[00:00:02] Rajan:  AI is reshaping businesses, and ignoring it could leave your industry disrupted. I'm Rajan, and this is Pivotal Clarity. We talk to those building or using AI, founders and engineers with real world experience. Our aim is just to cut through the hype and see where AI is truly making an impact. If you're a business or following tech trends, these conversations offer clearer insight than most of the press. Let's get into today's episode. Welcome to pivotal clarity. Today, I have with me Jesse Anglin, who's the CEO of Rapid. It's an AI product development company. And just like the name goes, they help you generate an MVP in ninety days. So one of the most interesting that I learned from Jesse recently was that they helped build manufacturing system in thirty days, and that helped replace about 300 jobs. Jesse, why don't we start there? Tell us what did you do. Yeah.

[00:00:58] Jesse:  By the way, thank you for having me on. I've listened to several of your podcasts, and I love the work that you do. And so so that project was very interesting. And it brings up really this very singular, very interesting thing about AI, which is that a very small amount of work can reproduce a very large amount of work, human effort. Because I had a friend explain to me like this, and it was actually brilliant. He said a long time ago, when someone went to take a picture, the problems that needed to be solved were chemistry problems. And chemistry problems require a certain amount of effort to solve. And then when digital cameras were created, they took a chemistry problem. They made it a mathematical problem. And because computers are incredibly good at doing math, they basically just took photography and, like, blew it into the next existence. Well, in a lot of ways, with LLMs and synthetic intelligence, you're taking an organic computing problem. You're turning it into a digital computing problem, and then that allows you to perform tasks at scale. And so with this particular business, what was really interesting was he called us up. He had already automated his entire manufacturing floor. And so there was, like, two or three people that monitored the automated systems, And an order would come in from the sales team. They would punch it into the software that ran the automated manufacturing floor. The parts would be cut. They would be packaged, and they would be shipped autonomously and head out to the person. Right? And so what he had done ultimately was he took what used to be, let's say, a thousand person operation, turned it into a 300 person operation with robotics and AI and some other elements. Now I wasn't there for that part of it. But what he didn't think was possible was taking the human portion of it, so the sales team, the SDRs, all the incoming requests, and actually apply some sort of a system to it, in this case, an agentic system. They could look at all of those emails, answer phone calls, do all the things that needed to do, and then feed data back to the manufacturing floor to get everything out. So when him and I were talking, I was like, no. No. No. Like, this is actually not a complicated task because he started showing me the emails that were coming in and what needed to actually happen with the people that were doing the work, and it just wasn't very hard. Like, I could get a standard out of the box LLM to do it, like, really, really easily on the fly without any extra training, without anything. You could just do it out of the box. And so I told them, listen. Give me a week to prove that I can actually replace these people, and I will give you a proof of concept in a week with one developer that proves that I can replace your entire Salesforce, at least your incoming Salesforce. And so we took a week, built a little proof of concept with a small amount of data that didn't take into account any of the edge cases or weird things that happened or that could happen and showed it to him. He's like, oh my gosh. This is actually gonna work, isn't it? Yeah. Like, a %. And so we took that same developer, worked for another three weeks, and then delivered a product that actually allowed him to take his entire incoming Salesforce, which was 300 people that were basically receiving emails, phone calls, all this other stuff, and inputting that into the automated manufacturing floor. We're able to take all of those people and replace them with one simple agentic system built by a single developer that honestly works twenty four hours a day, seven days a week, has absolutely no problems. Like, it was a massive upgrade.

[00:04:16] Rajan:  And it was a super fun project. Yeah. For somebody who's experiencing this sounds very magical, but, you know, for someone who's going through the replacement, you know, that sounds very scary, isn't it?

