Welcome. Welcome. So, how many people here are from Ann Arbor or Michigan, or close to Michigan? Show of hands. Awesome. How many folks came from further out than the Ann Arbor area? That's awesome. So, welcome to Michigan. It's an absolute delight to have you spend some time with us, this 24 hours with us, to participate in innovation and building something on the absolute cutting edge.
But I wanted to talk a little bit about this Jaki story. Like, why is it happening? What's happening in the market right now that calls for a disruptive technology that allows us to build more effectively? Like, what's going on? So I wanted to start this talk. Actually, very recently I was giving a talk about AI and education, and he's actually in the audience right now. Raise your hand.
So one of the members of that, that just showed up at the talk, said, you know, I actually, I was like, I heard about this talk. I saw there was some TEDx talk that you gave like years and years ago. And he was like, after I saw the talk, I just had to see what the heck is going on. And I had to come hear you speak because during this talk I made a prediction. And this is in 2019.
And my prediction, and you can see a glimpse of it on the slide there, is that the way that conversational AI and AI should work, we have to up our game, right? We have to aspire for a human-level, human-in-the-room interaction experience. And this was at a time when we had Siri and Alexa. You remember, set a timer in 10 minutes or turn the lights on, type of rules-based AI, and that was what the market had. That's what Apple released to hundreds of millions of their customers. It didn't work. We didn't have that AI inflection moment.
And so I was talking about what we were working on to produce that moment. Of course, we didn't have as much money as OpenAI, so we weren't able to achieve it at the time. But I thought I could open with that. And then in 2021, I wrote a piece for Forbes magazine. And in this piece, I talked a bit about this idea, right, of how we should be building software in the AI era. And the idea, I called it at the time, collective intelligence, where we should see those systems really functioning like ants in an ant colony, where they would be modular and individual neural units working together to solve a problem.
Later in the article, I described these as robots or ants, and you know, they would have a modular system. They can be walkers walking around your data and then collaborating at the end to solve your problem. And that was the tenet of the Jaki idea. And then in 2022, we had the OpenAI ChatGPT moment, right? This changed the world. I've never seen a rapid adoption. The internet wasn't even like this when the internet came out. You know, it was kind of a rocky road. There was the dot-com bomb. But AI with this moment, it just is continuing to space rocket into the market.
And indeed, a very peculiar thing happened. Every single area of computer science was now reorganizing itself and re-understanding itself as an area for AI. So processors and computer architecture for AI, human-computer interaction for AI, programming languages for AI, everything for AI. All of the world's energy is working fiercely on this problem. And then a couple years later, there was the advent of agentic AI, right? These things really are manifestations of that insight from a couple years prior.
So what now? What's the new insight? Well, I've been talking about something since 2023, and it really comes with an understanding that for the market, the model problem is solved. Right? The most pressing challenge right now isn't getting the model, because we have the model and we're slowly improving that model. The hardware problem is solved, but I make the case that we haven't invented how we build software in this era. So we're going to be talking about that, but there's a couple interesting things I hope to achieve by the end of this talk.
Three things in particular. Number one, Nvidia, OpenAI, and Anthropic, they're already cooked. They're done. And they don't know it yet. So I'm going to make an argument for why I make that crazy claim. The way that we and AI, because the AI has the same tools that we have to build software, the way that we're building software is already dead, right? It's on borrowed time. And so hopefully I'll convince you of that by the end of this talk. And the last piece is there's a new era of how we can build software that's available today.
Right? So I want to leave you with these questions, these claims, and convince you or not convince you, it doesn't matter. I think these are the questions that you take with you as you move through the world moving forward. So let's start with a couple trends. Okay, there's actually really interesting trends. Everyone knows what's going on now, but it's actually hard to understand where things will go. Folks are waking up with crazy anxiety. There's token anxiety. I don't know if you've heard of this phrase. Has anyone heard of token anxiety yet? Yes, I have it all the time.
