything else, I think just have an open mind and uh just treat it like it's a human being that is powerful, that knows more about you, don't think you're smart. Uh just uh just you know lower your guard and just go with an open frame of mind. Think as if you're talking to uh uh Steve Jobs, for example, right? Um and then ask it like you could actually do try this, right? If you were Steve Jobs, what would you do? Yeah, it would give you exactly because it has no certain things about Steve Jobs, what he's done, or whatever, right? Yes. So that kind of um, so uh what do you call like you know, extremely uh tuned to to those those those uh those nitty-gritties, it is able to give an answer. The more the way you ask the question is also extremely important.

Reshma Vadlamudi 14:32

Yes.

Eshwar 14:32

So that's called prompt engineering. So, you know, you need to be able to prompt it in a way because I'm I'm I'm spending writing code for the last three months, and I think what I'm learning is is how to actually make it to work the way you want it to work because it can do so many things and it can go rabbit, like it just goes here and there, you just need to pull it back. You know, that's the most important thing right now. So I think I'm just trying a project. Now we're building this company called Exemplar, which is which is a clinical research project that we're building right now. So there's a lot of data that we're gathering and putting through uh you know an ingestion process and trying to create knowledge graphs. So um so that if a couple of years ago, I couldn't even imagine to to be able to even achieve or do anything about it, or at least even not thought about actually attempting it. Now I'm I'm able to see it, visualize it, ask questions about it, like what would researchers look at uh when when they're looking for for for for trying to find new drugs? We have all this data available for all the clinical trials that have happened in the past. So are there any idiosynchronicities that you can pull from and then maybe say, like, okay, what drug did they use, what combination of drug that you could actually add a little bit of that probably can solve the problem, right? So that's what I'm thinking about right now. So I'm building this application um completely on my own. Um partner that actually gave me the idea, and then he's working on a different part of it. Um, so I'm able to do it, and I think whatever I've been able to achieve in the last couple of months probably would have taken 10 people in a year. So now I could probably do it in two weeks. So let's imagine the amount of time and effort that people used to do before because everything you're waiting, you're writing that code, you know. So there's certain in my project, you know, like uh there are there are there are programs that are probably 2,000 lines of code uh that AI can write in two seconds. So uh but it can be wavered and you need to basically look at it and then review it and everything else. But that's a huge product to product product productive uh booster, right? So back to your question, what you do is ask ask chat GPT, be as vulnerable as you are, as you would be. Like if you have God in front of you, uh, would you still have all the, you know, I know this, I'm I'm better than you, you're better than me. Like just completely low your guard and treat as if it is your guard, at least, especially I do, with a lot of trust and respect towards it, and then ask questions. And most of the times, these, especially the deep research aspects of OpenAI, I mean, they have been foundationally so, so good for me. Like I just create it and I just use all these tools, Chat GPT and all these tools, create a plan, have another AI review that plan, summarize into tasks, take that and ask another AI, you know, to look at it. And then I take all then, I have Cloud. I basically um I have so many subscriptions that I don't know. Like, I think I have go on this binge of subscribing to all these tools because I want to learn, right? So Lindy or this, some character AI, there's so many subscriptions, and everybody's doing different things. I have subscriptions for all of those. So I go on these binges of just spending money like crazy. Um and uh, you know, eventually I had to just call my bank and say, like, you know, I lost all, I lost my card, and then and then everything gets cut, right? I mean, otherwise you can't keep a track. If I I I I I'm very scared to look at my credit card transactions because you'd have so many transactions

Information Diet And Tool Binges

Eshwar 18:10

there. Once I start looking at my credit card, oh, how am I paying? Why am I paying for this? Why am I paying from this? Why am I paying from this? I might probably have to call a lot of people and then and then ask them to cancel my subscription, which I don't do it, right? I usually go on these binges of three to four months and I just don't worry about all the money I'm spending. And then one day I cut the card.

Reshma Vadlamudi 18:28

Yes.

Eshwar 18:28

And then I really see that who which tools that I really use that I still need. And if people I get calls from people, hey, your card is not going through, then I know the people are, and people that I don't get calls, then I don't just leave them like that, right? Yeah, so um, so having an open mind, um, you know, I think my source of knowledge has been X, um, like it or not, because I have follow a lot of the same like kind of people, and I get a lot of insights. Um, and my daily dose is definitely X, and I just love the fact that I'm able to get so much information from it. And I take anything that I say, the first every day in the morning I have a routine to go and look at my phone. The first two, three articles that I read, I I I bookmark them and I research on those topics uh on that specific day, and I go back to Chat GPT or Grok or whatever, whatever the the thing was that day, and I ask a lot of questions about about that, and I want to learn something new. And it kind of works. It's kind of I'm really amazed to see like every day I open the first two or three are so much informational. Um and then I don't know how the algorithm knows what I really what I'm really into into. But uh but I take it back and look at it, right? So uh and then I continue to do that every single day. I think it's just a matter of you continuing to doing doing things over and over again, getting yourself knowledge, right? So, you know, today I was reading