[00:04:26] Jesse:  Yeah. It really depends on what side of the equation you're on. I mean, for a business owner, AI is kind of a dream come true because you can build a synthetic human intelligence that does tasks for you. It never has a fight with its wife. It doesn't stay up too late drinking. It doesn't have bad days or good days. Like, it's always on time. It works twenty four hours a day, seven days a week with the same attitude. It always does the same quality of work no matter what's happening. Like, for a business owner, access to this technology is amazing. For the human beings on inside that equation, it is, I think, in some cases, even tragic. That's where, like, one of the things that I'm very passionate about is helping people who are going to be replaced by AI to upscale their current level of understanding. Because, ultimately, at the end of the day, anyone who can work with AI and understands how to use it will not be replaced because we're always gonna need people maybe not always, but for right now, at least, we need people that can interface with the machines and make them work and be efficient. Like, AI really augments your very best people instead of replaces them. And if you can be one of the best people and be augmented by AI, your job is probably safe. If you can't, then

[00:05:36] Rajan:  you're gonna have a tough time.

[00:05:37] Jesse:  You're gonna have a tough time. Yeah. And it's only gonna get worse because the technology is only getting better. Yeah.

[00:05:43] Rajan:  So when a client walks to you, a customer walks to you and makes a request, walk me through the pipeline. What does something from the request to what you ship? How does it look like?

[00:05:53] Jesse:  If I'm gonna talk about this, I gotta tell you about the coolest thing that I'm working on right now, actually. So let's say you've got an idea for a project. Right? The traditional way of going about this is you call up, talk to somebody who's almost like a consultant. They try to understand your project and figure out the architecture, the front and the back end, like, all the different pieces. Because we have to figure out how to attach, like, a value to your project, like, the amount of time it's actually gonna take to get it done. Right? Because if I said, hey, Rajan. Like, thanks for the awesome idea for the project. Write me a blank check. I'll start doing work, and then you just trust me that I'm not gonna overcharge you, and it's gonna be good. Like, no one can work inside of that environment. So you always go through this discovery process, estimation process, and developers just always get that wrong for the most part. It is a guessing game because you're trying to predict what the future looks like, the problems you're gonna deal with, all of that stuff. So either developers will underquote and then hope and just try to deal with those problems as they come up, or they'll weigh overquote so they don't have to. And so what we ended up doing well and there's actually one other problem that is important to remember. Your estimation is only gonna be as good as the guy who gathered the information to put it together. And so if your SDR or your salesperson that you're talking to doesn't understand what it is you're doing well enough to be able to communicate that to, like, the guys who are actually estimating the amount of work it's gonna be, it'll just be off. And these are all the normal problems. Everyone deals with them every single day. It's not weird. It's accepted that these are problems that are gonna come up. And so we started working on basically an AI estimation and discovery process that's an agentic system where you've got one agent that you're having a conversation with. It'll be live on our website, hopefully, here in the next two weeks or so. But you've got one agent that you're having a conversation with, and then it's got all these other agents in the background. One's acting as a senior full stack developer. One's acting as a back end developer, a front end developer, an AI developer, a blockchain developer. And so we've got all these different agents that then start contributing to the questions and the consultation with the client. And so as the client's having a conversation just like they would with, like, chat g p t real time if you use the app. So you're having a conversation. It's asking you all the relevant questions, helping flush out your idea, pulling everything out that it needs. That system then takes all of that information, basically, the transcript, and it architects the back end that estimates fully the product. So, eventually, in about two weeks, the process will be someone gets ahold of us. They decide either they want to talk to a person or they don't. I'm hoping that more people wanna actually talk to this agentic system more than wanna talk to a person. I would want to personally. They get their idea estimated out. We figure out whether or not we can fit what they want to do inside their budget, our timeline, our resources that we currently have available. Once we get all that stuff figured out, then ultimately, we onboard them onto a development team that specialize to help them build whatever it is they're building. And because we're doing custom software development for the most part, I've got different guys that I try to keep focused on different areas. And so if you've got, like, a voice idea that you're trying to build, I've got a group of people that have done a lot of work on voice. If it's image recognition, it's gonna be a team of people focused on that. So we assemble the right team for you. And then for us, the name of the game is move as quickly as possible with the best foundational code you possibly can because we don't like building, like, throwaway code. We like to actually have enterprise, like, a really, really solid foundation for expansion in the future. And so we'll spend a fair amount of time on the architecture, really understanding how everything needs to be laid out, and then we get to developing. And the idea is build as quickly as possible so that you can get to market as quickly as possible so that you can get feedback from your users as quickly as possible, mostly to avoid the mistake of entrepreneurs who'll spend their entire life savings and a bunch of investor money building forever a product that nobody's interested in using. And so for us, it's all about trying to get to that initial finish line as quickly as we can so that they can get user feedback, start making adjustments, create a product that people enjoy using.

