My Complete AI Operating System for Business

Eric Siu (0:00)

In this video, I'm going to show you the AI operating models that I run in my business. So this works for any business that is scaling, whether you are enterprise, whether you're a mid-market or you're starting up, because you need to have these operating models. You need to have some form of operating models that will work for you. So I'm going to show you how these workflows look. It doesn't matter which harness you're using, it could be anything right now out there. For the sake of this example, I'm going to be sharing Grokbot with you first and foremost, and I might have some other diagrams and workflows to show you as well. So your harnesses are going to best understand how you work. And so for me, I've got like six pages here that I can show you first and foremost. So this is how our bot fleet works. So whether you're using Grokbot, whether you are using DOTS, whether you are using Muse, whether you're using Hermes, for example, and there's all these other agents coming out, you want to make sure that you have these operating flows that are repeatable. Because if you don't have an operating system, just like if you don't have an operating system within your business, it's going to be very hard to scale. You want to make sure that the most important things are operationalized. And the cool thing is the computer's remembered its stuff, the agents remembered its stuff. So in terms of how the bot fleet works, this is for us. So for Grokbot specifically, and even for DOTS in its current state, you want to make sure that you have a human up top, and then there's a chief of staff, the main bot that hands out the work. So then what happens is afterwards with your team, you have these team bots. So you can make bots for yourself, and so I certainly have bots for myself, but once you get bots working for yourself that your team can use, like if you have a talent bot, let's say you have an AEO SEO bot, you have a GEO measurement bot, you have an outreach bot for sales, you have one for X, you have one for M&A as well, then you create these team bots, and then you deploy them into Slack, and so your team can talk to them, and then they have proof of each job, and you have a morning board with kind of a summary, and it just kind of gets repeated, and the work continues over time. And these bots are just going to get better over time, because we aren't even at RSI yet, which is a recursive self-improvement. So I believe that's coming relatively soon, maybe in the next year or two.

And then basically we have 40 bots, and then you have one human, which is honestly kind of the bottleneck over here. So that's why you want to have one thing that I'm going to show you will help kind of move things along here. So there's another concept here known as 131 So for your bots, if there's a 131 situation where if you have one big change, it needs to come with three options and a pick, and it needs to show the risk level of this.

So if any big change happens that your bot is maybe proposing or maybe you're suggesting a big change, the bot will write up the change, the three options, the one pick and the risk level. And then I say yes or no, or whoever's managing this says yes or no, and the change goes live. So that reduces some of the cognitive load on the human that's operating the bot. So one change, three options, one pick, nothing important changes without one yes. And this is always a godsend. This is how you want your people to work with you as well. So they're not just dropping problems off for you. Now, the thing that I mentioned earlier within this video was the whole idea of you not having to go back and forth with your bots all the time. So Lauren Tan, who created this process around how she's able to ship thousands of PRs, pull requests per month, she works at SpaceX.

So Trusted Continuation, I'm continuing to refine this over here, but the whole idea is that every bot gets checked twice each weekday and each one gets a clear label. So at 8:31 a.m., there's a check over here, 3.51, there's an afternoon sweep. And then the whole idea here for each bot is, hey, did the work actually happen versus the daily target? Was it safe? Did it learn from Eric's feedback? And then the weekly number move. So if you add these things for the bots, it's going to continue to get better over time. And then advanced means the bot moves on, Eric does nothing, you recall, Eric decides, bot gives it a pick, and you decide what to do. The bot names the blocker and needs a check again, and it skips its drop. So the new rule is weekly number behind pays two days, bot sends one blocker plus one fix in the yes line, daily batches are pre-approved inside set caps. So what does this do exactly? So the reason why it's called Trusted Continuation is that the reason you need to go back and forth with your bots, at least as of this recording, is because you don't trust it enough. But what if you could have these checks where your trust could build over time? What if your trust can compound over time? Just like how your trust compounds with the human? Well, you can set that up. That's why you need something like a Trusted Continuation flow. By the way, if you're enjoying this video right now and you need help with AI implementation, go to singlebrain.com. That's what the B is. singlebrain.com. This is where we help people with ad creator scaling. We help people with AEO SEO scaling. We help people with producing outbound email infrastructure for cold outbound, for example, or any type of outreach. And these are all things that can be managed inside of our agents that live inside of your Slack and Microsoft Teams. So singlebrain.com and we'll see you on the other side. And I would recommend that any of you watching this right now, you take this video and you dump it into a, you just send this video to one of your harnesses and you say, you know what, I want to go ahead and do what he's talking about over here. And you take the post from what Lauren Tan is talking about, whether it's on X or her YouTube videos, you take it and then you compile them together. And then you can make something like this, right? And then this thing's just going to get better over time. And that is how you build things, right? You learn about good ideas and that is how you build things today.