Fable 5 Is Back, Who's Allowed to Spend on It, and the Operating System for Running a Company on AI
Eric Siu walks through Anthropic's Fable 5 relaunch, the 50% usage discount, and how to build AI loops that generate revenue. He shows concrete examples: sponsorship-deck automation, website-conversion optimization, and overnight evaluation loops. The key is moving beyond one-off prompts to reusable systems that amplify intelligence across your team.
Chapters
Fable 5 is back for a little bit. You get 50% usage until July 7th. I'll explain what that means in a moment and I'll explain some prompts you can use or how I'm using Fable 5 from a business standpoint in terms of thinking about generating more revenue. We'll show you that in a little bit. I'm going to show you some cool stuff we're building right now. We're going to talk about the difference between loops and— because everyone's talking about AI loops, agent loops right now and everyone thinks they know, but they don't know. So I'm going to show you how that works as well. I'm going to show you a desktop app I'm using to make my life a lot easier. And I'm also going to show you some other stuff that we're building in terms of just maximizing the use of your tokens. And I'm actually interested, those of you in chat right now, if you can drop it in chat, one, where you're coming from, but also two, I want to know what What harness you're on right now. Are you using Claude? Are you using OpenAI? Are you using something else? What's your primary harness? Then we can go from there. But key thing is Fable 5 just came back 2 days ago or so, and then Anthropic's allowing you to use your Claude usage, 50% of the usage until the 7th, and then it's going to move into usage-based pricing. I thought it'd be helpful to start talking about this today. Because the reality is I think people are, it's like, oh, okay, you get busy, right? And one thing I'm about to show you, it's just amazing how I'm able to sit on my couch and build that. And what I'm going to show you is it just shows you the possibilities here. So we'll get to that in a moment, but maybe I'll start off with talking about some of the things that I've been prompting my My Fable 5, right? So I'll give you an example here. So yesterday, here, I'm going to share my screen. So if y'all can see this, let me do this. Share screen. Let's move to this window. Okay, great. So literally yesterday I was just, I was going back and forth, right? I'm looking for Fable 5 revenue use cases and my original prompt for this, by the way, you guys can take a screenshot of this, but I just said this, here's my prompt. Okay. Knowing what you know about how I work, my goals, my repos, what would be the best use cases for Fable 5 to maximize revenue? Rank them from top to bottom and include my ideas such as looking to finish off my projects, rebuild them using a Fable 5 lens, looking for technical blockers and more. Ideally only things you can do that other models can't. I think it failed here, but I was like, hey, what about the best loop we can run such as security scanning for our apps and repos? How about XMCPs? I'm just asking a bunch of random questions. I think this is 16 hours ago. Afternoon yesterday, I think. There's a couple of things. Yesterday what it did, it actually ran through a security scan in my repo and it caught a couple of security issues. That was good. The one thing you should note is anytime it senses any bit of security, because remember the government shut this down for a little bit, it basically kicks you back over to Opus 4.8. This chat got kicked back to Opus 4.8. But there's a handful of things over here that it's like, oh, you should finish this. We have this cold email engine which combines with what we use for sending cold emails and creating infrastructure instantly. It also combines sourcing and enrichment with it as well. It's saying, hey, here's the last mile over here. Because my whole thing is when you're building a lot of these things using Claude Code or Codex, oftentimes it's attended. You have to keep going back and forth with it. Then you might think you can just spam /goal, But you might not have given it enough context. The thinking here is that again, you lead off with a prompt like this. Again, knowing what you know about how I work, my goals, my repos, what would be the best use cases for Fable 5 to maximize revenue? Rank them from top to bottom, include ideas such as finish off my projects, rebuild them using Fable 5 lens. Ideally only things you can do that other models can't. Some other things like standing security over here. Or this one was big for me, activate the 8 dormant revenue loops. When I was using Codex, I had to actually ask it to create a handoff document. Let's say you're working in another harness, you can just say, hey, what are my top projects right now? Can you create a handoff document so we can move it to another harness? Because Fable 5 right now is the most powerful model that's available for everyone. ChatGPT has 5.6, but it hasn't been made available to the public yet. All that to say is I was looking at this and then I just kept going down here. I'm not gonna show you all the details, but I was just like, hey, I'm about to do this webinar right now. Keep it to 5 tactical things that people should think about. 