Eric Siu breaks down AI loops as business tools, not just engineering concepts. He walks through a framework: find the problem, build the loop, measure results, then escalate. Real examples include content systems, meeting prep, and task resolvers that pull from Gong, Google Search Console, and GA4.
Chapters
Everybody's talking about Loops right now. You have the creator of OpenCLAW talking about it, you have the creator of Claude Code talking about it. So Loops are sort of like as big as the step from source code to agents was. Loops are the step from agents to the next thing. It's just as important and as big a step. Keep in mind, all of these people are technical. And so how do you think about this if you are not an engineer? If you are a CEO, you are a founder, or let's say you're an operator, let's say you're a marketing leader or an operator, how do you actually apply this to to business. And so in this video, I'm going to deconstruct how it actually works. I'm going to give you a practical example on how we actually use it and how we think about it for one process around AEO and SEO. And you can even steal that if you want to. I even have some artifacts I'm going to show you as well. So stay till the end of this video, and by the end of this, you're going to figure out the right framework into applying loops, not just for building, but actually building your business. Okay, so this is a business loop artifact here. And basically, now you know that winning with AI is not about prompting better anymore. Okay, so we are looking at the, the old ways. Basically, you're chatting, you're prompting, right? People get obsessed over what's the prompt library that you should have, right? And then you have to be— you have to attend the session, meaning that you have to be paying attention to the session itself. And then, you know, over time, you have to just continue to just repeat this over and over. Now, the system doesn't compound this way, right? At least, at the very least, if you're putting together some type of goal or some type of workflow, and you're just repeating it over and over. And in many cases, it could be you're setting up an automation in Codex, for example, or you're setting up a routine Or a loop inside of Claude code. That's really what it is. And people are like, oh, well, how's that different than the /goal command? And in many cases, I don't really think it's that different. I think a lot of people are just kind of spamming these commands now to get things done. And just keep in mind, these will be typically token intensive. But if you're using something like newest version of GLM, which has been— people have been raving about it, maybe you can run that on a local model and you can just spam the crap out of these things, right? This is the old way of doing things. And so these are the 4 traps keeping AI stuck in demonstration mode, which again, you have to kind of stay on top of it. The new way is you find, you detect the recurring workflows and you look for fresh market signals, right? So an example of this is we actually have one where every single week it's mining all of our customer calls and figuring out, hey, what are the things that should be made, that should be created into workflows, right? So look for calls, they'll look for kind of the gaps because we're talking about building the services as software firm, like this AI-native firm, but why in the world would we still continue to sell retainers the same way? And this forces the muscle, this forces leaders to— sure, it can cite all this data, but once you find this stuff, and assuming that you have the right leadership in place, well, then you're going to be able to build the next step, right? So generating the first useful asset or action from the signal. For me, when we generated one of these, we basically looked at the Gong calls and I was like, oh, well, it looks like these clients right now are really healthy right now because they do really like the way in which you are helping them on the AI side. These are AI-native conversations. And these are some of the single-brain clients that we've had, right? Versus on the other side, you can also see the ones that maybe aren't so happy right now and why they aren't so happy. And this is how we can figure out the workflows that we need to be building for them, right? So again, this is, this is where we start to build things, right? Whether it's pages, content, or outbound tests, or even just changing workflows for, for our team, or just building these workflows out and say, hey, Mr. Client, we're going to do this for you for free, but if you want more stuff like this, just let us know, right? And then, then we're compounding our workflows. We're not just— it's not a very linear thing where if we add one customer, we have to add a certain amount of headcount to it, right? And then from there, it's your job to measure. So you're connecting the output to business receipts. It's not just about no longer about refactoring your codebase, for example, which is important, or looking for bugs to make sure you no longer have any bugs. Like, that's all important, but a lot of the stuff that's happening right now that people are talking about, I'm just like, why is nobody applying this to business, right? And so visibility, traffic, pipeline, all this stuff over here. And then what happens is, After you kind of made the judgment calls, then you would just make sure that you escalate this, you can, you can basically set the standard. And so talk about evals, right, kind of defining what the definition of done is, then you can, you can just keep repeating this mechanism over and over. And so what I just showed you earlier is we have a content ingester. Okay, so you look over here, we have something that will ingest my content from YouTube, my podcast, it'll ingest it from other areas as well. Maybe I'm speaking somewhere, it'll ingest it from conversations that I'm having internally in the company, because those are typically the That's the best content that you're going to pump out there, right? And then you build, you measure, and you escalate, right? So these are— this is what like I have in my AEO and SEO intelligence loop. And this is just what continued compound over time. It's not just a content calendar. I'll give you another example. One thing that I set up was not just a meeting preparer. So that's cool. But if it just looks at your Google Calendar or whatever calendar app that you're using, that's okay. But I want a pre-meeting context filler that will basically Look at Gong calls, HubSpot, it'll look at this just for anybody on the team, right, for