It's the anxiety that, oh my gosh, my agents aren't doing anything. I'm wasting my life, right? I need to keep the agents running. So there's all kinds of anxieties going on, and we don't know what's going to happen in a month. We don't know what's going to happen in a year. But there's three trends that I think we should pay attention to, and this is going to help us structure where we are now. Right? Let's re-understand where we are now. So one of the important trends is this is how many people understand agentic AI today. This is what people think is real, right? Oh, of course, I've got my coding agents, Claude Code, Codex, Copilot, co-work, agentic in-loop AI. That's how we're understanding it.
But there's something game-changing that happened this year, January, around January time. There's another category of AI that's even more disruptive, right? Man, I had a beautiful picture of Jensen, you know, Jensen Huang, Nvidia CEO, and he's got claw hands. You guys have seen this picture. I should put it in the slide. What am I doing? But OpenClaw came out, and I can't understate how much this has revolutionized the way we understand the role of AI in our society. And indeed, tenets of this release, out-of-loop agentic AI, is really what's causing massive market shifts.
You've got a $280 billion selloff in the software market. People are trying to figure out what's the role of software companies because this thing was building software for itself to help you solve your problem that you just texted it. And how it works is it's out of the loop. The human's out of the loop. It's your friend. It's your human in a computer that you can talk to, and it can use your computer to solve problems, write software for itself. Go right, that's the new generation, and everyone's going after it.
So we have a flurry of new claws. This out-of-loop AI, Hermes, looks really interesting. It's an AI that evolves itself and rewrites itself as it works for you, as it gets to know you, and it lives remotely, and you're out of the loop. You just chat with it like a human. So the world is still catching up to this moment, but this moment is already leaving us. And I could tell you inside knowledge inside the biggest companies on the planet, from Nvidia to Microsoft to Facebook to Google's, inside those companies, they're reorging themselves right now to build for that world.
There, projects are being canceled left and right. People are like, no, we're not going to work on GPUs anymore. We're going to work on APUs, agentic processing units, which look like regular processors that are really good at GPU because it turns out the agents love to GPU. So that's the workload of our time, right? So everything's changing with this, and we're just catching up. And you want to know what's interesting, guys? We're going to build stuff in this day.
I honestly think the easiest win in the market right now is to build an out-of-loop agentic AI for a specific purpose for non-technical people. If you do that, you will be a billion-dollar company. If you execute on that, you will be, because right now, who's trying to set up a claw for yourself? Was it easy?
Do you think your mom and dad can do it? Set up a claw. Okay, so whoever said yes, I got to meet your parents. They're amazing. But it's so you have to essentially be a software engineer. You have to understand what a curl command is. You have to work in a CLI. You have to know what a gateway is, and then you have to configure the bot in the client side by going through settings and a Discord bot. Hey, OAuth keys. You have to set all of that up to get a claw going.
Today, all of these claws are like that. There aren't apps in the app store that is a claw that solves one problem. Let's say, what's the problem the humans have? Figure out what my emails need to be, you know, in the next two days, right? Just that problem as an app, as a claw running in the cloud, massive unlock. But my point is this is changing the market now.
An interesting thing that happened when this came out is Mac minis were sold out across the country, right? There was a thesis that AI is going to live in the cloud. We're going to be hooking into tokens, and everything will be delivered like that. But there's a thirst in the market to bring this technology home and have it on your desk to control it. We want to control the AI. Another important trend,
Kimi K 2.5 came out this year, earlier this year, GLM5, MiniMax 2.5. I used to give this talk a year ago, or a similar talk a year ago, where I'm like, the open-weight models are catching up. I can't even say that anymore. They've already caught up. They're benchmarked against Opus 4.6, ChatGPT, Codex 5.4, and they're beating them, and they're free. They're MIT-licensed open-weight models. So we're already there.
The only reason we don't talk about it more is America doesn't like to talk about Chinese technology. That's like the only reason it's locked out of the enterprise. You know, we work with large enterprises. There's a mandate. You cannot touch a Chinese model anywhere on any server because those weights might be evil and steal our secrets, right? Because they don't understand technology. But the technology is there. And guess what? There's a company we all know that built something phenomenal with one of these models. Can anyone yell out the company that leveraged these models to build an amazing product?