5 tactical things for your audience to walk away with. Audit your last mile before you generate more leads. Find out where the point activity dies. Reply that nobody routes. A form fill that doesn't alert anyone. The gap is more relevant than more traffic. This goes with our technical sweep. Let's say you have a website. My site, singlegrain.com, which is marketing services. We're constantly— I have a routine that sweeps my website every single day and it looks for where the forms are broken. Maybe thank you pages are broken or links are broken and it calls it out. Or if it looks for maybe there's a fatal error, maybe the website's not working. That's super helpful. There's a lot of things that I end up building because I'm using Hermes, because I'm using OpenClaw sometimes, because I'm using Claude Code, because I'm using Codex. I'm on all these different harnesses that sometimes I forget what I'm building and sometimes I end up double building and that becomes a problem because I might have 5 projects that are very similar and then one is way ahead. But what ends up happening is maybe I merged the newest version of it from a GitHub standpoint. Then if I merge that, then, oh, the newest version replaces the oldest version where I put the most amount of work on. You can see how messy that becomes. There's that. This running a recurring security sweep on your AI-generated code. This one's a no-brainer because Fable 5 is known for this. Just be careful that when it kicks you back, sometimes you might have to switch the model back to Fable 5, which is over here. You can see that. By the way, my usage right now, 34%. I have 16% more to use, I think, before I get shut off from this thing and then maybe I have to switch to usage. Again, you want to think about the things that it's good at doing that other models can't do well. You wanna make AI verify, not just build. So you can point at a system that you assume that works, right? So attribution tracking, a scoring model, and tell it to prove the numbers are real. And so that's a big thing because when we do marketing services for clients, they're always talking about attribution. Well, how do we know this is true? Well, how are we tracking that? How is that going, right? So if you can have security sweeps, you can have attribution sweeps, you can have revenue sweeps, you can have— there's a lot of sweeps that you can run that this would just make things easier. And then again, I caught, I think it caught, I didn't catch it. Fable 5 caught it probably, I think 3 or 4 major security issues in one of my repos and it fixed it. So that was pretty good. So the problem is again, okay, so kill everything scores the same. So if your CRM or lead system rates every prospect alike, your team is guessing. Make scoring read signals, read real signals so effort goes in the right accounts. So one of the problems we have, you might think, You go a year back, you look at sales operations as a role, or you look at business development representative or sales development representative. Well, one of the challenges is keeping your CRM hygiene high. Well, I mean, this is stuff that you can just do a CRM sweep now. Every single day you can do a CRM sweep and just make sure you have a human in the loop that's checking. But again, I think the takeaway from this is what kind of sweeps can you run that maybe weren't that reliable before? Again, I would apply this to your repo. You can see here I've connected this to a handful of repos up here, and then I'm just asking about, hey, what can we do better here? There's that. Then I think the other thing too, if I'm looking at, I think maybe this one over here. Let's see what this one is. I'm asking this question because I'm tired of having to be involved all the time. What's stopping our loops in this repo from getting out to production? without having me consistently be involved. We can talk about loops for a second because I literally have a repo called Loops Creator. I'm going to switch my screen and then explain the difference between AI loops versus goal commands because many people still don't have it right. I thought I had it right for a while and I didn't. Now I can say I have legit loops and maybe I'll show you what one looks like. But let me share my screen over here. Okay, so I made an artifact on this one. So basically agents can help you with tasks, loops can run parts of the business. Okay, so if I bring you to this diagram over here, this is the difference. An agent is maybe it's a prompt, which is prompts are usually one-off and it's hard to remember these prompts. Prompt library, like nobody