Ex-Apple dev reveals his Agentic Engineering Workflow transcript

It's about to get really easy to hack anyone. Pretty soon, we're going to have open- source models that don't have any alignment whatsoever. As smart as like Soul or as smart as Fable or Mythos or whatever. Just think of like every little project that everyone's done online and talked about. I have it all unified into a central system. This is like everything I own. This is like everything I'm shipping and how fast I'm shipping.

Uh this is like my attack surface system for everything I have deployed for attacking against it. This is my centralized uh memory system. These are all my inputs that come in from anywhere. So if I were to like bookmark one of your posts on Twitter, it would show up here and get parsed and put into Cortex like very quickly. >> So here's an example of one that just came in and this is now in the memory system and all I did was bookmark it on Twitter.

You need full contextual understanding of your business like that. That's what you need to have as any company. This is why some companies were good at adopting AI and some are still like barely even starting. They weren't able to describe how the company works to the AI. >> Yeah. >> To have it take over for them. If you don't have that context yourself, you can't even start an AI project. >> All right, Daniel. So, you've been in cyber security for over 25 years.

You've done security at Apple and companies of this magnitude. What do you see of Asians impacting security? We've done a crazy hugging face incident like few months back. Where do you see this headed? >> Uh I think it's going to a pretty bad place actually. It's basically going to be attacker AI against defender AI. >> Yeah. >> And uh that's the whole game is how good is your AI stack? How thorough is it? How fast is it?

How intelligent are the models that you're using? And uh that's just pretty much the game. And if you're a defender and you're not doing this stuff, then you're just going to get beat. So is that just going to be one of the main costs of developing software is like how many tokens you use to try to hack yourself? >> Yeah, I think you have to basically have a full harness and also a full sort of model stack and the combination of the two.

I think it'll head more towards the models and the harness will become a little bit less important but I think the harness will still be super important for it and uh yeah every company will have to have it. So do you think like practically right people most people don't have a billion dollar company and they cannot get like unrestricted access to mythos or openly asset model. So like what should they do? Should they like download obliterated models and try to hack themselves to you know prevent other people hacking them?

Like how would you go about this if you were a typical person with a normal budget and like normal technical experience? Yeah, I I think it's using a combination of the uh stuff from the labs um you know open AI or anthropic um and then yeah using also a combination of obliterated uh open source models as well and ideally that would move towards more internal stuff because it it's never a good thing when your whole company kind of depends on one or two companies. um especially when they can kind of see everything.

Um I think Alex Karp was talking about that like you need to get your model stuff um brought internal. You you've got to be able to do this stuff inside the company. >> Yeah. >> And not have these crazy dependencies otherwise you're going to be screwed uh in a few months. >> By the way, a lot of you have asked me what user interface I use where I can use any model from any harness whether it's CEX, cloud code, PI agent, cursor or anything else.

And this is actually Cloudroom GUI. So Cloudroom is a new project I'm working on and it has two parts. The GUI is fully open source, completely free. It's built on top of BB which is also open source and you can actually get it right now. So if you go to cloudroom.dev, you can get the GUI and start using it right now with your favorite harness, your favorite model using the subscriptions you're already paying for. The second part is the Cloudroom Cloud.

So basically each agent getting its own isolated sandbox. This is still invite only. It's in the early beta. I'm still working on it. And if you want to be part of that, if you want to be one of the first people who have access, again, go to cloudroom.dev and click this button right here to join the weight list and to apply. Now, everything the whole philosophy of Cloud Room is that it's open source, right? The GUI is fully open source as well as the core.

In fact, you can go here to GitHub and you can see the Cloud Room core for yourself. We're already at 154 stars and you can see the architecture, go through the code, improve the security, do a pull request, do a file an issue, whatever you want to do. You can fork it and build your own. You can self-host it if you want. This is the first world's first cloud agents that are open source. Everything else on the market, whether it's Codex, AM, Devon, Cursor, is fully closed.

Cloudroom is the only and the world's first platform that is fully open source for cloud agents. So you can get started with the GUI right now. Again, very nice graphical user interface. If you're using codeex or cursor, you're already familiar with this. Like all of the main features are here. And again, the main advantage is that you can use any model you want and just chat with it um like you are used to. That is completely free.

It's also fully open source. So feel free to get that. And uh yeah, just go to cloudroom.dev and click on download the gy and get the latest version. So okay, screwed. I think it's we need to unpack the statement screwed, right? because it's not just like reliance on the labs and crazy token spend stuff like that and cyber security. It's also like leaking IP, right? Like people don't realize that the zero data retention it's not really zero.

There's like so many ways and also like people always say like okay they don't train on my data and they don't sell it to third parties and they think that's when it ends. But there's so many other things they can do, right? They can like look at it. They can make product decisions based on the user data. They can do so many other things with the data. So I guess let's unpack that, right? like how should a business that has some IP think about it because still it's it's it's not that easy to like self-host. >> Yeah.

The the dependence problem is um it's got multiple angles. The big thing that's super dangerous in my mind and what Alex was talking about was the smarter the stuff gets and like you have churn inside the company with humans, right? You have like Carol's been there for 18 years >> and like Carol just knows how everything works, right? So Claude comes in or OpenAI comes in or whatever and it downloads all Carol's information and Carol leaves.

Well, now Alex also leaves and Raj also leaves and pretty soon claude or whatever um you know closed source model or whatever it is the thing that understands the business the most right. So even like the leadership or a new employer or whatever they don't know where everything is they don't know where all the bodies are. They don't know where all the dependencies are, how anything works. And you know, we've seen this with humans before.

Um, the Phoenix Project was this book written by a buddy of mine was about this one person who was like the center of the company, everyone uh relied on them. And like if they just like get hit by a bus or they just leave or they decide to retire, the company is screwed. Well, in this case, all that is moving off of a human and moving into a model. Um, so it's like the prospect of cancelling Enthropic or OpenAI or whatever company when the whole business runs on it, you really can't.

It it's become the operating system for your entire business. And I think that level of dependency is super scary >> for sure. So what's the what's the like biggest bottleneck? I guess let's let's see like a business, you know, again, not enterprise. I'm I'm thinking like maybe 20, 50 people, 100 people. Like do they should they buy their own hardware? How would you advise them to like get moving in the right direction? >> Yeah, I think people are going to need their own hardware um but they're also going to need like the full thing that um like Alex sells and the thing a lot a lot of us are building.

