Your Team Is Using AI Wrong - Build a Company Brain Instead

Everybody is talking about loops.

Everybody's talking about skills. Everybody's talking about evals. Everybody's talking about company brains right now.

But nobody is sharing how they're actually implementing it right now.

So I spoke at a private event recently, and it was invite only.

And these are— we're talking 8, 9, 10-figure operators that were there. And nobody's really sharing this stuff. So I wanted to share that conversation that I had with this audience with you. So you understand how you can grow a lot faster. So how we're using company brains right now, how we're actually using these loops, how we're about graph engineering ultimately. There's a lot of numbers, there's a lot of case studies that you can pull from, from within this video. What I'd also recommend is that you take this video afterwards and then you dump it into your harness and you figure out how you can skillify it, use it for your organization so you can use it to grow a lot faster. So without further ado, let's get into it.

Before I start, I just— we did a quick show of hands real quick. I just— quick show of hands. Who here is actually using like a Claude Code or a Codex? Good chunk of you. Okay. Keep your hand up if you're using like a Hermes or an OpenCLaw harness. Okay, like 6, 8 of you maybe? Okay, great. So I think what I want to focus on today is more around how we actually go about compounding our knowledge and how you go about doing that. A lot of people right now are talking about, they're talking about skills, they're talking about evals, they're talking about loops, like what are these things? And that's what we're going to talk about. So we're going to talk about this concept of building a single brain and If you think about it right now, most people, yes, we talk about we're using AI, most people in our companies are using AI right now, but not many people are actually compounding. So maybe it's like 9% of people right now that are actually compounding using AI. And that's a problem because if you become really good at it, that's not good enough, right? It's not good enough if you're compounding at this level and then your team's compounding at this level because that gap's going to get bigger and bigger. And so I just, I hosted a dinner the other day with a couple of operators here in New York, and these are 9, 10-figure revenue companies. And the conversation right now is like, how do you compound knowledge? How do you bring everyone along? And also, how do you adjust compensation when it comes to AI? Because the AI-pilled people, they're driving a lot more output now, right? So these are all the conversations that are happening. Maybe I'll share some of those stories as we go along here. I'll try to cram as much in these next 20 minutes or so. So the other narrative that you're hearing quite a bit right now is, Oh, AI is going to take jobs. AI is taking jobs. Anthropic's like, oh, 50% of jobs are going to be gone. And there's this whole job loss narrative. AI is eating commerce right now. The way that we're doing work is different. But your job, that whole narrative, I don't think it works because the reality is we're going to be doing a lot more. Because if you look at the number of jobs that we have today, 60% are net new since World War II, right? And technology has always enabled more and more. So when you look at The industrial age. Sure, we lost 8 million farmer jobs. Well, what ended up happening? Almost 50 million new jobs came out from that. Oh, the spreadsheet came out. Oh my God, everyone's going to lose their jobs. That didn't happen, right? What ended up happening was more jobs ended up forming from that in terms of finance jobs. And then if you also look at radiologists, for example, oh, AI can scan, can do all the medical scans now. We don't need radiologists anymore. Guess what? you have more radiologists than ever. So I think we're going to have net new job creation. But what's more important in these next 2, 3 years or so is how do you actually harness the power of this stuff, right? And so what I would argue right now is that we don't necessarily have a people problem as much as we do as a knowledge problem. And so how do you compound that knowledge? And so one thing that we've been doing internally, and I'll give you the high level on how you can do this, is we just started working with our AI inside of our Slack, inside of our Microsoft Teams. And the moment you can connect your different tools, tooling in there, and I don't even know if you can see this right now, but you can basically ask it whatever questions you have with your paid media that you're running right now, your email marketing that you're running right now, your SEO, your AEO campaigns that are running right now. And not only that, it becomes a multiplayer game because those of you that are using Codex or Claude Code right now, you're using it in a single player. way, right? Those of you that are playing with OpenCLAW Hermes right now, you're still using it by yourself. And sure, you're becoming a lot more powerful, but your team's not benefiting. And so by having one of these little single brains or company brains living inside of teams, your people end up getting smarter with you, and then your company ends up compounding a lot faster. And so when we have this working in there, where it's asking about, hey, what's our ROAS on this campaign over here? Hey, can you spit out more campaigns? Can you spit out more creatives? What's going on with AE over here? Can you— can we double down? Can we triple down on that campaign? Everyone just ends up moving faster because you're able to transfer knowledge a lot faster. And I remember I was reading something maybe a month or two ago in terms of how humanity has really sped up to where we are now. And how many of you actually think our brains are bigger than they were 50,000 years ago? Nobody, right? It's the same. So what actually ended up happening? What happened was humans started to trade. And it's not only just a trading piece, it's when humans start to trade ideas, we start to move a lot faster. And so that's like, I'm like, well, that's the same thing with this. If you start to trade ideas faster inside of the surface that you're used to communicating in, you end up moving a lot faster. And by the way, this might seem complex to you, it's not that complex. We build this out for our clients right now, it's simple. We're just using Nemo Claw, which is NVIDIA's enterprise runtime, and then we sync it up with Hermes.