[00:09:48] Rajan:  Looks like a lot of the initial heavy lifting is is around the right kind of understanding and right kind of estimation. Then that moves on to before developing, you're doing, like, design and architecture, and then, you know, you're doing the actual development. And then you're making sure that there is some amount of user validation that is there so that they're not building a product that nobody is gonna use. So in this entire process, like, how does it differ, like, now with AI versus how it was done, let's say, two years ago?

[00:10:15] Jesse:  I would say there's probably one major difference, and that is that for the most part, none of the tools that exist currently for, like, helping a developer write code are good enough to replace a really good developer. Like, all of my best developers are going to write code that is far better than anything that an LLM can produce. Now that's changing really fast. Like, the code quality is getting significantly better and better and better and better. But what AI can do at the moment or a large language model can do at the moment, depending on the one that you're using, is it can write code probably as well as a two year developer, give or take, something like that, year and a half developer. Well, there's a lot of stuff that just needs to be done. Like, the really, really smart, difficult, hard, complex work as a developer is actually not nearly as much as people would think. There's just a lot of things that need to be done. I kind of equate it to, like, putting in the electric for a house. You could hire someone who has absolutely no experience to go run the wires through the house. It doesn't take a lot of knowledge. It doesn't take a lot of expertise. It doesn't take a lot of anything. Like, almost anyone could do it. But when you actually go hook it up to the main power, you wanna make sure that you get it done right or it's going to create massive, massive problems. And so what AI has done is it has taken the time consuming part of the project, which would be running the wire through the house. And it basically just does it for you, and it probably does it as good or better than most of your, say, freshers that you bring on or interns, whoever you bring on there to just do that. You're one year to your guys. And it'll work as hard as 15 of them. And then the senior developer or the guy that you have that you're really trying to augment their talent, they can go in and do everything that's really critical. And so when you get rid of all that time consuming 80% of the work, it basically just makes you way, way faster with way, way less people. And so really the overall effect is that it's a lot less expensive to do it and you can get a lot done in a lot less time.

[00:12:17] Rajan:  What's the difference in development time before and now?

[00:12:20] Jesse:  I would say that it depends on the project. But if you're just looking at, say, a run of the mill standard project, you probably reduce development time at a bare minimum by 30%, but probably maximum 80% at times, maybe even more. I mean, I've seen some projects where it would have taken a team of four guys three months to do, and it took one guy one month to do. And so in some cases where the idea is relatively straightforward and simple, you can just knock it out of the park. But it really depends on the complexity because I would say there's even some projects that we've taken on where AI isn't able to help at all because the idea is not intuitive, and it's very novel. So AI struggles to work on novel concepts. I mean, the way that it writes code is by looking at code that's been written, and so it was trained on code that's been written. If you give it an idea that really has not been where there are not enough examples in the training dataset, it's really struggles. And so on those projects, it doesn't help. It still does. I mean, it does a lot. It just doesn't help as much.

[00:13:23] Rajan:  So in this new AI centric development pipeline, where are things breaking?

[00:13:28] Jesse:  That's a good question. No one's ever asked me that question before. I mean, I don't know that I would be able to quantify that in the sense that it seems like everything that was breaking is still breaking. Like, you have regression issues. You have, like, all of the normal issues you'd have in development, but I don't actually see really anything new showing up. It's all the same problems that developers have always had.

[00:13:48] Rajan:  So you said you talked about this example of what took four developers three months to do. One developer did it in one month. That's like a massive compression. But is that the best example that you've seen? I mean, what is an automation success that you've seen that, like, truly blew your mind?