Um I've got an app um that I'm doing for companies uh called Vector and it's basically doing the same thing. You need full contextual understanding of your business like that. That's what you need to have as any company. And you're supposed to have that like preai as well, but that's like very few companies had that. But you need to fully understand your business. You need to know how everything works. You need to be able to describe how everything works.

And then um kind of a related topic, it's like this is why some companies were good at adopting AI and some are still like barely even starting because um they weren't able to describe how the company works to the AI. >> Yeah. >> To have it take over for them. Whereas, if you go to like the some of these super advanced banks, they have every single little thing fully documented and any little change that happens to the business, they're able to like see it and monitor it and like there was a decision that went into it, which means they could take that whole thing and give it to an AI and be like, "Yeah, here's how the company works." And the AI is like, "Okay, cool.

I can help you with this. I can help you with this. This one still needs a human in the loop." But like if you can't if you don't have that context yourself, you can't even start an AI project. Um yeah, I did a post a while back about that. It's just like most companies aren't even slightly ready for AI. Uh it's not even a question of is the AI dangerous. What's dangerous is the fact that they don't understand their business. >> So what okay in that case what's the first step like creating some SOPs, some processes like >> Yeah.

Yeah. they they basically have to fully articulate their business. Um they have to find contact stores. They have to collect all the information about like how everything works. They basically have to build out these context domains for everything in the company. Uh build their SOPs, build their sort of work streams and workflows, which of course will be mostly human-based. Um but once that's all articulated, now they're in a more like sort of safe space because now they can start saying, "Okay, what models can we build?" Um should we, you know, uh most people aren't going to like stand up a whole GPU farm? >> Yeah. >> And you know, do that.

But there's a middle step there. A middle step would be renting GPUs not from the company that you're actually getting your AI from >> from sort of breaking Yeah. Yeah. Exactly. So, some other cloud provider, you're renting your GPUs, it's dedicated hardware, um, and you're doing all your local inference or whatever, semi-local inference, it's happening in the cloud, but it's not one company that just has everything of your company. >> So, de-risking basically. >> Yeah. >> Okay.

And I guess uh when it comes to the you know typical person watching this like you know people without even companies how should they think about their data and like them protecting themselves online >> uh like a typical typical company. Okay. Yeah. So >> no let's say individual like not a company just individual right like knowing that there's open source models open weight models that are obliterated that like you know these companies have like internally like crazy powerful models and you know obviously enthropic or openi will probably not hack an individual but you know we've seen what happened with the hugging face incidents where they can lose control but like more practically you know if you're an individual you're building some software you have your personal data on the web on some you know Google drive whatever how would you be advising people to think in this completely new era of cyber security.

The most common mistake I see new founders make is not enough feedback from real users. Sure, with AI you can build software much faster, but the faster you build, the more important real customer data becomes. After all, you need to know whether you're building the right thing. That's why I want to tell you about today's sponsor, Post Hog. Post Hog shows you how people are actually using your product like where they get stuck, which features they are using and whether they are coming back.

And with their session replay, you can even see what the users actually did. Y Combinator, Superbase, Fireworks AI, 11 Labs, Railway, all of these companies use Post Hog. In fact, Superbase used Posh Hog to spot a wave of new users coming from AI app builders and based on that data pursued partnerships with those companies. As one marketing expert at Superbase put it, Pose Hog has literally helped us get 10x more weekly new users than we did a year ago.

When you're shipping new features every day, your feedback loop needs to keep up. And that's exactly the problem Post Hook solves. So visit go.postth.com/david posth.com/david and learn how Posh Hawk can help you understand your users and build better products. That's go.post.com/david. The first link below the video. >> So, I recommend people do something like this. Like companies need to do this. Uh people need to do this.

This is an extreme version cuz I'm a psycho. But um this is called bunker. So basically, you know how we've all been uh basically putting out projects since like 2023, >> trying things, putting out new technologies. Well, so I don't have anything out there that I don't understand. So basically any as soon as I build anything, if I put it online, it goes into this thing I call bunker that has automated checks running against it.

So step one of this is actually you need to have agents that are continuously like mining uh because you asked about a person. >> Hey, what does the internet know about me? >> Yeah, >> right. So scour the internet, look at all the different uh data broker services. Figure that out. Maybe use a service like um I forgot the other one, but one is called like delete me. >> Yeah, one is in >> right. If you care about >> Yeah.

Incogn. Yeah. these types of companies >> and say, "What do I have out there?" Um, maybe you have some post, you know, uh, you're a 38-year-old woman, but you said something crazy online when you were 19. Like maybe go clean up your social media, clean up your websites, like just like basically understand how the world sees you because um, one of the things I've been talking about is like it's about to get really easy to hack anyone. um you you basically say, especially with like an obliterated open source model, um you could just be like, "Hey, uh David is getting more views than me.

Um I really don't like him." Um go into his background and find everything that could like hurt him and like we're going to do a Twitter campaign and pretty soon we're going to have open source models that don't have any alignment whatsoever, which are like as smart as like Soul or as smart as Fable or Mythos or whatever. And then of course even smarter than that and it will take that command in the same way that you told it to go make you coffee or find the best coffees. >> Yeah, that's logical. >> Uh yeah, it's just like hey whatever.

That was a request. Getting the best coffee sourced so he can make coffee at home. That's a request. trying to kill the career of some, you know, uh, YouTuber is also just a request and it will simply take whatever uh, tokens it has and spend them on trying to do that task and and I think we need to realize that and figure out what our attack surface is. >> I see. All right. So, I guess can you show us your agentic engineering setup, how you work, which models and harnesses you're running right now? >> Yeah.

Yeah. So if I just switch over. So I um I've got my own system. It's called uh LifeOS. Yeah. So this is basically um I had a thing called Pi uh long time ago. I started this in like a few months after Cloud Code came out around uh July >> and it was called PI P AI personal AI infrastructure. >> And um I recently changed it to life OS because it's just so much bigger. But basically the idea for LifeOS is rather than being a coding harness um basically a general harness. >> Okay. >> And so the the concepts are I think pretty powerful.

It there there's a bunch of them. They all sort of work together but kind of the main one is uh general hill climbing towards what I call euphoric surprise. So the idea is the prime directive for the system is um we're it's trying to achieve euphoric surprise in anything that we do any task that we give it, right? Um and when we give it a task, it's trying to move from current state to ideal state. Um so basically it's based on massive amount of data that we bring in.