Right.

which is the harness that I use personally, and then we build it on their infrastructure, and then we help tweak their permissioning, governance, and all that stuff. You can build this yourself if you have a team, like a competent AI-pilled engineering team, you can build this. I'll go on a little tangent for a second. I asked a question at that dinner that I hosted the other night. I was like, what's the one team that you'd completely rebuild from scratch if you were to redo it?

today?

What do they all say? Everybody at the table said this, their IT team. Everyone said their IT team. Why is it that— why is it the IT team? Because there's a lot of inertia, resistance for whatever reason within their IT teams, right? And so, kind of similar to that, I would also say that there are a lot of companies right now where their product teams are actually slowing them down, right? So wherever you're not seeing it happen, I would just say that one of my friend's companies, product team is not moving quickly enough, so they hired a team of cracked interns that are moving way faster than the product team, and the product team hates them, right? And here's the funny thing, they ended up firing their product leader, right? They ended up, and then another one quit, and then now their AI-pilled interns are actually starting to move over and take over those roles. So that's a whole nother conversation. My whole point here is if you want to compound, you got to start to work like this. I think everyone's going to start to work like this within the next 12 months. 'Cause there's already a lot of competitive solutions that we have in the market already, and I think anybody can build this. So the other piece here is that, you know, a lot of companies talk about, you know, it's always a process problem. It's never— it's 95% of the time it's a process problem, it's not a people problem. The problem with that is how many of you have actually created processes out there? Like everybody here has, right? But how many of you actually reread the process that you create? Probably not many of you. Maybe the operational people, yes, a little bit, right? The most, one of the highest leverage things that we do right now as a company is we just skillify everything that we do. If you do it more than once, we skillify it. So we just take a process, we make it into a skill.md file, and then we allow people on our team to fork them, right? And I'll show you what I mean in a second, because we have an entire skills marketplace that we created for the team, and it allows us to kind of monitor who's active with it, who's making different versions of it. So if you're skillifying maybe an email marketing process that you have, maybe an SMS process that you have, right? Maybe a paid media process that you have, stuff that you don't want to do, right? Robot stuff that you shouldn't be doing in the first place, you're going to be able to move a lot faster, right? Because then these things, these skills are very portable and you can hand them off. So that's what a skill is. And then I'm going to talk about evals and loops in a second too. And loops, I think, are the most misunderstood thing when it comes to AI at the moment, because everyone thinks they understand what a loop is. It took me probably, I don't know, maybe I'm not that smart. It took me probably 2 weeks to understand what a loop is because you had the co-founder of Claude Code talking about it, and you had the co-founder or founder of OpenCLAW talking about how they don't prompt anymore. All they use is loops now over and over. And I'm like, I don't get it. And everyone started to give their different definitions of what a loop was, but nobody really got it. I even talked to a guy that wrote like an article on X about loops and got 4 million views. When I talked to him, he had no idea what he was talking about too. So I'm going to share with you guys what my definition is, and we'll see what happens. And I even have a loops creator skill that you guys can steal. When I share a QR code, maybe towards the middle of this. So what I have on the screen over here, this is one of our creative agents, and I truly think anything I show right now, anybody can build, right? I was talking to— I'm in town for investor meetings right now, and some of the VCs were asking, can anybody build this? I was like, yeah, absolutely. Who's going to maintain it? Who's going to optimize it? Who's going to manage it, right? Just 'cause anybody can build anything right now doesn't mean that you're not gonna need people that are actively managing these things, right? So this is one of our creative agents, Picasso, and it took one of our clients from, I think, 5 creatives a week to hundreds of creatives a week, right? And these are— this is an example of a company called GymPass, or now it's called WellHub, but formerly GymPass. And this allows us to spin up a bunch of different versions here. And We'll share these slides with you afterwards in case you can't read this, but the whole idea here is that, you know, step 1, like a skill. What are the hero concepts? What's the brand kit look like? Step 2, what variations should we generate based on your ad library or based on maybe what your competitors are doing or based on other data that you have? Step number 3, brand safety eval scores every frame. What is an eval? I'll get there in a second. Number 4, exports ready, and then you have that staging to your ad account, and then you can ship more ad creatives ultimately, because a lot of you that are, because we're talking about commerce, are