themselves. It'll look at the emails that they sent too. So based on all the context that they have and whatever context that they found out about the company or whoever else they're going to meet with, it will suggest next steps to make the most of that meeting, right? So now you're not just having a meeting prep thing, it's the ultimate meeting prep thing where you have maximum context going into that. And then if you have maximum context, then that means if you're just one step ahead or you're two steps ahead, whether it's a sales call or recruiting call, whatever it is exactly, You're going to get bonus points for going the extra mile, right? And a lot of the stuff that happens in life when it comes to relationships is about just going the extra mile. And another example of a loop would be— we call it the resolver. What the resolver does is, for example, I use my Hermes agent, I use Claude Code, I use Codex. And the challenge with using all of those is that I start to lose contact. I start to forget about all the threads that I have. I start to forget about all the outstanding items, and some of them might not be important anymore. But the challenge with that is If I start to forget, then that means work doesn't get done, right? Ultimately, you're making a lot of these things into loops, but not everything should become a loop, right? I think when it comes to engineering, you can have a lot more loops, but when we're talking about business, for example, and maybe more judgment is needed, you're not going to try to loop everything, right? But you do want to continue to move the ball forward. So maybe at the end of the day, a resolver, it helps you prioritize and deprioritize things. It helps you figure out what work units should be consolidated, deleted, Or updated, right? Or even maybe you were creating net new work units there. So I find that to be very helpful. And because I have that problem, I'm sure many of you watching this right now also have that problem. Like, how do you just continue to move the ball forward, right? And then this can also be Slack or email things that are outstanding that are really important or critical to your business. So that's that. If you're enjoying this video right now on creating loops for business, this is exactly the type of stuff we do for clients. If you go to singlebrain.com with the B, you can find out more. Again, singlebrain.com, and we'll see you On the other side, everybody likes a 2x2, right? So you have a 2x2 over here. So start where recurrence meets measurable value, right? And so if you have something that is, let's say, low judgment load going to high judgment, that's down here. And then you have high business impact, right? If we are talking about something that is a strategy, like a strategic call that you have to make, so high judgment and low repeatability. So AI can help you brief the decision. So you can have a human in the loop here. It's high business impact, but it requires high judgment, right? So this is where you're putting something together initially. And then you can also make a judgment call inside of Loop to help it continue to get better. Yeah. And then I would say that once you kind of made this call here, then you can figure out how you go about— you can— you've built this loop first, right? Then you have a human kind of make the call, but then you— it continues to inform this piece over here, which is the loop engine, which recurs over time, which is visible. It's tied to revenue, right? Example would be the AEO and SEO content engine, right? And then novel work where it's high judgment but low business impact. Well, maybe you're not going to do that as much as these other areas up here. So you want to stay high business impact because we're talking about business. Now, quiet ops, low judgment, low business impact, repeatable, low judgment, you run it with exception alerts, or maybe you hand it to someone else that's managing your AI automations right now. So, you know, for us right now, I kind of mentioned the AEO SEO pages. So we might repeat this every single week, right? It also depends on how mature your site is. Sales objection mining, that might be another idea that you might have. And maybe that's a daily one, maybe that, or maybe you have one for case study mining as well. And maybe that's happening on a weekly basis. So you have to define how often you want these to repeat. And then, you know, for AEO SEO, like what is a receipt? Well, it's pipeline. We want more pipeline. For sales objective mining, your win rate should go up, right? Monthly board memo clarity, right? And then new brand narrative. This might be just like a brand new strategy, qualitative, right? Which is more strategic calls. So like you have a framework here on how to go about thinking about these loops, not just for engineering or building, but for business as well. And once you're able to apply a lot of these concepts over to business, It's going to be really hard to stop you, right? So the loop is real only when it has receipts. So we're talking about measuring business outcomes, right? Like, I want answer visibility to go, I want qualified pipeline to go up. Otherwise, this is a bunch of theater. And, you know, people are talking about, uh, no, it's no longer about token maxing anymore, it's now about token minimizing, right? And so I don't know if I buy either of those. I think it's just about getting real business impact. Humans touch the standards, thresholds, and exceptions, thresholds and standards. And you can see here, If a loop proposes, if a rule is hit, then basically the whole idea here is that it's just going to continue to work, right? So it'll ship a test and you have human judgment and then it just repeats over time. And then some things maybe you might not need that much human judgment after you repeat it for a while, right? But I don't think it's a very smart thing for you to just go about and just YOLO with these things in the beginning because you can waste a lot of tokens and also you end up, it's a bunch of theater because you decided that you didn't want to put the time into it. You decided that maybe you had Maybe other things that were, that were higher priority, which is completely fair, right? But in many cases, this should be solving your stuff, right? So there's a worksheet down here. You can screenshot this if you want, but the whole idea is find, build, measure, escalate. This is the worksheet. And so you want to find the workflow. That's number 1. So what happens every week and still depends on a human pushing it forward, right? Does it have repeated inputs? Has a clear