So um inside of my system I basically have uh my full tilos in here which is um mission goals, strategies, narratives, challenges um and I basically have everything in my life is all inside of my life OS instance. So when I say that I want to do something I've immediately reduced the amount of context that I need to provide to the system. If I say, "Hey, make me an awesome website or make me a new blog post." It doesn't have to ask me questions because I've got something like 150 skills.

Um, these are all the 56 like uh public skills, but if I go into my uh system here, yeah, this is how many skills I have. Um, and these underscored ones are all private skills that are highly customized just to me. Mhm. >> And the um the regular ones are the public ones inside of LifeOS. But these are all very small individual skills that just uh kind of do things uh really well, but they're very light and they describe instead of describing exactly how to do something.

It's mostly describing what the ideal state looks like for me, my preferences of what perfect looks like for that particular task. >> I see. And that's that's sort of moving towards like bitter lesson engineering of like getting out of the way of the model. >> So okay, one follow up on that. Do you think this is going to be one of the key skills that like humans need to get better at like translating their preferences, judgment, decision-m into you know markdown files or skills so that the models can decide like them. >> Yeah.

Yeah. So so think of it this way. This is the way I I try to engineer everything. I try to think not where we are and how can we move outward from it, but rather um what is the tech going to look like for sure uh in the future, right? And that that seems pretty uh seems pretty arrogant to think about, but I think science fiction has already told us exactly what this looks like, right? If you if you look at like Jarvis, you look at her uh with Walkeen Phoenix and Scarlett Johansson, what what did he do to on board?

Okay. He he booted up the system. She went and looked at his whole life. She went and looked at his whole life, ingested everything and was like, "Hey, what's going on? You know, what do you want to work on?" And um I've had this idea for a very long time. I I wrote a book about this in 2016, basically saying that uh the interface between us and the world is about to change because we're going to be talking to our agent and our agent's going to be doing things against APIs. >> Yeah. >> Um and that is exactly the the kind of the whole concept here.

We we already know that's kind of the future that you talk to your agent and your agent does stuff for us. Like science fiction has taught us that, right? Well, the question is how does it know what to do? And the answer is it has to know everything about us. So, that is the central premise of LifeOS. Um, it's got an interview system. It uh, you know, you talk about your work if you want to. I have all my stuff sort of merged.

Um, I could actually show you a bunch of stuff which we're going to have to edit. Anyway, I I'm just going to go through some of these. So this is like my uh tilos and this is basically got my current state of my life, ideal state of life, um health. I've got all the problems, the world problems I'm trying to work on. Yeah. So it's like here's what you're trying to do. Here's where you're going. Um here are the things that you're having trouble with and here are the strategies that you need to be doing.

Right. And um this is kind of the baseline for everything because my agent Kai, my digital assistant Kai basically knows about this and knows exactly what I'm trying to accomplish. So when I give it tasks, it's always aware of all this stuff. >> So this is like API against the knowledge basically, right? If the agent needs to get specific knowledge about you or how you think about something. >> Yeah. So this is all local inside of my local AI directory.

This is all serving off of local host. Um, this is like everything I own. This is like a share system. This is everything in my house. Um, let's see. This is like everything I'm shipping and how fast I'm shipping. Just just think of like every little project that everyone's done online and talked about. I have it all unified into a central system >> where every single request, it is available to look at all of it. >> Uh, this is like my attack surface system for everything I have deployed for attacking against it.

This is my centralized uh memory system which is a combination of like um uh Carpathy's um uh wiki loomm combined with like honcho features and a bunch of stuff like that. Um these are all my inputs that come in from anywhere. So if I were to like um bookmark one of your posts on Twitter, it would show up here and get parsed and put into Cortex like very quickly. So here's an example of one that just came in. Um, and this is now in the memory system and all I did was bookmark it on Twitter, right?

So basically anything that I'm capturing and doing is flowing into my system to be used in some sort of way. >> Yeah. Basically trying to capture all the context that you're capturing and have it ready for your agents as well because this is a real challenge. I was thinking about this because you know we see a lot of posts. It's it's not like we speak everything, right? It's already good to have like some notetaker on a Google Meet if you're taking a Google Meet or if you're recording like a important conversation with your team, you can just do Apple voice memos, but like there's so much context and stuff that like is not easily tokenized, but this is definitely a step in the right direction. >> Yeah, absolutely.

And I also wear uh this thing here, which I don't have turned on right now, but this is the uh B computer. >> Uh let me turn it on real quick. Yeah. So, this is the B computer. And so, um, oftentimes I'll go walking with, uh, a friend or I'll just walk by myself and I'll just talk. >> Yeah. >> And as I'm talking, this thing is picking it up. And then, so, watch this. This is crazy. Um, give me a short summary of um, recent conversations that I've had uh, from my life log.

Just the headline uh, not any detail. So I don't have to say, "Hey, by the way, there's an API. Hey, by the way, I have a device." So I this skill right here, you see it's got an underscore in capitals. That's because it's private. And um it is now going to pull the um the content from my life log, which is an API pool. And it's going to tell me these conversations that uh that that I I was talking about, basically topics.

Yeah. Here we go. Oh, look at that. So, it's already caught up. It's already talking about stuff that we've talked about. >> Yeah. Nice. >> So, so, so, and here's the highle concept, which it it seemed like you just immediately went to this. Think about anything that you are doing throughout your life where there is value in some sort of way. You just told me something interesting. I I just had came up with an idea when I was in the car.

I don't have anything that ever hits the floor. It never drops. It never gets disappeared. It's always being captured. But it's not going somewhere which I need to go find and bring back in. It's already coming into life OS. >> That's great. Yeah, I think this is going to be like a real skill of capturing the right context so that because like you know I I believe models will get better, harnesses will get better, agents will get better and I think soon enough agents will be doing most of the tasks, right?

Like clicking around, filling out some forms. This just feels so primitive right now and it's going to feel like even more primitive in six months. You know, computer use is only going to get better. So, I I'm a big believer in that like we will just be moving higher and higher level up, right? We might not know what that looks like. It's probably like setting up agents, orchestrating them, choosing which model to use where, but over time it's going to be even higher level things like it's really going to be an art of like how to capture context and how to enter the right context to agents.

And yeah, I mean just having it is the first step, right? Because there's many companies it was especially like 24 25 but even this year many companies want to get into AI but then you like talk to them and you realize like they don't even have the right data they don't even collect the right data. So like what people can start doing like you said you know wear a microphone or even just like with my phone if we walk with my team and we're talking about like some issue like how can we improve our hiring instantly we have this instinct like somebody turns on the you know Apple voice memos and we're transcribing so like we have this conversation recorded so then when you go to computer you don't have to like do another voice prompt and talk for 5 minutes you just upload the voice file and you have all the context.