constrained on the creative piece, right? And sure, you can use the HiggSEO MCP, that's all plugged in here. You can plug in whatever MCP you want, whatever API, whatever 3-letter acronym that you want. It's fine, right? Point is, we've been so used to working in such a manual way that it's hard to undo what's led us to be successful, right? And what's also interesting to me, just a little side story here, Is one of our— when we're hiring these interns right now, by the way, these are like 17-year-old people, 20-year-old people, they're running circles around our experience team. And I talked to them because I gave one of the interns his own Hermes instance when he started. And one time it was not working one day, and then he was like, it wasn't working for a few hours. He's like, dude, I can't do work. I was like, what do you mean? He's like, dude, I don't know how to work like this. This is like drinking soup with a fork. I can't do it. And I'm like, okay, well, like, then that's a huge paradigm shift because he onboarded himself using one of these agents, right? Whether it's the Single Brain that I showed you or the Picasso, this is the new way of working. And kids don't know how to not work like this. In fact, when he sees how I work, sometimes he looks at me like I'm stupid, right? So I think we have to undo a lot of these old habits that we've had in order to move faster. So this over here is what we call our Skills Dojo, and I think every company needs to make one. I don't think it's that hard to make one. In fact, I think I have my GitHub repo where you can steal the code for this. The whole idea with the Skills Dojo is your entire team, they can upload their skills, people can fork their skills, they can download skills, they can leave upvotes and downvotes. So you want to make sure that you give your skills a home, because then the other thing is we have an analytics section too, where you can see which teams are engaging, which individuals are engaging, who's not engaging, maybe who needs help, and maybe who needs to be probably, you know, shifted to another company. Um, but no matter what, you want to make sure that you give your skills a home because that is where you can see the compounding happen. And not only that, we, um, these, these skills, like this is on my, uh, X long-form skill writer, and I handed it off to someone on my team, this skill. He took that skill. I was basically getting on X in January, like 100,000 views a month, which is not a lot, right? Now, last month, I think we got 3 million or so, and it's continuing to compound over and over. And he's made my initial skill so much better. He forked it, so he made it his own version. He added new scaffolding to it. And to him, he's like, dude, Eric, I feel like I'm playing a video game. I'm like, yeah, it feels like a video game. He's like, there's no reason to play games anymore. If you can make work feel like a video game, everybody wins, right? That's a whole nother conversation. We're not going to get to that today. So, um, I have a lot. But this piece over here is super helpful when you're hiring new people. These are skill packs. What does this do? Here's a revenue pack over here. What that means is you hire someone on your revenue team, your sales team, your marketing team. They can have this AI revenue intelligence skill, this daily sales standup. There's all these skills that we use over time. Again, it's just a, it's just a repeatable process. That's all. And these skills can be fed into Slack. They can be fed into Claude. Claude Code, Codex, I guess it's called ChatGPT Work now. You can feed it into whatever harness that you're using. They're very portable. These are just portable processes. That's how you want to think about skills. So the next phase here is evals. So you have skills, we've kind of talked about that. Evals are your definition of done. That's your definition of success. Okay, so what is it? Okay, so this, uh, Maybe you might say it's a page optimizer skill. It's only successful when you bring a page down to 50 milliseconds, so a page has to load within 50 milliseconds. That's an example of definition of done, because if you don't define what success is for a skill, it's just going to keep going. It's going to keep rolling. It's going to keep using your tokens, and it's going to spend a lot of money. I'm sure some of you right now are spending 6 figures on tokens. Maybe some of you are spending 7 figures. Maybe some of you are spending 8 figures on tokens. So you want to make sure that you have some evals to keep your skills in line. And you also want to make sure that you're using the right skill, but also the right model too. So I like this meme over here where this guy's using a blowtorch to light his cigarette or cigarette cigar. Obviously right now you have all these, the best frontier models, ChatGPT, Solf 5.6, you have Fable 5. These are all great. They're also very cost intensive, right? So maybe 5% of the time you're allocating it towards those models. Maybe 15% of the time it goes towards, let's say, subscription, okay? Maybe 80% of the time you're using these open source models. And the good news is you don't have to use the Chinese models that much anymore because you have these American models coming out, the Meta ones, the Thinking Labs model that just came out yesterday, this American model. and maybe 80% goes there, and then you run it on your infrastructure, and you save a lot of money that way, right? And so I'm not saying the Chinese— by the way, the Chinese models are really good. My background is Chinese, but I'm not