[00:14:02] Jesse:  So I have a client we're working on a project. And in his industry, the POC that we put together, which really you could probably consider an entire project, took one of my developers about five days to build. This may not actually even be the right example, but I'll give it anyways. But this POC could today replace about 15,000 employees just for this client overnight. And not only the customer experience that their customers would have would actually probably increase by a pretty significant margin, and so there's some very low hanging fruit. Now this is more of a compression for a business owner, which I would say I think more like a business owner than I do like a developer because for him, like, you can't even really quantify that, like, on a cost basis. Like, what's a cost for a really, really good senior developer's time for a week? Even at the highest prices that you'd pay in Google or in San Francisco, like, a week's worth of time is maybe at the most, you might get, like, $10,000 maybe at the most, maybe 15 for someone who's crazy, who doesn't even write code anymore, which means that for $15,000 worth of a human being's time, you can solve a problem that is probably costing them I don't know. What's 15,000 minimum wage employees cost? Or even if they're all outsourced in The Philippines and India, like, you're still talking millions and millions of dollars a year, maybe $80,000,000 a year. So now here's the interesting thing is that with this particular client, and this is a problem most business owners are gonna face, can you do it? Like, if you do it, is it gonna destroy your reputation in the marketplace? Is it gonna cause problems with your stock prices? Is is it gonna create the kind of havoc that you can't recover from as a client when you have that kind of economic arbitrage that you can take advantage of? And so in the case of this particular client, he's not moving on it yet because it would be catastrophic. And I actually think there's a lot more business owners out there today facing that problem than are facing the problem of how do I actually implement and use this, which is why when it does happen, it's gonna happen crazy fast. It's gonna take a lot of people by surprise because as soon as one of the big companies really takes the time to automate the processes that they're using inside their company, and all of a sudden, they reduce their overhead burden at a very, very minimal cost. They reduce their overhead burden by $80,000,000 a year or a hundred and 80 or a billion dollars a year. Their competition has to do it. They have to. And so when that first domino falls, it's gonna create havoc worldwide in my personal opinion. And what the other thing I think is probably going to happen is the bigger companies are probably gonna fail to move fast enough, and it'll be a smaller company that comes in and just takes market share from them, and they won't understand how they're doing it. Because I've got a client right now or I guess is even really just a prospect because I haven't signed any papers yet, but I'm very excited about them. So it's a group of entrepreneurs in a very competitive space that have a lot of industry experience, And they got together for the sole purpose of starting a company that has AI as its foundation. And so they're starting a brand new company, and they're basically instead of hiring an s secretary and a sales team and SDRs, instead of going out and finding software that they want, they basically raised a bunch of money in, and they decided that they wanted to build proprietary novel software that at its very essence is an agentic system that runs their company. Because then they won't have 15,000 people to go and lay off when they want to bring an AI on. They're not gonna have all of this disruption that needs to happen. And I think that that's actually a brilliant strategy. He's my second client that I've gotten that has had this idea where they're saying instead of retrofitting AI into my existing company, they sold their existing companies to somebody else who wants to run a traditional company, and then they're using that as capital to start a new company in this new AI era. And a lot of them were inspired by what Sam Altman said. So Sam Altman said the other day, within the next ten years, I think he might have even said five years, we're gonna see one unicorn. So a company that grows to a billion dollar valuation in less than five years that only has one person in it, one employee. And I'm a % sure that's true just based on being as deep into it as I am. Like, I see it. The writing's on the wall. It's just a matter of who's brave enough to pull the trigger because it just creates a very, very interesting environment for planet Earth that we've never faced before, which to me is very exciting. But I understand why people are scared of this. It's crazy.

[00:18:30] Rajan:  What tools do you use? Do you use your own tools that you build? Are you using, like, you know, new tools that are coming in the market for doing a development? What does your tech stack look like?

[00:18:40] Jesse:  So for some stuff, we'll use our own proprietary stuff that we've built.

[00:18:45] Rajan:  These are agentic frameworks, or are they, like, you know, IDEs you're building?