Yes. Yes. Awesome. We're going to talk about that in a second. But before we talk about Cursor, there's something else that's really special. And this one's a doozy. If you listen to Apple in the market, what they're saying, their position and their strategy is to take AI out of the data center and put it on your desk. That's what their play is. They can't compete with Nvidia. They can't compete on AI. They can't compete when it comes to data center infrastructure. The only way they can compete is on your laptop and on your desktop. That's where they are.
So their strategy, their innovation is focused because of economic incentives. It's focused on getting you to buy their machines so you can have your own GPT on your desk. So the GPTs are good enough, the free ones. You can fine-tune it, keep all your secrets. You never have to go to an API. You never have to find an API key again. And at the end of this year, there's a product coming out. My prediction is it's going to do 100 tokens a second with Kimi K 2.5, and it's the studio, the Mac M5 Studio.
If you look into this, just Google it. The M5 is insane on the laptop. They've quadrupled and octupled AI capability. And so you can run very sophisticated 30 billion parameter weights, 120 billion parameter weights model on your laptop. And that's the sweet spot for capability. There's also a product called Tiny. So this is the size of an iPhone, and you plug it into any computer. When you plug it in, just like you know there's these hard drives that'll load software, and then you can use the software on the hard drive when you plug it in. It loads a ChatGPT window called Tiny AI window,
and you can talk to a 120 billion parameter model on this thing that fits in your pocket that's the size of an iPhone. This already exists. If you Google this, you're going to see it all over YouTube because they're doing a Kickstarter right now. They've already sent units out to all the YouTube streamers that's in the space, and they're talking about it. This is bonkers. It's $1,300 for a 120 billion parameter model running at like 15 to 20 tokens a second.
They launched on Kickstarter asking for $10,000. They're at like 2.5 million right now. The market wants this, right? They want to come out of the cloud. So I think Nvidia might have a problem on its hands, right? I think that whole thesis, Oracle just laid off 30,000 engineers or something. What's happening? I don't know. But these are interesting trends.
But one thing you could observe is that we want control, right? We want control, and we, the engineers amongst us, want to build. Now, there's a lot of discussion about, oh, well, our jobs are so screwed and everything's scary. You have to pop up. It is incredibly meritous, strategic, and smart to be a software engineer right now. You just have to understand how the field is changing. But it's the sickest field you could work in. Here's why.
These are the top market cap companies on the planet, right? So if these were countries, they would be like big, right? They'd be massive countries, right? You've got Nvidia at the top, of course. But if you look at Apple, Alphabet, Microsoft, Amazon, Meta, which is back in the top 10. Good. Good job, Meta. I don't know what you're doing because they fell out for a while, 16 or something. But anyway, those are software companies. The biggest companies in the world are software companies, right? Software is really the critical path of the flow of the oil that is money through our global economy. And that's going to stay.
People are still going to be buying stuff from screens. And that's, it's bigger than oil. It's bigger than healthcare. It's bigger than anything. We just have to understand how that field is going to continue to run. So we want to know how we build software into the future. And that's where we have a little problem. Okay, what's this problem?
Well, if you look at how we build software, the most popular languages on the planet is Python and JavaScript. We know this, right? All right, that's interesting. When was Python invented? 1989. Okay, that's interesting. When was Java and JavaScript invented? Early 1990s. So the stuff that we're using to build software today is from an era that looks nothing like the era we live in today. There's been a massive stagnation. This was before the internet, guys. There wasn't even internet. There were no cell phones back then, right? I don't know. Were laptops invented in '89?
No. Well, the computer came out in '87. Oh, no. No. '77 and '80. So it was like seven years after the PC came to the fore is when, you know, we started having Python. So we didn't even necessarily have laptops like we know them today. No tablets, but we're still using the same stuff to build software, and unfortunately so is the AI. And so that's led, how do we then build software for this new era? Well, what's emerged is the polyglot problem. So if you Google this, it's a very well-known problem.
To build an interesting piece of software, you got to, on the front end, you got a Next with your JavaScript, you know, API interface to a back end that's Python running FastAPI, and then you got your server SQL interface that you got to interact with. And then all of the startups on the planet are trying to solve one or two of those problems. Oh, Next.js tries to do front end and back end and the same thing, but you still got to go through the REST API. You're not type-checking across the API border, so you have to add validation code. And now there's a caching interface that you have to write by hand so that you don't hit API that you don't need to hit that are read-only, save the network traffic. That's what you're doing when you're building software.