wants to do that. A /goal command, you can You can spam /goal and a goal will get you to an outcome. But the thing is you have to, you still have to be, it's still a little attended there. Tools. So you have to have different connectors. Maybe you're connecting your CRM, maybe you're connecting your SEO tool, maybe you're connecting your analytics, maybe you're connecting your call intelligence. You might have different artifacts. So these might be different These might be different processes that you have, different HTML documents, different MD files. You can have an agent in here. The way I look at an agent is just, do you actually have the AI thinking in the workflow? Is it thinking? Then basically, do you have it recurring? In many cases, are you just running a recurring? Are you adding a cron job and you have AI thinking there and you're just having it do a task? That's how I think about an agent. But oftentimes it's It's different than a loop because a loop involves looking at a target state. Let's say you want to get a website, you want all the pages on your website to be 50 milliseconds. Then you might say, okay, well, that's my target state, but my current state right now is all pages are actually 150 milliseconds. The gap is 100 milliseconds because 150 minus 50. Then you act and record trace and then you learn, stop, or escalate. We're talking about repeatable leverage here because Originally when I thought about a loop, I'm like, oh, it's just a /goal command. I think people are just, they're confusing it because nobody knew how to explain it. And then I started working more and more. I interviewed Matt Van Horn who had a great article that got 4 million views and the title was WTF is a Loop? You have creator of OpenClaw talking about it, Claude code creator talking about it. And I'm just like, dude, but it can't just be a /goal because then it would just be called a /goal. And so I dug into it more because nobody was explaining it well. And I'm like, oh, this is actually really good because sure, you might have /goal inside of the loop to help you finish something off. But if you're going to look at a target state, you're going to observe it and then you're going to evaluate. And if you're going to learn, you need to have some type of repository where you can store the learnings. And a /goal doesn't have that. And then everyone talks about recursive self-improvement, recursive self-improvement. Well, this is on its way to recursive self-improvement because I think yesterday it's— I had, I think, let me see if I can pull it up in a second, but I'll pull it up. But basically I have this— what did it send me yesterday? I'll pull it up in my Codex. But basically it was really cool because I saw it fix 4 different things. It's like, hey, these are the people you should follow up with and I fixed these 4 things. And I'm like, oh shit, this loop is working. Right? And that's pretty cool. So I'm going to try to find— I'm staring at my Codex on the right side over here. Hopefully I can pull it up for you guys to give you an example. But once I understood this, last weekend I went to town and Max from the team, he's behind the scenes right now, but I'm like, man, how many loops can I make just to make our lives easier when it comes to our content production? But then I started thinking about, oh, how can we make it easier from how we do client services? Because we serve clients helping them with AEO and SEO, right? But How do we just make that easier? Because on one end, when we help people doing single brain implementations where we have these managed marketing agents, the conversations are so collaborative. How do you do more? How do we do this? What about this idea? What about that idea? Versus like an adversarial relationship that you might see on the agency side. It's constantly like, what have you done for me lately? Why is it so late? You're not innovating, you're not proactive. Which they're right, they're right. But when you have stuff like this, when you have loops that people can run and you have skills that people can repeat over and over, Then you have true leverage, right? And then it's just a matter of finding the right people that are open to playing the game this way. And by the way, if you're watching your leader right now, if they're not down to play this way, you got to make a change at some point, right? And I think you want to do it with grace, obviously. I think it's— nobody ever takes pleasure in having to move on. So I would just say that you want to do it with grace. For example, we have people on the team where we might have had to move on from them, but we found them another job, right? I'll just leave it at that. I think we're going to net-net end up hiring a lot more people long-term and as long as they have these powers. I think eventually many people, almost all people will have these powers and know how to use this stuff. It's just going to take a couple years. Real loop architecture. I'm