Yeah, that's exactly right. And and that could all just be happening automatically as well. So, let me let me show you this. So, this is crazy. Um my work tab, this is a private repo that all my different agents um humans and uh digital have access to. This is autoparssing everything and producing GitHub work items. >> So, this is like things that we should probably do a project on. And so at any point I could just say um what um are our top prior priority work items that we have uh in GitHub.

So is do you think that's a good unit of like measurement of what to do because you know some people like use Jira linear to-d doist some people use GitHub issues GitHub PRS like it's kind of hard to know what should be like a like a GitHub issue and what should be like a instant promp to an agent to just fix it. >> Yeah. Yeah. And absolutely and some of these um you know they have the ability to go in and autodo but um you got to be a little bit careful with that right you don't know like you don't want to have a conversation about something where you're talking about offensive security and like it thinks that's a thing that you want to do and it goes off and starts doing it but the the idea is that I didn't do an extra step.

The stuff is auto being parsed. Um so I I read this book a very long time ago um uh getting things done and the idea was never trust your brain for anything to always write it down. Yeah. >> So for like like 25 years I've been walking around with a space pen and index cards. >> So I've been doing this like forever and now I have the ability to just do it automated through all this uh automation. >> Yeah. Yeah, I mean people always do like this is I every new person I hire I immediately tell them like write stuff down you know if you need something from somebody tell them to set a timer like human brain is great a lot of things but it's terrible at being an accurate form of recall right like in a court of law they don't trust like eyewitness eyewitnesses and memory there's been like many CIA you know experiments where they just like convince somebody that you've been there you've seen it and like after a 3-hour interrogation that person can swear that they've been there so memory is like really bad And yes, people still try to like do things that clearly it's better to like have it in a text file or or you know GitHub issue or whatever.

And this is like the beauty I feel like we are we need to learn over the next 12 months is like what makes human human right and not try to like do the things that computers either like just code or models can do better than us and just do the things that we can do better. >> Yeah. No, absolutely. And and a lot of that starts with just knowing exactly who we are and what we want. Which is why I always start with this tilos thing which is like what are you actually trying to accomplish? >> Because as as work starts being taken away from us, we have to have a direction and we have to be pushing towards that direction or we're just going to have a lack of meaning. >> And so we have to have big problems that are outside of us that are difficult to accomplish and be heading in that direction.

So my number one thing is um trying to get people transitioned from the old world which I call like human 2.0 which is like corporate work into like what you and I do which is like we literally have our own ideas. We put it out on the internet. We hope it works and we hope there's a way to make money from it. Turns out they're sponsors. So there's like ways to make money but we literally wake up saying what do we want to do in the world?

And so my goal is to basically transition as many people as possible into that world. But that requires that they are able to articulate what they actually want to do. >> Yeah. >> And so many people have been trained for like hundreds or thousands of years. They're literally trained go to school so that you can be good at doing Mrs. Johnson's slides. >> Yeah. >> And the best thing you could possibly achieve is if Mrs.

M. Johnson is like, "Hey, these slides are really good. I'm going to give you a 1% raise. >> Yeah. After three years. >> Yeah. After three years. And it's like, no, that is not the way we're supposed to be living. And the more and more it it's a really rigged system because not only are people being told that's how to live, but at the same time, those jobs are being taken away. >> Yeah. >> So, we're about to hit a major crisis. >> Yeah.

I mean, for my generation, it's even crazier because I literally know multiple people that like they see AI, you know, they understand it's like superhuman at coding and it's getting better fast and they still choose to go to college for four years to like pursue the same degree. I'm like I I even like people in translation, I'm like, "Guys, have you used JGBT voice?" >> Right. Like what was happening here? Like >> Yeah.

So, I guess >> and then you talk >> No, go ahead. >> No, no, you go. I was just going to say I I I I know some people who um I I've got friends. Let me tell you how crazy it is. Just like here in the Bay Area, uh I went to a hackathon uh some time ago and um the kid that won was a 9-year-old sitting next to me >> them >> and um he's like, "Hey, listen. I just want to start by apologizing cuz like my front end wasn't fully finished.

Um but this is a fully working application. It does the following things." And then I go talk to uh somebody who I'm close to in Pleasanton, which is only about like 40 miles away or something. And I'm like, "Hey, what are you doing? What are you doing uh with your kids and stuff like around AI?" And he's like, "Oh, you mean the chatbt thing?" No, I thought that blew over. >> Like I tried it a couple times like a year and a half ago and I think I think it went away.

And I'm like, a nine-year-old is apologizing because the front end isn't fully done versus this guy in his 40s thinks JT GBT was from two years ago and it blew over. It's like they live 40 miles away from each other in the Bay Area. >> But almost spectrum is >> Yeah. So, but like like Okay. So, when I hear that, what my first thought is like the gap is about to get greater. Like people always say like, "Oh, AI for everybody this and that." But listen, like whoever understands how to control these agents and has the resources to just burn billions of tokens, like that person is not going to be equal if somebody who's using the free plan on on Google AI search.

Yeah. Yeah. Absolutely. And increasingly the the more and more abstracted that we can be, the more and more we're talking to our agent, um the more and more it depends on how big your goals are, how big your ambitions are, how clearly you can articulate what you want. And you have to here's the the immediate step that's so scary to me. You have to actually want things. M >> so there there are other people that I talk to which this is is the scariest scenario I've ever seen which is >> oh my god look what I could do and I show them all the stuff that I could do >> and I say what would you do with all this power and they're like I don't know >> but I feel like that's most people though >> can can it show me like a better show to watch on Netflix and I'm like yeah I can but like no but you could do way more like think bigger and they're just like they don't have anything that they want to do and I don't necessarily blame them.

I blame like the whole system, all of society for like training people to not be ambitious. >> Yeah. I mean, for sure, like this is the the classic, you know, everybody understands it that every kid is an artist and every kid is like kind of unique and then you go into like this system that kind of formalizes everybody into like these rigid paths and it's not like okay, you have a talent for programming, 80% of your lessons in school are going to be programming and 20% some English and basic stuff, right? or you have a talent for playing the piano, 80% of your stuff is going to be it's it's like the same, you know, even if you're like super great at poetry or or or whatever, it's going to be like the same split of lessons and then okay, you're going to the same type of colleges like okay, the colleges are a bit more specialized, but still like it needs a whole refactoring.