How much of that has anything to do with the app you want to build? Zero. Right? So this is the landscape of the best tools you can use today to build software. If you're smart and savvy, you'll know all of this stuff because this represents our market across the different things you have to do to build an application. 44 plus languages, DSLs, and configuration languages available to you if you want to be state-of-the-art.
Now, you'll say, well, I don't have to use all of it at the same time. What's the minimal thing I can use? Well, if you want to build something simple, the simplest full-stack app, you still have to go across around six languages. Six different things that you and the AI has to do to solve a problem. And guess what? The more code you have to write, or the more tokens you have to generate, the more complexity you're dealing with. And complexity is bad. It's bad for you, the human, and it's bad for the AI.
All right. So what we've been working on, that's the story of the innovation that is Jaki. Now, languages are hard. That's why it hasn't been done in like 40 years, right? New programming stacks are really, really hard. And there's something that developers won't, there's many things developers won't tolerate. They're very skittish, right? Ah, I'm going to have to give up my favorite libraries. Oh, no. I can't, I don't have to learn all this. No.
Well, you see, the beautiful thing about this particular creation is it's interoperable perfectly with every single Python module and library in existence. It's interoperable and perfectly compatible with every single JavaScript, Node, npm package in existence. The compiler understands all of these languages and introduces abstractions to abstract away things developers shouldn't have to deal with. It also generates pure machine code, and it's one language.
Wow, that sounds kind of crazy. Well, it exists. And what does that give you? Well, this is what you have to do to do a to-do list today using the best stuff. SQLAlchemy, you guys know Python, of course. If you want to talk to AI, you got to use Python. You can't use Next.js, unfortunately. It's just not the ecosystem. React for the front end. You know, this is what a to-do list application looks like in the best languages you can do it today with the savviest stuff. I vibe-coded that. And then I vibe-coded this. And this is that same application, same functionality in Jaki.
Well, that's nice. And that's the innovation. There's also some other innovations. There's innovations for programming abstractions and constructs for agentic AI. If you want to write a prompted system today, your code looks like this. Prompts everywhere. In Jaki, there's language technology so that you can get better performance from the AI with less code. And you're leveraging types. Types. Models love types.
If I'm type-checking your front-end code and your back-end code in one language, that means there's no interface for the humans to deal with. It's like when we were doing assembly programming and you had to figure out where you're going to put your registers in memory so that your functions can find the data for the parameters, and humans had to do that. That's exactly what we do when we create a front-end back-end API, where what do you put where so that the back end could understand it and the front end.
Because this was solved in computer science, I think so. People are like, why now? Well, why didn't this happen for 30 years? I think it's largely learned helplessness. This is what we got to deal with, guys. I'm sorry, just try to make it a little better. Learned helplessness, and then you have to find folks willing to bleed to make it a reality. And that's what one special thing I found is a community of folks that are willing to bleed to make this real. So, and I should add a picture of them. It's so beautiful. Like, I'll share it on LinkedIn so you can see some of the team.
So today I'm going to be unveiling two things. It's kind of a secret, but a lot of people probably know this already. So this gets into how we're going to present ourselves to the world, right? How do we come to developers to say, I think we can help, right? Because it's a bold thing to do. So we created two things. So I'm going to share that with you now. But let me just share for the folks who are business-minded.
One thing that you can do when you design across the whole stack, this is vertical integration. This is what Apple does, is you can co-design, and that's important. We can co-design the language with the agent, with the model, with the tooling ecosystem. You can co-design it to help software do better. So I'm going to leave that nugget out there. But the first thing is a coding agent we call Jaki Coder, right? So this is a coding agent that functions much like Claude Code. It's a VS Code extension. It's in the marketplace right now.
Well, no way I'm messing up my future slide. It's a coding agent that can live into your VS Code extension, into your Cursor, etc. And it's actually pretty nice. You can actually understand what it's doing nicer. So it has a lot of niceness in it that feels better. I, you know, I'm a Claude Code ninja. You know what I'm saying? And so we got all the stuff to be nicer. No, this is very early. We're just launching it now. So it's alpha, right? Okay. So if you hit a bug, you know, that's how it is. But that's going to be okay.