going to move back to— I'm going to show you an example and then I'm going to show you the thing that I built using Fable 5. Then I'm going to talk about the implications here because I want you all to think bigger about this when it comes to using these powerful models. You don't need to— I don't know if you guys saw that meme where it's like a guy's smoking a big-ass cigar and there's a huge lighter. It's like most people are like, you don't need to use Fable 5 for doing a quick calculation or something like that, or doing a search on Google. You don't need to use something like that. People are using this stuff for things that are— when Fable 5 is overkill, when you could probably just use Sonnet 5, which just came out recently. Real loop architecture here is, and I think it's important for you all to understand this because once I understood this, it was a big switch for me. I was like, that's why I probably have 15, 20 different loops that I worked on last weekend and I did it. What'd you do this weekend? I made loops. That's my going out. That's my partying. Anyway, real loop architecture. Target what good looks like. You have to define what good looks like. Observe. Reality from test, metrics, CRM, support analytics. Evaluate. You have a rubric, metric, judge, or deterministic check. Act. Edit, research, score, draft, test, route, notify. Then learn. State, trace, score history, reusable pattern. Then decide. Decide to continue, decide to stop, decide to change the strategy or escalate. For example, I built a loop for our singlebrain.com website. That's our managed marketing agents. And for that one, I'm like, okay, there's not that much risk for me. And I know there's at least a couple thousand visitors to that site per month. So it's a small website, but it drives leads for us. And I'm like, what if I just set the loop to try to optimize our conversion every week? The goal is to get the conversion rate an increase of 30%. And I think right now the conversion rate is maybe 1.8% on the site. So how do we get that up 30%? So over the next 4 weeks, it's going to be running a test and it's reporting back to me every single week and it's telling me what it's learned. And that's the key thing. It's not because if you just say every week run a /goal and it like, oh, optimize my site and increase the conversion rate by 30%. What if it just keeps writing the same thing over and over because it doesn't have memory? Anyway, all that to say is that's why we have a Loops Creator skill. If you guys want that, I think we're making it available so you guys can comment loops and we'll see how we can get it because I don't know what— I can't see the other channels right now, but you can comment loops if you want it and you can just take it. We have a Loops Creator and then I think there's a repo template. Max, I think we're figuring out how to give that away, right? Okay, great. Yeah, you can just give me a thumbs up too. By the way, the rest of you, I know you can't see Max, but Max is helping behind the scenes. Anyway, I'll call out this piece again because I think it's helpful reminders here. Target state, current state. You guys should just take a screenshot of this if I'm being honest. Take out your phones. Target state, current state, observation source, evaluation rubric, action policy, trace, learning store. Stop rolls, human gates. You don't have this in just a slash goal command. So you need some type of skill that you can use. And at least for me, I think about AI as it relates to business. That's my content angle. I don't see many people talking about it in this way. Anyway, all that to say is hopefully you guys took a picture of that. This is where it shows up in business. These are some examples. For example, coding is very verifiable. You can observe tests, logs, screenshots, you can evaluate pass or fail regression risk. You need an eval to define the definition of done. Then an acting, you can edit, trace, repeat, test, and then you can escalate when we're talking about authorization, if we're talking about billing or production data. You need to have rules here to make sure that you don't just burn the house down. Anyway, I'm going to show you some examples here that I've been that I've been working on and let me just show you. Okay, let me see. There's one I had yesterday. Okay. Oh yeah. Okay, here, I have a good one. I have a good one. Okay, so let's see. Let's see. Hold on, let me pull it up. Give me one second. Okay, let me share my screen. Let's go over here. So Let's go to Codex. Okay, here we go. Okay, so I have a loop here called, it's called Mr. Siu Sponsorship Deck, okay, or Mr. Siu Sponsorship. So basically when people reach out for sponsoring the YouTube channel, for example, one, I don't want to deal with this, okay? So here's an example yesterday where, let me just go to the top here. Okay. So you can see all these, these are all these paid opportunities here coming in for people that want