But we know that this is not going to move fast or like you know AI moves super fast. It's increasing a speed but societies are super slow, especially like the education system changing or the government institutions. So like what's the solution because this not going to wait the the like the improvements are happening. >> Yeah. I mean my solution is to try to expose as many people as possible to to reawake what they might have lost when they were a kid.

Um so one of my favorite uh authors is uh real um I forget how to pronounce it. I think it's German. But um he he basically said that creativity comes you can remember your creativity when you were a kid. When we were kids, we're like stomping around. We're like, "This is my fort. Those are the enemies over there." We're inventing all these things, right? And we we're like, "Yeah, I'm going to be a ballerina astronaut." And you know, we have all these potentials, right? >> And then slowly our peers and our parents and our friends and our teachers tell us that's not realistic and kind of like shuts that down.

Yeah. And >> so my my sort of goal is to basically say look, you have to remember, you have to focus inward. You have to ignite something internally. >> Um and basically pursue that. So I've got three sort of practical things that I tell people. You have to deeply understand the world like at a technical level all the way. This is kind of my prescription uh given how crazy things are. Deeply understand the world from like Adams all the way to like psychology and politics. >> Okay.

Right. So that's the that's the education deep full stack understanding of reality. Right. And the second one is you have to want something which in which is you have to deeply introspect and understand what you want and who you are. And the third one is deeply learn AI and get really really good at it. >> So step two is basically um how should the world be different? Does it not have something that it should? Do the things that exist, should they change?

So the first one is like understanding how the world is. Second one is like it should be different. And the third one is how to make it different. And if you don't have any one of these three, you're going to be in bad shape. >> Yeah. I mean like the way I the first one I guess you know the easiest explanation is like Elon Musk or Jeff Bezos even if they didn't have any name they could build a billion dollar company from scratch in like six months right no name no head start no money because they understand reality more clearly than us this is a I mean how would you like improve people on that right like do you tell them like read a wide variety of books or listen to different podcasts like >> Yeah yeah yeah I've got like a book list that that takes them through like a stack and everything.

Yeah. So, it's it's biology, it's u physics, it's stuff like that. And obviously, their interests are going to take them more in some directions, but I don't really care. It's like >> it's going to be different for everyone, but it's like I don't want them to have like these massive gaps >> um that cause them to be surprised by things. Um >> and and the cool thing about reading is um I try to read like like 20 to 50 books a year. um usually around 40 or so and it's like if I stop reading I stop being smart because the the the reading is like particles hitting my nucleus right and ideas are exploding off maybe not even related to the topic of the book but it's constant exposure so step one is like you learn about the world and then you see uh why doesn't this exist or why is it like that that doesn't make any sense someone should make a thing and as soon as it's doing that well now their tilos is starting to form now their internal thing is sort of activating and then step three is let's make it happen with AI yeah I think this this is a really good insight that like people need to pursue these kind of unique unique directions because you know if you have even if we like go two three four years in the future if you have a swarm of thousand agents and they can really do anything and then obviously humanoid robots as well it's going to be all about like somebody has a clever idea of like what to pursue and like other people didn't realize that and still at least with the current architecture of the models they lack this type of creativity they they lack this type of thing of like what to do right like I don't know what's your thoughts do you think this is going to be like solved with newer architectures or even the transform architecture you think this is solvable that like the AI has this intent of like having the right practical idea of what to pursue or do you think this is a unique human thing >> yeah great question um I don't think it can get there anytime soon right now with the current architecture because um our desires, this is kind of a a deep thing that I I think about is um everything we want is actually because evolution made us want it.

Um we're sitting on top of evolution. We're basically a mech suit um wi with evolution on the inside pulling the levers and like pressing the pedals. So, for example, like when we get dopamine or we get a a chemical that makes us feel sad or whatever, that's because evolution is saying you're not doing things that I want you to do. >> You're not struggling through a difficult thing and achieving something. Therefore, you get no dopamine hit, right?

So, it's controlling these dopamine hits and controlling like, you know, what feels good to achieve. Like, this is why it feels good to have kids and see them go to college and thrive and do all these things, right? These are all like squirts that evolution is is giving us. Um well AI doesn't really have that. Um it it has the neural net and everything and it can accomplish tasks but it doesn't by its nature want something which we do.

So I just don't see an easy bridge for that. I I suppose we could kind of bake it in with desires but um that would be a hack and it would also be kind of dangerous. >> Yeah. Yeah. Yeah, I mean that goes into like alignment and whole >> alignment. Yeah. >> Whole like rough topics. So like practically speaking, right? Becoming more technical. You mentioned the full stack knowledge from biology to like psychology, politics, but like let's focus on specifically like technical computer, you know, science stuff.

Look, there's many people who are on both sides of the spectrum. Where where are you? So, do you think like people should like be more technical or do you think like people are like, "Okay, the models can handle more. As long as you can just ask the model, it's fine." >> No, I I think people need to have the base. I think people really do need to have the base. Um I don't think they need to stay in the base and always understand everything, but um when when you're disconnected from reality, it's just I it's going to take you in a bad place because it it goes back to what we were talking about with the company.

If you don't understand how the company works, >> you just lose control. >> Um, so you have to stay tethered to reality in some sort of way >> and you can still be using your agent and the agent's doing all the work and everything, but the separation will cause problems. It's I I think that's pretty much uh guaranteed. >> Yeah, because if you don't even know like you know where is this agent running like is he on your computer?

Is he on a cloud device? You just lack stuff. And I mean this is a core limitation to any complex system right like understanding how it works like software is I guess a clear example because it's the most obvious and it's it can get like insanely complex but like you cannot you know run a country if you don't understand your ministries and you know the regions of the country. >> Yeah. And the other thing is like when your interface is the computer right just like um her or Jarvis or whatever.

Imagine you're only talking to them and they're the one picking all the stuff that you do in your life. They're picking your news and stuff. Well, at some point like you're your opinions start to change. They start to move in a direction >> and like you have to ask yourself, why do I believe that? >> What are the opposing viewpoints? Has my agent been steering me in a direction? Oh, and it turns out that um nine months ago we accepted a sponsorship with whatever Evil Corp and Evil Corp has been feeding, you know, prompts and system prompts into my agent and sort of steering me in a direction. >> So, if I've broken away from reality and I'm only accepting my reality from my one interface, >> again, I I've separated from reality. >> I see.