And then the other thing, which is crazy, this thing is nuts, is Jaki Builder, right? Jaki Builder is essentially, you can think of it as like a Lovable for Jaki. And there's a lot I can go into about these, but I'm just going to, we're going to play with it at the end. We're going to play with it at the end of my talk. But Jaki Coder and Jaki Builder, when will it be available for folks to use that kind of thing? Well, we're going to make it available today to all of you guys. And it'll be available today and forever. Yes. YES.
And it'll be linked here. But let me talk about one more thing. I was going to demo now, but I think we're in a good flow. I'll demo at the end. Let's talk about one more thing. We were talking about those three statements. Oh, OpenAI and Anthropic is cooked. Let's talk about why they're cooked. They're cooked because they can't do what needs to be done, and they're going to have to bleed to get it done. And this is a principled understanding of the world.
Now, I started with, oh, you know, you talk about a few things, a few things, a few years later it's happening. So Anthropic and OpenAI, they don't know they need to control the language yet. They don't know that, but I think they're going to know it soon. So Anthropic acquired Bun. How many people in the room has used Bun before? Yep. So Anthropic owns Bun now. Bun's open source. I don't know why you spend hundreds of millions of dollars for something that's open source, but Bun's open source. They acquired it. Bun is a tool.
It's the tooling around JavaScript, right? Okay. Interesting. They want to control the tooling. They want to be able to co-design agent and tooling ecosystems. Interesting. What does OpenAI do? They acquired Astro. So there's some kind of war going on. They're acquiring open-source tools that everybody uses to get their code written, part of their software development stack. They're starting to prioritize acquiring those for hundreds of millions of dollars. They don't know it yet, but soon they're going to realize they want to own the whole stack.
Right now, what do they have today? They don't know yet. Maybe they'll know after the talk if they listen to me. But what do they have today? Okay, they have a model. Well, is that interesting? Is having the model valuable? They're dropping hundreds of billions of dollars to train these models. Well, let's see. Cursor just released Cursor 2. Composer 2. That's interesting. What's Composer 2? Composer 2 is their own model, right? Their own model. They trained it themselves. They fine-tuned it. They used all the data from you guys coding in Cursor to build this. And this is to compete with your Opus, your Anthropic, and your Codex models.
And it's really freaking good. And it's really good because they started with Kimi K 2.5. Kimi K 2.5 is that MIT-licensed model from China that nobody wants to accept is amazing in America. That's what Cursor used to create an agent, a model. Basically, the value prop of OpenAI and Anthropic, Cursor just did it with a Chinese model, and it's beating Opus 4.6, and it's almost at GPT 5.4, but it's way cheaper. It's blazing fast, right? It's that MoE goodness. And they released a paper on how they built it.
So what do you think is the first thing I did when I, once I saw this paper? This is like how we trained Composer. So the first thing... Oh, then somebody want to actually try to gander and guess the first thing. Go ahead. Download the PDF. Yes. Okay, that is true. You actually beat me, right? So the answer was download the PDF from this intelligent-looking guy in the front. Right. The first thing I did is I tried to reverse engineer how much it cost them.
Okay, because I was like, isn't it supposed to be expensive to train a model? To pre-train a model, it's expensive. To fine-tune a model, it's actually dirt cheap, and it's getting cheaper all the time. So from our analysis, it's about 100 to 150K they used to build their own model in their own environment that they controlled and named Composer 2. So what does OpenAI and Anthropic have, and what does Jaki have?
We have the open-source ecosystem and stack of their dreams, including the coding agent. We have the language. We have the tool ecosystem. We can co-design this in a way that can accelerate AI. Now, there's some studies we're doing right now, and we're at 11x less tokens to create the same app. 11x. I wish I had put the data in the slide. It's very early. 11x. So we just kind of got started on that problem, and we're 11x less tokens to build the same application, which isn't a surprise because you saw the code size difference, right? We're going to continue working, but Summerhead is in Jaki Coder, and so it's going to be interesting to see how that goes.