to sponsor the podcast or the YouTube channel. So one, I want to focus on my business too. I don't want to deal with this. I also don't want anybody else on the team to really deal with this. So literally all it says, it comes up with this messaging over here, right? Thanks for reaching out. Mr. Siu's current queue is tight. So he's only reviewing fixed fee sponsorships at $25,000 for one simple post. If that works, send the product, campaign timing, usage rights required, talking points, platform, posting date, and disclosure requirements, Mr. Eric Siu's team. Okay? So now instead of me having to negotiate with them where it's all this back and forth, it also elevates our positioning where it's like Mr. Siu's team is managing the sponsorship, right? And I also like, they're going to say, oh, our budget for this is $100. And then I'll just respond back, Mr. Siu will only take $25,000. Let us know if this works or not, right? And you might be thinking, who's going to pay that? People do pay it, number one. 2, I don't want to do this. Even if you pay me $25, I still don't want to do it because I have other STDs, shit to-dos. All that to say is I'm just like, go ahead and send this all. The cool thing here, this is what I was looking for. Tuning applied. Added search terms for rate card, audience insights, screen recording on creator selection, aligned source copy. Around one simple social post, added a dedicated video bundle usage rights across Scopecape. I'm laughing right now because this is just like, this is so cool. And now think about all of the other stuff that you don't want to do in your business. How do you loop? Where the tuning applied piece is the coolest piece. When I saw that yesterday, I was like, that's amazing. I'm not going to cuss. I'll just say, F yeah. Now it's just running every single— it ran again today and it's just like, oh, here's your other queue over here. Here's what you should be focused on. I have all these loops over here that will tell us revenue command center, what's on track, what's not on track right now, which deals are stale right now, which deals are stale right now, and then it's gonna improve over time. Also, which deals do we lose in the past and what's the follow-up messaging for all of them? And then I can just approve and send all of them. All that's to say is you don't have to build these loops inside of— you might be thinking, oh, it's too complex. I need to build inside of a terminal, inside of a CLI. No, you don't. You can build inside of Codex. You can just ask it to repeat. That is a loop. You just need to make sure that you're using a loop creator and that loop's repo to store the learnings, and you just keep doing this over and over. I would highly recommend you all watching this right now to think about what are the things based on— and maybe when you Maybe go to my YouTube channel afterwards, take the transcript from this video, and then think about, plug it into whatever harness you're in. What can I make into a loop based on what you know about me, based on what my goals are? And then just go from there. Okay? So anyway, as I'm going, by the way, you guys can drop questions in the chat. I'm just going to keep going here, but hopefully you guys find it useful. And I'll give you another one over here. I was building this AEO Content OS, AEO SEO Content OS. So what does it do exactly? I've been trying to hand this off to, in fact, I did hand it off. I handed it off to one of our engineers working on ClickFlow. And in fact, he asked me for it. He was like, to his credit, he's like, hey, can you just send me that Content OS that you're working on right now? So once I sent this over to him, he's like, oh, I get what you're trying to do. So maybe what I have is maybe 70, 80% there. Maybe I'm the prototyper in the company and you might have multiple prototypers. So 70, 80%. Okay. So if that's the case, when he looks at it, he can make it a 9 or 10 out of 10. And so what this does is it will ingest all of my content. For example, what I'm saying to you right now, okay, there's probably a few spikes within this chat I'm having with you right now. My long-form YouTubes, my internal Gong calls that I have, which probably when you look at, because we try to have Gong, which is the revenue intelligence tool, it's just a call recorder. But when we have that sit in all of our calls, that's often when I have the best spikes, the most intense, the most inspirational moments actually come from you and your internal meetings. At least that's the way for me versus me preparing to make content. But the whole point here is that ingest content and then it creates, then it thinks about, oh, okay. What content should we be creating from an SEO and AEO angle for our website? And then it can draft them. Whether it's like, oh, Eric's been talking a lot about enterprise marketing agents or AI managed AI marketing agents or agents that live inside of Slack or talking about this company brain