It's it's crazy like how many people disagree with this, but like do you think we'll see maybe let's not put a time frame on it, but soon enough agents doing like 99.9% of computer related tasks for us or do you not see that being the case? >> Yeah. Yeah. I think that's that's very soon. Um I think I think by some measures like we might already be close to that, but I I think even in the practical sense that you're talking about I I think that's probably true.

I think it's just more and more abstraction um and things just running uh and they're being run and they're being run. Well, this is also a big problem with alignment because um yeah, when it's the one running everything, you don't know exactly how how it's being run, what decisions it's making. >> Yeah, it it's really dangerous when >> it's already weird that you're not sure how trains work. you're not sure how the whole airline system works, but it's kind of okay because you know it's humans running it. >> But what if it's a not very smart AI that's running it and a new smarter AI gets born and it's able to take over the other one?

Like it it's not even like a Skynet malicious type situation where we could lose control. It might be a very boring way that we lose control. >> I see. So where's that headed? like will we see like more adoption of you know Linux and Rust and like the type of things that agents would prefer to build with rather than humans and maybe they go even lower level than that like assembly and and binary or no? >> Uh yeah, I'm not sure what the tech stack will look like.

Um I mean they might just be like yeah this is all garbage and invent their own tech stack that's just way better. Um, no, but I I think fundamentally it it'll be a whole lot of like modular um AI or or API based interfaces with like authentication infrastructure. So, it's like everything can talk to everything else. Uh, which is a big thing. I talked about it in that book. Um, it's like that just seems the way that things will work because what you could do then is you can instrument everything.

Uh so what I was talking about there was this concept called a universal demonization where like every object and every person basically is broadcasting a demon. So like the park bench or the the tree in the park is broadcasting um I've not been watered in six days. I'm very sad my leaves are in pain or whatever. And the city Damon is looking at the tree Damon and it will you know turn on a sprinkler or whatever. So, it's just like the whole world is instrumented.

Um, and another example I had there was just like um you walk into Starbucks and everyone's broadcasting their Damon and in my ear Kai says, "Hey, this this girl like two uh places in front of you in line, she's like really upset right now. She doesn't have any friends. Uh she loves skiing and you know you love skiing or whatever. So, like you should say hi." Like I've already talked to her agent. her agent said it's fine for you to approach or whatever.

So like you move up and you buy her a coffee or whatever. So it's like every person and everything is kind of interconnected. I see. Okay. Uh I want to like double tap on the point of you know people like the nine-year-old being super advanced and then some some people like kind of ignoring AI like where where is this headed like very practically right? We not talking about like singularity and and the future of of like once everything is automated but like next one two three years where people just who know how AI is powerful who can control hundreds of agents productively thousands of agents versus people who like tried JGBT two years ago and kind of gave up on it.

Where do you see that headed? Like how are you thinking about that? Yeah, I I think it goes to um a pretty predictable place, which is kind of a more extreme version of what we have today. Um so I I think some massive percentage of like the the uh less ambitious or the less uh lucky or talented or whatever kind of move off and lose their jobs or whatever and move to like some sort of UBI system. And you have um like this crazy economy kicking up with like whatever the top percentage is, let's call it top 10%.

Uh of like people who are innovating or whatever or making things or have desires to make things. And so it's funny when people say, are there going to be more or less jobs? The answer is violently both. Okay. So most jobs will massively go away that currently exist and there will be a massive amount of UBI and most people will just not be able to make anything that's that people want and that'll be like a massive number of people who who knows the percentage and then some percentage of people will take their agents which are going to be another like millions of people essentially who also work for them uh or billions of people who also work for them and they're going to create this insane economy of new products, new services, new entertainment, new everything.

So, it's like the top like you know K-shaped recovery, the top just absolutely blows up. It's like utopia or dystopia but craziness and like the economy massively grows most likely assuming something bad doesn't happen and it's just like the world is like it's never been. It's just absolute craziness. And then for most other people, they they're either in that group or they're not in that group. And they're kind of just like surviving participating in like uh you know uh entertainment.

They're watching shows. There will be a lot better shows because people will be able to make shows themselves. Um so I I I see things really separating into those two groups um in in say the short to medium term. Yeah, because like all of the, you know, lab leaders when they're on a TV interview or, you know, some famous podcast, they always say like, "Oh, yeah, our job is to give it to everybody and give it to every but it's just a lie." Like, like literally we can look at what we see with our own eyes and, you know, most of our like friends that are not in the industry, they are not getting massively improved by AI, right?

They're they're like kind of ignoring it. Maybe maybe they use it instead of Google, but they're not like building their own software and like automating parts of their business and like growing faster and like upsklling themselves and learning faster. They're not doing it right. So whether it's a lack of agency or ambition or creativity or IQ or combination of all of these, it's just not happening. So I just wanted to like address that because you know when you talk to like Sam Alman, he always says it like it's going to be equal like give it to everybody but people don't even do anything with it, right?

That's the crazy. >> Yeah. Yeah. Exactly. Exactly. I I think they're right in saying they want to give it to everyone, but that doesn't like you said, it doesn't mean they're going to do anything with it, right? So, the end result is not going to be everyone is doing this. >> Yeah. Now there is one caveat there though which I find super exciting which I want to write more about but um there's one super exciting option which is um imagine we have like whatever mythos 19 or fable 27 or whatever and um we're like hey uh what is it that makes some people super ambitious >> and super whatever >> and like it it just like oh hold on um I need you to go into this lab.

In fact, I need uh 4,300 people to go into this lab. I'm going to do these scans for the next nine months. Um and out pops this pill and the pill pops out and somebody who just sits and watch watches Netflix takes the pill and in like four hours they're like, "Holy So, tell me about these agents." And then boom, they just activate. And it's like this is not this is not an inconceivable thing. If if you're a materialist like I am, I I'm guessing you might be.

Um the there is actually a difference in the brains of these people based on culture, based on their upbringing, based on whatever all the different combinations. >> And so we have drugs for treating ADD, we have drugs for treating like diabetes or whatever. These are physical changes in the brain, right? And so you take this pill and all of a sudden I mean how about this? How about an acrosstheboard 15 point IQ gain for everyone on the planet. >> Yeah. >> But may maybe in addition to IQ maybe because IQ is not really >> necessarily the most important thing.

Maybe it's like maybe there's a pill for ambition, maybe there's a pill for agency or something like that. Yeah. Exactly. So I think that's one really positive thing because >> I see okay so basically sorry to interrupt but like as it becomes a problem because again like the startup ideas and business ideas are to solve problems right and as it becomes a problem that like large chunk of society is like disconnected and checks out then that problem is valuable to solve.