owner? Has a measurable business receipt? Uh, number 2, build the first asset. So what output can AI generate that's even useful? That is even— that is useful even before the loop is perfect. All right, draft a page, campaign analysis, or action, which is what I just showed you earlier with the AEO SEO thing. That is very much— that's a draft, that's a packet for me to review first, and then maybe I have some feedback and go back and forth with it, and then maybe I might want it to go to production, right? Human standards documented. So again, you need some form of eval here, measure to receive. So what metric proves that the workflow improved? Ideally, you have one metric to focus on. So leading metric, business outcome, and then escalate judgment. What rule decides when a human gets pulled back in? And again, that depends on the type of work that you're doing, risk threshold and approval path, right? By the way, like, even this artifact that I made for you over here, like, this is a skill that I use. And what happens is every single week I have a loop that will— I was on my way to lunch today and basically have a content machine interviewer and it'll interview me on the most interesting thing that I've worked on, right? And this actually, this is how this video came about. From me dictating that, and now I have a video, right? And then I even made an artifact with it. But I even expect that to become a little more, you know, a little more hands-off in the future. By the way, if you want to grow your AEO and SEO traffic a lot faster, very competitive right now, right? You got to use ClickFlow. So ClickFlow, it gives you 5 free articles to start with, and they're high-quality articles, and you should see for yourself. But not only that, there's actually an MCP and an API server that your agents can use to scale out your content. So it can connect with the analytics tools that you have. Maybe a certain level of SEO tools that you have as well. And by doing this, a lot of our customers using this right now have said that they have not only reduced their time to get this work done, but their traffic is growing faster. So again, you want high quality, you want to scale, you got to check out ClickFlow. We'll see you on the other side. Okay, so here's an example of a loop that I'm actually working on right now. So this is talking about building an AEO SEO content operating system, right? And the way this loop actually works is it will look through A handful of different sources like my Granola, for example, or my Gong. And then it'll use one of the tools that we have, ClickFlow, use the API key, and it'll also pull from GSC. So Google Search Console, Google Analytics 4, you can pull from Gong as well. So raw transcripts from sales calls or customer calls, right? And then you can also pull in the Ahrefs API for validation around SEO and AEO metrics. And so once you are, once you're able to validate this, it can, it's basically, here's like a packet, like it's just going to continue to run this every single week, right? So It's going to look at, hey, like based on what you're talking about this week or based on the, based on what you're, you're talking about from a business standpoint too, maybe you should go after these manage AI marketing agents pages or AI marketing agents pages or company brain pages or AI marketing agents versus marketing automation. We know these are all concepts that people are going to start to search for. And if there's one thing that we, that we've always done well over the years, it's actually using a workflow like this, right? The challenge with this workflow in the past is that you have to repeat it over time and people get busy with their, with having to do other things, right? And so you can see it's pulling from different, different sources over here. And then it's talking about strategic evidence and what we should go about, how we should approach this from a sitemap standpoint, you know, adjacent live pages, like what should we be updating, consolidating, keeping, right? And so all I'm saying to you right now, you don't even need to know AEO or SEO, but this is much more of a business process, right? And so this is helpful because once you review this and then after, actually right after this, I'm going to make sure that I have the right connectors and I'm just going to have this run over and over and over, right? So that is a loop. A loop is something that you don't, you don't have to re-prompt all the time. It's going to continue to work. And then sometimes it's still going to ask you for human judgment as well, because something like this, it still needs human taste. But the reason why Loop works really well for engineers, for example, is that you can say, you can define what the definition of done looks like. Okay. Let's say you are looking to refactor your codebase. Okay. So that's, that's one thing that, that's an endpoint that's very clear, or you're looking to, to, to get your, all of your pages, you want them to be less than 50 milliseconds, for example, right? Those are all very measurable, verifiable things that can be done. Now with this type of stuff, maybe not so much, right? Maybe like for this, you can say you want to have— you can give the LLM some judgment, but the more judgment, the more clear you make it, the better, right? And so each and every week you're making it a little clearer. Maybe you want to optimize, or maybe you want to optimize 100 pages, or maybe you want all your pages to be under 50 milliseconds because that helps with SEO. But for me, as an example, maybe my definition of done in this scenario is I want to make sure that we're scanning through the entire website to find, and then also my content ingester to find opportunities to update, delete, consolidate content, or to create net new content, right? And figure out how to optimize things from an AEO standpoint to make sure that, you know, our things, do we have a good schema markup? Do we have FAQs at the bottom of it, right? And as AEO changes over time, how do you continue to adapt? And so, you know, I think everything with loops right now are that it's basically on the way to recursive self-improvement, which is where these things can get better over time on their own. But that's why a lot of people are talking about not prompting anymore and doing loops, right? So What I just gave you was an example. And so I actually want to hear from you in the comments. Do you like this style of video where I've done a little more prep with the AI help with my little mini loop here? So that being said, hope you enjoyed this video. We'll catch you in the next one.