[00:18:50] Jesse:  Are you building, like, plugins too? No. I mean, the thing is if other people wanna go and create, like, IDEs that integrate with, like, Anthropic or OpenAI or Walmart or whatever, let them do that work. We'll just use it. I'm not above that. I don't wanna build all my own tools. And so, like, we use Copilot because it's relatively easy to do. And there are some other frameworks that guys are playing around with. A lot of it's kind of experimentation that changes a lot. I mean, for probably the last twelve months, every three months, we would switch back and forth between OpenAI as, like, a flagship model and Anthropic as a flagship model because depending on who had pushed the latest updates, one was always better than the other. And so, like, all of my guys, they've got an OpenAI account. We've got an Anthropic account. We'll swap back and forth. We use a lot of open source, but most of that is all coming from Meta because there's so much good work coming from them right now with the llama series that they're pushing out and things along those lines. And we'll use, like, knowledge graph for rag systems, and there's a handful of them. And all of them are changing. Like, that's one of the things that right now in this space, like, if I was going to lay out my tech stack that I'm using today, if you and I did another podcast a month from now, it would be different, almost for sure, because people are making such great improvements in the technology that's available for developers to actually use that if you decided you just wanted to use one and they weren't innovating at the same speed as their competitors, you would get left behind, and you'd be the victim of that. And so, yeah, so it changes.

[00:20:20] Rajan:  Got it. So other than OpenAI and Claude and like some meta support models, are there like IDs that you use? Are there like tools that you use, like in your development pipelines? I know you said Copilot.

[00:20:35] Jesse:  Ultimately, I'd have to go and ask my guys which stuff they use. I let them have a lot of freedom with what IDs they wanna use and things like that because we are very much an experimentation company. I mean, I know, like, Versus Code is like, everyone kind of already uses it, and there's so much that you can do inside of it that I would guess that if I hold every developer that currently works for me, probably 95% are using Versus Code. And it's really, like, how they have it set up and what stuff they're using. I mean, you maybe you see guys using JetBrains or things like that, but I would guess it's not as much. I haven't used JetBrains in a long time. I don't even think since the AI side of things came out, I'm assuming they have integrations with AI and stuff like that. I would assume so. Otherwise, they're dying, but I'm not even sure.

[00:21:18] Rajan:  How do you test this AI generated code?

[00:21:20] Jesse:  So one of the kind of core principles at Rapid is that you cannot trust AI to do anything yet because it's not trustworthy. It hallucinates. It doesn't actually write efficient code or do things like that. And so, really, we test it and we trust it in the exact same way that if I had a one year developer that came on board or a fresher that came on board, like, we've got systems where we write test cases, uh, surpass those test cases. Senior developers go and do an audit. They take a look at the code that you've written, make sure that you're pointing in the right direction. It's not any different than working with a group of more junior developers because at the end of the day, AI code really looks like it's been written by a junior developer most of the time. The other thing that we have discovered, though, is the prompt engineering side of things. I'm assuming you've played with AI models enough to realize that probably three quarters of the result that you get is going to be based on your prompt engineering, the context that you give it, like, how you actually go about telling a machine to do something for you. And so for us, like, one of the things that we've really focused on is how to actually do the prompt engineering when we're using these LLMs to write us code so that you get a better output. Because if you go into Claude and say the classic example would be, like, create the game snake. It'll do it. And I bet you 95% of the time, it actually creates a game that works. But if you go into Claude and you take five hours and you think through exactly what you want in the game snake, how you want it to work, how you want the functions to work as a senior engineer, and you think through the architecture of it and you give that as a prompt to Claude, even a one shot prompt, the the output is probably gonna be 80% better, maybe 90% better. And so a lot of my engineers' time is actually focused in those initial stages where you're getting the code, like, just written out. A lot of their time is actually spent on trying to architect it, understand the code from a fundamental, foundational point of view, and then give those instructions to the AI in a very clear way so that it produces better code. And then also doing chain of thought prompting and using other I mean, really, whatever's effective for what we're doing, but not just one shot prompting it and being lazy. Most people who've used AI a lot like, one of the problems is that you get really lazy. Right? You're like, write me an email, and then it doesn't work. And you're like, man, you suck at writing emails. Well, no. You suck at prompting. Like, if you actually sit down and you give it good instructions and you kind of narrow down those neural pathways that that you're about to go and push your tokens through, you can get really, really high quality token outputs. You just have to take the time to do it.

[00:23:56] Rajan:  So how does pricing work now with AI? How has it changed from how it was done two years ago for you?

[00:24:02] Jesse:  In some ways, two years ago, you basically just paid for your own GPUs. They were easy to get because nobody wanted them, and the demand was really low. But the AI's kind of sucked. Maybe not two years ago. Maybe I'm thinking more three years ago. But, I mean, you would go and grab whatever the latest research was from OpenAI on GPT two, and you would go run that yourself on your own GPU somewhere. And so it was, in a sense, kind of more expensive in some ways depending on what you're doing. I mean, it certainly wasn't worth the output.