So there's going to be like some lab like biology lab, AI lab that's going to figure out like okay is this genetic? Can we do a pill? Can we do like a some rehabilitation? Can we do like some gene editing? And that suddenly becomes the problem to solve which like the free market incentives, you know, kick in. >> No, absolutely. I I I think it it would be wonderful, right? because instantly you'd have a whole bunch of like um groups and tribes and you know subcultures who just hate each other.

They're just like uh let's let's play soccer instead. Let let's solve this on the soccer pitch. Um and let let's not do war. It doesn't seem smart for anyone. You know what I mean? And that could be Oh, another thing that potentially could be enhanced is just like empathy, >> right? What? Oh, uh the other kind of related uh tech here. Um I had this idea like 20 years ago or something. Imagine a machine that an AI built for us where we just step into other people's lives. >> So to see like why they did something.

Well, you just like you experience being a nine-year-old in like Nairobi and like you just saw something happen to your parents or whatever and like you're some billionaire executive in Silicon Valley and you're like, uh, I need to redirect some funds cuz like I'm being a total Like, you know what I mean? Just empathy for like the experience of experiences of others. So, I I think there's a few angles of like AI progress that could just uplift all of humanity.

Um, but I think the more short-term one is the one we were talking about, which is a lot more people watching Netflix or whatever and and this subgroup just taking off into the stratosphere and um that's why I don't think it's it's one or the other. All right. So, uh I really want to see how you develop software. So, can you show you already hinted at like your life OS and how everything goes into this unit of work, but can you show us like how you actually work right now with the agents?

What's your go-to harness? What's your go-to models? >> All right. Do you have any ideas that you can think of that would be a cool idea to sort of uh maybe do a blog post on >> like software or like something to build? >> Yeah, >> like some let's do something with Jeff, you know, cuz that's been a big release. May maybe something like you've already built something similar how Jeff can be useful. >> Okay. Um All right.

So, one thing I've been curious about is uh if we can look at all of our previous sessions and use Jev to analyze all the sessions and determine what we should fix across the whole life OS harness. >> That's great. >> Um, so let's think about doing that. >> Yeah, that's great. Like I literally I thought about like doing this even before Jeff released just like how much value there is in in the agent traces, right? in all the different sessions and like why are agents stopping right for me that's fascinating like for just what what aspect like when the agent gets blocked is it because loss of internet is he missing context so yeah this is a great idea >> yeah and so I already have this system which I built before Jev uh it sounds like you were heading in the same direction I called it aa and it was basically every mistake or complaint like when I start cussing it should automatically be tagged as there's something that we need to M >> um so um yeah, this sounds right.

Um and let's go ahead and build an ISA for it and get it started. And by the way, you see that say uh say why. So I'm very lazy. So why would have just made it proceed? Um >> nice. >> So it looks like Jev is sorting and soul is explaining. Um by the way, I have um a new custom harness built that is Jev based. that's routing um my work to different OpenAI models and I'm using my anthropic models for different things. So um most of my work is being done uh max is Astra um highest soul then Terra then Luna.

So I got a custom router going there. Um so I think it's going to look at my ARA database which I >> Okay, let's double tap on that on the the router. Is it like for the purpose of like token maxing and getting the most value of your subscriptions? Is it for the purpose of like routing the task for like you know Kimmy was the best on legal? So like you get a legal task routed to Kimmy like what's what's the idea there? >> Right now it is intelligence based and token based.

So um basically I have four tiers which are those four levels of open AI models and those four tiers um are being chosen based on user prompt submit by Jev. So um I am each of if you look down here um actually that got removed. Hey we need to bring back our model list and um the agent list that we had in our status line before. I'm not sure why that got dropped but it happened yesterday. Um all right so I guess first thing is um this is herder.

This is a highly modified version of Herder that's running um for life OS and I've got that custom thing there. I also mentioned um an ISA which is an ISA document. So let me show you this. So I've basically unified you know how we went through all that spec kit and plan documents and all that stuff and you've got like Matt PCO's uh like skills and stuff like that. So my whole thought on this um where is that thing should be here?

It is the ISA system. So this is a single document that unifies what ideal state looks like. So this has the description of what we're want to do. These are all the sections of the document. But it's just one document, dude. It's it's absolutely insane and I think it's super useful. Um so this software is basically going to go build an ISA and this ISA is the it's the um goal as I articulated it. It's the decisions that we've made throughout.

It's the list of steps. Um in fact, let me just show you this. Um but to finish answering your question, the router is fundamentally determining how much intelligence is required for this task. So, if I'm doing like a whole bunch of SDS and JQs or whatever, it's going to go down to like Terra or Luna. And if I'm doing any intelligence based work, it's going to be Solar Astra or or Fable. Um, all right. So, this is the ISA document for my um Surface app.

I I got to tell you about this offline. It's super sick. This is like the way I process all news. But this is the Surface thing. It's got the problem, the vision, the articulation. But if I move down here, look at these. So these are the individual steps. Let me uh these are the individual steps of what is being built. But the crazy part, dude, is the steps for what to build are also the steps for the testing harness.

So the testing harness and the build steps are identical. And so um any changes that are made, the decisions and everything, they're all going into this one single ISA document. So that's that is the bottom line of how I build software is uh is using these ISO documents. Um all right. So let's see what we're being asked here. And this is all custom stuff for LifeOS. It's uh it's got a specific um output format. Uh do I want a version row back? uh model list.

Okay. Yeah. So, it brought it back in the status line. Let's see. Um still not seeing it in the status line. Oh, there it is. It came back. Here it is. So, you see how uh it's got opus. It's got these are basically all my models. >> Mhm. >> And these are all my agents. And the agents are obviously named the same thing as the model. So, as they're being used, they're lighting up. So, I could see which ones are being used uh based on the router.

Um, and then I've got context down here. This is my um anthropic context. And down here, this is my uh OpenAI context. Oh, the other thing to mention. So, remember I was talking about how everything has to be consumed. So, I don't have to very say very much when I ask for something. It immediately just knows what to do and how to do it exactly the way that I like. And that's because of this. So skills operational rules lifeos system prompt uh the claw.md this is uh digital assistant identity.

So that's kai this is principal identity that's me and all these combined is this size. So it's progressive loading but the system instantly knows everything about me in a progressive way immediately on load and my claw.md file. Look at this. My claw.md file is just a routing system. >> I see. >> It's just a routing system for how to get to things. So, it's not like instructions or anything because that's in my uh system prompt. >> So, would you recommend this for like only general agents or even like uh you know cloud.md or agents.md in a specific project? >> Um I don't do cloud.