saying don't use them. But here's the funny thing. At dinner the other night, one of the guys, he runs an AI CFO company. They have a lot of users right now, about, I think it's like $50 billion assets being tracked on their platform. And we're all talking about the Chinese models, and it doesn't matter, like you're talking about People running multi-billion dollar companies sitting at that table. And we're like, why don't you just use the Chinese model? He's like, yeah, we're one of Anthropic's biggest spenders. And he's like, I can't use the Chinese models. I was like, why? And he's like, because if they start to ask about how you can save money on taxes as a wealthy person, it starts to get CCP-like answers, right? So he's like, I can't use that. So anyway, tests. I mean, you should be buying your own infrastructure. You should be renting your own infrastructure. You should be testing out the frontier models. And I also believe those of you that are leading a company, operating a company, working inside of a company, you, everybody in this room, you have to be the chief AI officer. I don't believe that this is something that you can just hand off. Oh yeah, I have someone on my team handle it. You have an AI person. Yeah, that doesn't work. Anybody I've seen try to do that, you can tell there's a different dynamic when they're talking to other leaders, other founders. Everybody here, like, you have to take this on yourself. And I don't think it's that bad at the end of the day. It's just like learning how to use the internet. It's learning how to use email. Maybe a little more work than that, but not that bad. So evals, we kind of talked about this a little bit. So a skill will produce for you, you have an eval that grades, and then we're going to come over to loops. So this is what a loop looks like. Here's the steps for the loop over here. And I have a better diagram for the next one, but this one's pretty cool too. So the first part with the loop is you— and a loop, again, is a workflow that compounds, a workflow that learns. So you might have the first step where you have to define what your target is, and then it goes all the way down and it observes what the target is and it learns over time. And this is actually a better diagram over here. So on the very right side of the loop, you have your target state. So that's what you're trying to hit. So let's go with the example again. You want every page on your website to be 50 milliseconds, okay? That's the target state on the right side. Right below it, observe the current state. So let's say every page on your website, maybe your average is around 150 milliseconds right now. Then you have the gap. So 150 minus 50 is what? 100. So that is the gap. And then what happens is the loop learns over time. That's all it is, and it gets better. And so— I'll give you an example of a loop that we use. So one is called, I call it the Mr. Siu Sponsorship Loop. Okay. And the reason it's called that is because one of my friends, he used to have his assistant responding to people saying, you know, Mr. Siu cannot take this deal. Please, please, please come back with a better price, right? Because once you do that, it elevates your status. Here's the thing. One, I don't want to respond to these people. 2, I don't want anybody on my team dealing with this. So this loop, what it does is it looks through my email and it responds to people every single day. And it says, Mr. Siu cannot take an offer less than $25,000 for a sponsorship, right? And the reason it's $25,000, one, I don't want to do it, okay? Two, there's some people that would pay for it and I would gladly have it fund my content team. And the most important thing about that loop is every day I look at it, it says, we tuned these 4 things, we tuned these 5 things over here. So I can see it getting better every single day. And it sounds complex. It's not that bad because I think in a couple slides I'll show you where you can get access to the skills dojo that we have. I have that loops creator in there so all of you can use that. And here, how did I build that loops creator? Yeah, I just sat with my Claude code and my Codex and I just kept going back and forth. Like, that's it, right? Now the difference between a loop versus an agent is an agent, you might have different connectors, different tools. Maybe you have your Klaviyo, your Shopify. Maybe you have whatever other tools you have, Google Analytics, right? Your CRM, whatever. And then you have different prompts within that agent, and then you have different artifacts or answers within there, right? And that could be like agents can handle a lot of one-off work and repeat it, but they don't get better over time. So that's why it's— that's why you have all the people in Silicon Valley right now talking about loops, and that's why they're so important. If you think about loops, by the way, the people on my team right now that are using these loops, like there's a social media loop that's being used right now. There's one that's being used for YouTube as well. This makes everyone an A-player, right? This is how you can truly compound your team. And then you have, like, I like to say to my company right now, like, I think I have 2 cultures, and I say this very openly. One culture, you have the AI pill people, okay, that are flying really quickly. And then the other culture, not so much, right? They're struggling to pick this stuff up. But I think our job as leaders or people that are working in companies right now, it's our job to create the conditions for success. So, Can you create a Skills Dojo for