And so hopefully I achieved these three things. The first crazy statement was Nvidia and Anthropic and OpenAI are cooked, and they don't know it yet. That's my case. I'll leave you with the question. We don't have to answer the question, but this is food for thought. The way we and AI builds software is dead, and it needs to die because it's been since the 80s, since we've had something new, and we need something new, right? And then the next era of how we build starts today. All right, so that's the talk. Let me switch gears to the demo.
Hopefully you guys are convinced. Okay, so I'm just going to go to this. This is the Jaki Builder. Unless it's just a site, right? Oh, wait. No, you can't see now. Maybe I should mirror. Let me mirror. Let me mirror. Change to mirroring. Yep. All right, cool. So this is just jaki.org. It's very easy to install and get going. You just run this two stuff, and you'll have a full-stack app on your machine. You probably could do it right now, but don't do it right now. Just watch this demo.
So if you click here, this gets you to the Jaki Builder. This is the splash for Jaki Builder. And then let's just go, let me play with it. So let me first go to the dev one because the dev one is the siloed one for me to do this demo. It's the same exact stack. We just have two instances of it. So I think we could just, let me just... No, no, not with Google. Let's do GitHub. We're coders here representing. Okay.
All right, so we're in, and then so there's these templates and whatnot, but this is the platform, and you can talk to it to build something, and you can pull your code out, get it to GitHub, whatever you need to do. So let's just create a little thing here. Let's just call it demo. And this is going to take a minute, right? Because what's happening right now is we're spooling up an actual instance. Wow, that was faster than it was this morning. So we're spooling up an actual sandbox in the cloud for a full-stack application.
So we're going to run this. This is going to be the demo, the little starter code in this template. And one thing I want to emphasize is this is not a front end. This is going to be, it's installing here. You can see in the output. So Bun is installed in the sandbox, and it's loading, and it's probably doing Bun install right now of the packages and so forth. So this takes a minute. It's loading up a full-stack application, and it's an actual instance in the cloud.
And then I'm going to do something really fun when it loads. This is a slow part I was talking about. I'll do something really fun. Do you guys know that QR code generator? Okay, so this is a full-stack app. So there's something I want you guys to do. So I have this URL here. How do I do QR code generation? Here we go. Let's do this thingy. I don't know if this is the right. Is this a good one? Do you guys know? QR code generator. Don't try to sell me stuff. Accept all so this goes smoothly. Okay.
All right, so let me zoom in on that. So you guys just go there. Go there. Right. Oh, let me get this out of the way. I know. That's how I like to do it. Surreal. They're going to be using it in like an hour. So this thing better work, right? So go there and let me know if it loads. Yeah, cool. Feel free to create a fake account, like just A, and you just have to have four characters for the password. Create a fake account and go in there.
Actually, I'm going to do this part again with a better app. This is like just a starter code, but I don't know. Let me know if it doesn't work. If it doesn't work, it's fine. We'll fix it here. Okay. Cool. Good. It works for people. So now, wait. So what's happening now is you guys are all logging in to a sandbox of like an app that this is actually building right now, right? And so, okay, cool. But let's change it up. Let's say we've never done this before, so this is going to be interesting. Can you change the highlights in this app to blue? Let's see what happens.
So now we got Jaki Coder is about to come up, and sometimes it takes, you know how this goes, right? The clock, you know, Claude Code will be working, it's generating tokens, trying to solve your problem. Let me look at this. Let me look at that. All that goodness. So this should start running, and I trust that no one's actually trying to like break the app. Okay, good. I trust you guys. I can trust you guys. Okay, so the engine needs to change how much blue need to lock in. So what's happening here is looking at the app, stuffing agents, what the experience is like. You're like, I don't even know what the is doing. Oh, there it is. Blue. So reload, reload your website.
It's blue. It's blue, right? Did it hot reload? You didn't even have to reload it. Because this is hot reloading, right? All right, cool. Let's do something else fun. Let me just log in. Let's think what else you guys want to do. Let's see. Let's... Jason, Jason, sign up. That's my super secure password. Okay, I'm in. All right.