[00:24:31] Rajan:  But do you compare that with your development? Like, you think about this as a revenue per developer or things like that?

[00:24:38] Jesse:  Well, think about this. How many output tokens on four o can you get for every 6¢? It's like maybe a million tokens or something like that. I forget. I think it's a million tokens. How much would it cost to you how much would it cost a human being to sit down and write 750,000 words, we'll call it, reproduce the same tokens? Like, it's unimaginably cheaper. Because even if I've got a guy let's say I've got a guy somewhere in India that I am paying almost nothing to because he doesn't know what he's doing. He's a fresher. He's still in college, and he's willing to come work for me for almost nothing. Even that guy costs 50 times more because it's gonna take him a day, two days, three days, four days, maybe ten days to write the same amount of code that OpenAI four o can write for me for 6¢. And no matter how cheap he is, his ten days is not 6¢. It just isn't, and there's no way that it could be. And so human beings can't compete with the output as long as the output is the same quality as a human. It's not possible. It is so much less expensive that, like, you can't almost can't wrap your mind around it. So one of the agentic systems that we've recently built was a system for creating blog articles. And so if you go look at my website and and you start reading the blogs, they're actually really good. It's a very complex agentic system that creates all the blogs. It optimizes them for SEO and for the purpose that they're being created, make sure that they're actually good to read, puts graphs in, and all that kind of stuff. In the last year, I think maybe we've written 1,600 blogs, something like that. If I hired somebody to write those 1,600 blogs, even someone really, really cheap that wasn't going to do as good of a job as the AI system that we have. I mean, to write 1,600 blogs, I bet you they'd charge at least, let's say, $25 a piece maybe at the very least. Well, I can produce one of those blogs with an inexperienced writer that actually brings people to my website. It converts them into real into real leads. I can write that for 15¢, 8 cents, something like that. And so there's the math, 15¢ into whatever you can get someone to write a 1,600 word blog for and a really good SEO optimized one. Like, you can't compete with it as a human. I've been trying to tell people this for a long time. No one believes me, but it's unimaginably better than humans are at the things that it's good at. And most people, their response is, well, I can't even count how many r's there are in the word strawberry. And I'm like, no. No. No. Like, I get it. It's stupid. But at the things that it's better than you at, it is so much better than you that you can't wrap your mind around it. And with the right prompt engineering, you can do a lot more than people think. I mean, just a lot more than people think.

[00:27:18] Rajan:  It just radically changes the unit economics. But is there a objection that you get? What is the biggest objection that you get when customers come to you for doing development with AI?

[00:27:28] Jesse:  For some reason, human beings like to have human beings do stuff for them. I'm not entirely certain why that is. But I remember when I first started playing around with it myself and I started talking about it, I had this kid. This was back in the chat GPT two days. And one of his main jobs was writing these discovery documents that needed to be very detailed, 30 pages, tons and tons of work. It would take him weeks to do one at a time. And I noticed that his productivity all of a sudden increased, and his quality got a little bit better. And it was like, man, I need to give this kid a raise. Like, he's doing a phenomenal job. Well, what I had found out was that he had been behind my back and hiding it from me because he thought he'd get fired. He had been using large language models to start doing his job for him, and he had basically just became a prompt engineer. And by the time I talked to him about it, he hadn't produced a single document for me for probably three months, four months. The AI was doing all the work. And you know what's funny? When I first brought it up with him and I talked to him about it and asked him, and he was honest, he thought he was gonna get fired. And so me as a business owner, I see someone who's producing a higher quality product 10 times faster than he ever has. He had to quit work. Like, he went from working ten hours a day to working two hours a day because all of a sudden he had tons of time on his hands. But he was doing more for me than he ever had before, and he was just afraid he was gonna get fired. There's this mentality that us humans seem to have where we want people to produce, I guess, in some way, shape, or form, but the AI world's going to change that. And it was funny after I had that experience with him and a fair amount afterwards, I had a conversation with the company and said, listen. All of you guys have to start learning how to use AI to do your jobs better. Otherwise, I don't know that I have a place for you here. Like, if you can't augment your own talents and the things that you're best at and use AI to make those things better, then there's lots of other jobs where you get fired for doing it, so go work at those places. But here, we're going to use it. So oftentimes, clients will come in with the exact same mentality where they're like, well, I don't know if I want a computer writing my code. It's like, well, what if the computer's better at writing code than the humans that are on the team? Then do you want it writing your code? And so I would say that's probably the biggest objection, which is easy to overcome because I just tell people, listen, at the end of the day, you're paying for a result. It's very simple. Like, either you get the result that you want and you're happy to pay for it, or you don't you're unhappy to pay for it. That's gonna happen whether or not your developers are using AI or not. And the truth is just because of how AI works and if you have the right systems in place to monitor it and make sure that it's producing a high quality product, you're actually gonna be happier with what you get if you have an AI work with you because it's just gonna be faster, cheaper. You're gonna get more done. And then for the most part, business owners are like, yeah, it makes sense. I'll try it. But I'd say that's probably the biggest objection. People are just scared of computers doing things for them, at least the same kinds of things that humans do.