them? Can you teach them how this stuff works? Can you set up these trainings every single week for people to get better? And if they choose to take it, great. If they don't, unfortunately, it's not gonna work out long-term, right? But it's also incumbent on you to repeat this message over time. So here's the QR code over here for the Skills Dojo. And this is completely free. I'm gonna give you the meta as to why we built the Skills Dojo and gave it away for free. So again, that Skills Dojo is our marketplace where we put our skills. Now, for me, having a B2B ad agency and having the single brain piece where we're implementing these brains, it's helpful for me to see if this thing is public, who are the users? Who are the people building the most interesting skills that I can reach out to, to work with, or maybe potentially hire? The other piece is, like, if I make this an open platform, I can also see which companies are using it, which companies are ICP fit that we should be working with, and then we can have an agent reach out to those people and then see if it makes sense to connect, right? So that's how you can connect all this stuff. How do you give a lot of this stuff away for free? In fact, a lot of our software now, we just have free versions for it because I see a lot of self-serve SaaS going to zero, right? Those of you that might be running a, running a SaaS company. So how can you give stuff away for free to drive distribution and how can you continue to promote it? And shout it from the rooftops and go from there. So a couple moves by Friday. One, think about one workflow that you want to skillify. And then second, what do you actually want to loopify? Okay, you can, I guess you can skillify, you can evalify, and I guess you can loopify as well. Okay. And then what you want to do afterwards, you guys should use the loops creator that's sitting inside of the Skills Dojo, again, free, right? And then you You give it a shared memory. And what that means is if you're going to make a loop, you have to make sure that it's sitting in some type of repository, right? Where it can pull the memories over time, where it can learn over time. And guys, I'm saying this to you right now, I'm not technical at all, okay? I don't even know how to code. If I can do it, I think everybody can do it, right? And you just go, like, all I do, by the way, like, I don't even watch House of the Dragon or Game of Thrones or any of these things. I'm just on my phone, like, just making new loops and just sharing what I'm finding from people's X profiles. to my Hermes and just saying, hey, how do we loopify this? How do we skillify this? Over and over and over again. And the beautiful piece is when you work with a harness like a Hermes, it knows you, it compounds with you, it has memory, right? And so it knows what you're looking for and then it'll pick out the best parts of what to skillify and it'll do it for you. Then you share it with your team and your team's like, oh damn, yeah, that's pretty cool. And so what I do now is I spend a lot of time doing that. I get something to maybe 60, 70% there. I hand it off to our product people. I hand it off to our client services people. I hand it off to our engineers and they make it 10 times better. And that's your job here. I think those of you that are watching, like, like learning here, your job as visionaries is to hand this off to other people so they can build with this, so they know that they have these superpowers. It's not to hog the superpowers for yourself. So that's, that's step 2. And then step 3 is you want to loop it, which is what I mentioned already. I'll give you like maybe 2 more examples of loops that you can create, okay? And I think these are also in that marketplace too. One, we have a CRO loop, which is a conversion rate optimization loop. And so for singlebrain.com, if you look at that site right now, it is literally running a test that it came up with and it's optimizing. And all I told it was, use my Loops Creator skill, and I want you to increase the conversion rate on singlebrain.com by 30% in the next 4 weeks. And every week it's reporting back to me and it's telling, telling me what it tuned. And it's also telling me if it doesn't think it's fast enough. And it shows me every, every couple of days, it'll show me what it's tuned and shows, it shows me how it's getting better. That's one piece. The second one is the AEO SEO loop that we have. Okay. So a lot of people are talking about AEO now, SEO, whatever you want to call it. And what that loop does, it ingests all of my content from my YouTube, my podcast. Even our internal call recordings too, because that's usually when you have your spiciest stuff, right? And it'll figure out what should actually be created on my website from an AEO SEO standpoint. And then it'll draft it up based on our templates, based on, based on who our ICP is, right? And then the editor can choose whether to draft that or not. So that work used to be the work of an entire team, an SEO team, an AEO team. That can now be done. Like, by the way, That's just AEO SEO. I'm just— that's CRO. How can you do this for email marketing? How can you do this for SMS? How can you do this for paid media? How can you do this for creative, right? And I think every single week you should be asking like, what's really annoying right now that's costing us a lot of money? What's preventing us from 100x-ing? So anyway, I think that's enough to leave you all with. And I think I will give you my slides over here if you want the slides. And other than that, thank you for your time.