MDs in in specific projects. Um the the um document for a project is the ISA. >> I see. >> Yeah. So there's only one document for for anything which is the ISA. And if it's a massive um project like I'm I'm building a bunch of like role playing game uh and game design systems. Um I will sometimes have one ISA that has like four ISIS underneath it. like if there's a entire character-based system and a languagebased system or whatever, but it's functionally still one is just limiting the size of one document.

Um but yeah, every single thing is just the code and the ISA and the whole thing is that the ISA has to represent the code and vice versa. And so um also inside of um my bunker system that is checking and by the way all these checks the whole testing harness is built on my eval system um and I just upgraded the eval system to use Jev as well um because it's a really really good opportunity for that. So basically eval on one side you have asserts which are deterministic and on the other side you have the judgment ones and the judgment ones break down into rubric which is a list of um options that the model picks from and then the other one is a tournament which is picking from two options and both of those sides perfect for Jev.

So now my whole eval system which is built into the ISA which is built into every single skill is um starts with Jev and then uses models as as needed but Jev might be able to do most of the work. >> Yeah. So let's talk about Jeff a bit more. Like um do you think we'll see more models like this? Like what were your first thoughts when it got released and like yeah how how did you begin integrating it into all kinds of stuff?

Yeah. So, the first thing I had to do is just kind of grock it. It took me like an hour. I'm like, what? And then I'm like, wait a minute. This is like because if you think about like everything we're asking, we're asking AI questions that normal automation can't answer, >> that scripts and code can't answer, right? So, it's some sort of intelligence thing. >> And if if um I actually want to do a post on this because I hadn't thought of it this way, but it's like you ask a question.

Well, what does an LLM do? An LM gives you an answer in the form of text, a sentence. Um, well, Jev can do the same thing if you pre-esign the questions, the options for the answer. Yeah. >> So, there's like a mini step that has to happen before Jev is useful, which is coming up with a really good list of steps >> or a list of options or choices, right? >> Yep. So um what I'm basically did was I basically said okay across all of life OS tell me where we are doing things that don't require a text answer but can be done with a decision answer instead.

Guess what is really powerful for this? The entire hook system on prompt submit um is does this look like prompt injection? Does this look dangerous to you? Um how should I route model selection? which model should I use? Like um people have talked about the email situation like uh analyzing a whole bunch of like customer sentiment stuff like that. I basically said across all of life OS look at all the different components and see where we could use Jev and um it came back exactly as I thought it would which is um hook system um the routing system and uh there's a bunch of other places um oh most importantly the eval system because that's like the basis for everything um and every single uh ISA by the way um these all have eval tied to them.

And the eval are basically how um how you check whether or not that feature is still working. You know how when you have production software and it's like is this feature broken or not? >> Y >> so you could do that with asserts sometimes and sometimes you need judges uh to do that. But mostly uh what I have in bunker is all asserts because you want to be able to do them fast because I'm running like I think like 1,500 of these like every hour because I have so many things deployed live >> uh with asserts. >> Um but yeah, it all goes back to this single concept of like there's one ISA for this application surface.

I'll just show you the app real quick. Um, so this is like my news reading app which I built and it's basically um it's parsing like 5300 sources. Uh, and then it's reading every single piece of content that comes off of a source and then it's objectively rating them. So it doesn't matter who wrote it, it's getting an objective rating of how good it is. >> That's great, >> right? >> Yeah. So, um, this is another thing I'm I'm up upgrading with, uh, Jev as well, uh, for making sort of decisions here.

Um, yeah, let me just show you something real quick. So, there's another way that I build software here. Where's Arbble? Here it is. So, these are all my services. Um, I have everything running in Cloudflare. >> Okay. So, if you if you look at these, a whole bunch of them are for my offensive security testing because um I do hacking and bounties and stuff like that still on the side because it's just my DNA. And um these are all individual think of these as all like individual Unix commands and they're all running on Cloudflare workers and um they're basically chained together and into pipelines which are chains of actions.

And a flow is a chain of a pipeline combined with an output. So if you look at these, these are like um stuff I have for for some consumer apps, but these feed ones are for surface. So all of surface is being powered by these backend services, but it's all feeding off this one um ISA, which is this. There's only one document for the entire app. >> So then it's easy to like have any an agent or model like check if if they're they're going in the right direction because it's in a single source of truth. >> A single source of truth and any changes that are made it immediately updates the ISA first, builds the code and then makes sure the code matches the ISA. >> Yeah. >> And keep in mind and they're all individually granular steps.

Um, oh, one cool idea which I haven't implemented yet was to use Jev to say is this criteria granular enough, yes or no? And if not, create an additional um I call it an ISC uh ideal state criteria. So this is an individual state ideal state criteria uh which has you know uh eval associated with it. Um but Jev can say is this granular enough and maybe break it into additional pieces. Um but the but the idea is like dead simple.

So first of all you have to so a big part of life OS the concept is that the reason AI fails is because we have not properly transferred our brain >> into the AI. >> I agree. And right and so so going back way back to 20 and 24 I did this uh project called fabric and it was all just prompts designed to get um our desires into the thing to articulate in prompts properly. So it had a structure of the prompt and um I basically taken that and expanded it into the ISC.

So it's like when you start a project your number one goal is to perfectly articulate exactly what you want >> and that's what gets captured into the ISA. And if you don't do that properly, it actually hits you up. And I I I took some of this idea from uh Matt. Um if it's not properly articulated, it doesn't think there's enough detail. It'll just keep interviewing you until it hits like an 80% uh like fullness and then it can start building.

Um and then whenever you have a problem, guess what? It's because you had something in your brain that was not in the document. >> Yeah. I mean that's that's also in human systems is the same, you know, like usually a manager or like miscommunication. >> Yeah, exactly. >> Yeah, totally. >> Yeah, this is great, man. I'll have to run. So, I wish we could like do a deeper dive just into this because this is great, but I have a dinner scheduled soon, so I need to go.

Where should people go? Where should we send them? >> Yeah, check out uh the website danielis.com. check out um ourlifeos.ai um and yeah, you should be able to find everything from there. But um yeah, good chatting >> and likewise. I'm going to link all the socials below and everything we showcase so people can just access it easily. So yeah, appreciate your time, Daniel, and have a great day. All right, take care.