[00:30:14] Rajan:  Yeah. You've been working on this from early on. I know you said that, you know, every three months things change, but if I still ask you to say fast forward twelve months from now, what do you think is this gonna be, like, the biggest change that is gonna happen?

[00:30:29] Jesse:  Twelve months from now, I think that we're actually going to get a general intelligence that is able to, in a sense, mimic human reasoning. Because I think that the next twelve months is gonna be focused on reasoning models. Because the biggest problem with large language models up until, like, OpenAI's four o preview that actually has some reasoning capabilities was that AI couldn't reason. It was just literally a machine that had input output, and there was no way to think about its output. It had to get lucky, or you had to be really good. The intelligence went into the input and so that you could get intelligence on the output. And since a lot of people are lazy and they don't wanna put intelligence into the input, they would get an unintelligent output. Because all prompt engineering is being smart about your input tokens so that the AI can be smart about its output tokens. It's not rocket science. It makes perfect sense. But what is interesting is what happens like, a human being like, I I can tell you something nonsensically stupid that has a little bit of wisdom, maybe a little bit of knowledge in it. And you can take stupid input tokens, and you can actually produce really intelligent output tokens as a human. We call that reasoning. Like, that is the ability to get information, reason about that information, make the information more valuable in our own heads before we then output the response. I think in some ways, that's what an AGI really is, is the ability to take in any tokens and then through a process of reasoning, make the output tokens more valuable than the input tokens. Now we have something crazy because it is not really mimicking how human beings work, especially, you know, because that's what we do. I think in the next twelve months, that's gonna be the biggest focus. And when that happens, I don't I don't even know how to predict what happens when that happens. It'll be interesting.

[00:32:10] Rajan:  Sam, I'm gonna wrap our conversation with one last question. What would your advice be to someone who's a developer that has not looked into AI? What should he be doing? I mean,

[00:32:20] Jesse:  he should either be looking into AI or figuring out what else he wants to do for a job. What would you recommend? How exactly he should do that? I would say learn prompt engineering. When when I say that, learn to have intelligent conversations with AI so that they can create intelligent outputs because the most valuable people over the course of the next twelve months are gonna be people that can have really intelligent conversations, have intelligent token inputs into AI. Because the more intelligent your token inputs, the more intelligent your token outputs. And my very best developers, really the very best people that I have and my most valuable people to me are the ones that really know how to have conversations with AI that produce valuable token outputs because they're maximizing their ability to get things done. And so I really do think prompt engineering and agentic systems, like learning how to work with and build agentic systems, are the two most valuable skills that anyone in the developer community could have right now for sure. Awesome. Jesse, thank you

[00:33:16] Rajan:  so much for coming on to the show. It was such

[00:33:18] Jesse:  a fun chat. Thank you for having me. I really appreciated it. That's it

[00:33:23] Rajan:  for this episode of pivotal clarity. This is an Opeka podcast. Opeka is an accelerator for global Indian founders building AI software companies. We're exploring the fast changing world of AI together with our listeners. If you like this podcast, you can find more on our website and other popular podcast apps. Subscribe if you want to keep up.

‍