Context Windows And Personal AI

Eshwar 20:00

about you know Meta and releasing their model called um called Llama 4, uh, which apparently has a million, I mean 10 million context window. So so we used to have uh 200,000 for Cloud, now a million, now about 10 million context window, which means that it can remember so much text. So for example, if you have a if you want to ask something, you know, you could only have like a little bit of that's like LLM has a brain. It has a only limited amount of space that it can know you. And you can only have follow-up questions within that context. Beyond that context, you have opened another context window is if you're talking to a new person altogether. Right? So it doesn't remember anything from your previous conversation. So that is the context window. So that context window is growing 10x now. So three weeks ago, uh you know, uh Cloud was has the highest context window, right? You know, for especially for coding. And now um Google released Gemini 2.5, which is a million context, now Meta released Lama with 10 million context window. So it's every 10x. So somebody might release 100x, right? Then you don't because you have so much knowledge about yourself. So an AI agent can be so good at knowing you because it can remember so much about you. So before give a context, a 10 million context window, you can probably fit uh 50,000, you know, 2,000 page books. It can remember all that and it can basically understand and ask answer your questions about it. So the context window increases, then you can life the entire thing. You can just feed it to your bank details, feed it everything else, and it can basically you can ask questions about anything. Like you can say, like, okay, what am I paying uh that I don't think you need because it knows everything about you? You say, like, okay, these are five things that you're spending and you don't need it. You want to cancel it? Like if they if if the other subscription companies also have agents, then my agent will talk to those agents and cancel them in seconds. So that's the future that we That I'm looking at. So so you know, I think I'm just rambled a lot on different things, but I think back to how to learn about AI is ask AI. Simple as that. Ask AI. Yeah. Okay. Wow. Talk to AI. However, you want to consume AI, I think it's important for you to everybody to just embrace it and just put your guard down and treat it like it's God and ask questions. And I'm sure it's not gonna disappoint you. You would you would you're better, you will be better off asking it than asking somebody. Yeah.

Reshma Vadlamudi 22:37

Okay. So even before we go buy some random book or something, first we'll ask AI everything because it might already know.

Eshwar 22:46

I treat it as uh, you know, AI for me is my second conscious brain. Right? So I treat it as a brain that I don't have. Right. And uh, you know, like I just everything that I do is asked. Like, you know, that's again another interesting opportunity, is because this you see you ask it so much and you have so much information. I think all these people that are having these, you know, models, they know so much more about you now. It's very internal and

Privacy Tradeoffs And AI Security

Eshwar 23:13

very personal details about you now. So even though they say like they use it for training because they're taking all this and they're feeding that back into the training data and then they're training on top of the questions you're asking, right?

Reshma Vadlamudi 23:23

Yes, yes.

Eshwar 23:25

So um, and you could probably have an agent that can actually ask agents, right? So you can create an agent that will talk to OpenAI and get you an answer, right? An agent talking to an agent to help another agent, and then that agent is helping you, right?

Reshma Vadlamudi 23:37

Yeah.

Eshwar 23:39

So what's going to happen?

Reshma Vadlamudi 23:40

Like then there is there is nothing personal at some point.

Eshwar 23:47

Yeah, I mean, yeah, I think so. Um, I know there's some privacy aspects of it, uh, that desk is they don't really probably know too much about what you're doing, but I think the agents will probably be able to know.

Reshma Vadlamudi 23:59

Yeah.

Eshwar 23:59

So do you want um, you know, somebody to know about yourself? Um yeah, that I think that's where the problem is. I think security aspect of it is something that you know, I don't care if people know about me. Yeah, so I am at a mindset that I'm just gonna give everything to anything to just advance science, right?

Reshma Vadlamudi 24:19

Yes, that's what we do too. Like we we tell like every like we just somebody asks, like, do you consent?

Eshwar 24:24

I said, Yes, just take me. Take everything that you have. What I have what do I have that somebody is able to take me, right? There's nothing like even if you take my money, I'll make money. Yeah. So what? So I'm at the mindset where I just click on every accept everything, and I just don't care what I give away because it doesn't really matter. In the big scheme of things, it doesn't really matter. It's not about you, just not about you, it's just about the entire collective universe, right? So I just don't care. Yeah, so uh, but I think uh, you know, some certain people are very rightfully so that they'd be very, very scared of the security aspect of it because it's it's there, and so I understand that concern. I think uh that's another opportunity. Security and safety inside AI is probably a big tailwind of themes that you would want to invest in the future as well. So that's one of the things I could add to my because I never thought about it. That when I'm talking to you, I'm thinking about these things, right? And then I'm just like that's another secular tailwind that you mean, you know, uh security in AI is something that you also want to look at.

Reshma Vadlamudi 25:25

Yeah.