And so in this platform, and really I'm supposed to be demoing all of the like, oh, you can see this and you can see that. So you can see the code. It's really busy in the sandbox. There it is. So we have some code there. You can look at the pages it's doing. So let's ask it to do something really, really fun. So I don't know, can you add a feature where the user can add posts and it's saved to when they log out and log back in?
Okay, so we're going to add that feature. We still see there's some errors going on. But so sometimes what'll happen when the agent's working because we're hot reloading while it's working. So sometimes it'll be like everything's broken and then three seconds later your app is back, right? So that happens all the time because the agent's doing its thing. It's actually changing the files. It's actually writing code and we're hot reloading it. I know when you guys are coding you have the same experience. Go ahead.
So for in the sandbox, so there's sandbox and deploy. When you hit deploy, it deploys to Redis, MongoDB, scale-out storage layer. Inside the sandbox is just a little SQLite for your development, so you can kind of develop and try things and so forth. So you guys are all connected to a SQLite right now, but once you deploy, it'll actually, it's much more expensive. It'll spool out about like ten pods in a Kubernetes cluster, and those pods include, of course, the database backend, but it also includes Prometheus and all kinds of stuff.
Actually, let me show you guys that right now. So we've got, so this is actually, if you go to forward/docs, this is what you're going to see. So you're going to get to, oh, we have LM telemetry in here. Folks are building like AI apps in this thing. So of course, we did a hackathon like two days ago. We spooled up one hundred and fifty instances of this thing because we're battle testing it for you guys today. First time this is available in the world. So this is probably still coding. I don't know what it's doing, but that's Swagger. This is a graph.
This ends up being very useful if you're going to do some object spatial programming, which isn't required, but it's a feature that you can access. There's this really sick admin dashboard. And then of course you have Redoc for the API, but what you're going to learn later is how it's optional if you want to code your own APIs. So one of the nice things, and this is, it's still going. The agent's still working. But one of the nice things about this language is one of our principles is we take nothing away from the developer.
So remember that before with lots of code and then after with a little bit of code. You can actually generate lots of code in Jaki too, and it'll work, right? And you could use all the libraries, Axios or whatever, to implement your own backend. You can implement your own databases. We take nothing. We take no powers away from the dev, but we just give them tools that gives them wings. So let me log in here and see if this thing is still loading. This is a major feature. So it's working really hard. It's writing whole new pages here.
So we have a new page coming in, a JSX element. We have a progress. So we should be able to see the progress report. Where's progress? Edit progress. So here it's almost done. So we can kind of see how much it's working. Main doing stuff, whatever, delete button handling. But one other thing I want to emphasize here, right, this is unlike anything you've ever seen in Lovable or Base 44 because this is a full stack application.
There's a REST interface on the back end. There's database. All of this will be functional as you guys code. So see now it's actually running stuff. We also have an eval loop in here. So the eval loop is, it's using Agent Browser. If you haven't heard of Agent Browser, it's a phenomenal tool. It's using Agent Browser to try what it's building. So the agent is like, I built it. Let me click around. Let me add some users. Let me make sure. He did say he wanted it to be saved and the users to work. So it's doing all of that work. It's generating tokens.
But it'll ultimately be eleven x less tokens. So there's a lot of fun things we can do, right? But I will have to be conscious of time. You guys can go start logging in right now if you want and start coding something. The workflow that I think is the best that I recommend: start in Jaki Builder. Have some fun. Then just, it's integrated into GitHub. So then you could just create a GitHub branch or a repo. So if you see here, this has, you can connect it with your GitHub and then just create a branch.
And then you can pop into your comfortable environment, whether it's VS Code, Cursor, or whatever, anti-gravity, whatever you guys like. And then you just go to the app store, you install Jaki Coder, and you'll have the same agent in your favorite IDE after you prototype and build out as much as you want in this platform. And then you can go from there. And so there's like a whole list of features I didn't get to cover. You know, you ask around, we all know kind of what's going on. Lots of us knows about, there's PWA, progressive web apps.
So you can have a very nice PWA experience that'll run on mobile like native. There's this long list of features, right? So anyway, so that's the demo. Thank you guys all. Hopefully I at least opened up the aperture of curiosity for you guys as it relates to these questions. And with that, I'm happy to take questions.