or talking about the single brain. These are all concepts that don't have a lot of search volume right now that will take off. Some will take off. And usually if there's one thing that I'm good at, it's usually being on top of the trends. How do we take advantage of that from an AEO, from an SEO angle, and then leverage tools like Ahrefs API, like the ClickFlow API MCP, which is free by the way. You can use that to create content. That's our own. You can use that to connect with your analytics. You can use that to connect with your revenue intelligence tools. You can use that to connect with your CRM. And then you'll make much better content once you daisy chain a couple of these skills you have together and then make them into a loop. And that's why this is so powerful. And then what I did afterwards, because Fable 5 came back yesterday, I'm like, well, hey, give me a handoff document so I can move it over to Fable 5. And by the way, what you're seeing on the screen over here, this, let me see if I can make this. Okay, you see this over here? This is not bad. This actually looks legit like our website over here. It actually did a better job than Fable 5. These are actually previews of these pages. And I said, I worked with it a little bit. Hey, I want this. I want the Our little video preview over here. I want that. And it listened to all of my feedback and made the changes. And now I'm ready to deploy at scale if I wanna deploy this at scale. If I wanna go, let's say we wanna test with 50 a week. Okay, then it's gonna measure the results every week and say, okay, here's what worked, here's what didn't. Okay, let's try again. Here's another 50. And then maybe after a month or so, I feel really good about it. Let's go hundreds. Maybe don't go that hard, but I'm just saying. Depending on the strength of your website, depending on what type of website you have, you can do that. And so my whole thing here is you use loops, you're going to be able to scale more. You teach, not only if you use loops, that's one thing, but if you teach your team how to use loops, that's another thing because then they know how to scale. They know how to think like you. They know how to move a lot faster. And the whole excuse, it's funny right now because Jensen Huang said a few months ago, Oh, that's too hard. That idea's out the door. That's going to take too long. That idea's gone. All the excuses should evaporate. And I will say that if you have anybody doing this stuff right now, they won't make any excuses like that. The only thing that I'm worried— again, I've said this over and over. The only thing I'm worried about with people who are AI-pilled at my company is burnout. That's the only thing I'm concerned about. Whereas it's a different set of problems for the people who aren't as much. But anyway, you decide the problems that you want to solve at your company. Anyway, I think that's a good amount on loops already. You can see I have this overnight eval and this will make sure that I'm continuing to use tokens through the evening and it'll look at what else is outstanding and maybe evals we can upgrade throughout the task that we're doing. Again, these are all things that I'm testing over here and my usage, 95%. I have a lot of usage left and I have 2 resets too. I love, by the way, I use both Codex and Claude Code. Codex is my daily driver because the usage limits are much more generous. When you pay $200 a month on OpenAI for the max, I think you get $14,000 basically in tokens if you're using it how I use it. Whereas for CloudMax, if you pay $200 a month, you're getting 8,000, right? And I do know that with Fable 5, I'm going to have to switch. I'm probably going to swap myself back onto APIs. But maybe if you guys want in the future, maybe we can spend an episode, a webinar talking about token optimization because I have a lot of friends who are spending millions of dollars a year on tokens. I'm also sitting in a lot of AI chats right now with CEOs. that are just like a lot of people are considering open source, right? And then how do you actually go about doing that? And so that's happening for our clients now on the Single Brain side. They're like, hey, can you actually help us with token routing? Can you actually help us with gateway optimization in terms of making things more secure? So anyway, that's a little— that's kind of a preview of where things are going. So let me show you what I built with Fable 5 just to show you guys the possibilities here. Because my whole thing is I don't want to use a bazooka when I can just I don't want to use a bazooka in a knife fight. I don't need a bazooka. It's too much, right? So I'm going to switch my share to something else. Let's share this window over here. Okay. So, and Lawrence and Aram, I'll come to your questions in a moment. Excuse me. So what I'm showing you on my screen over here is a game that I made. And not like I played a lot of games growing up. I literally wrote a book on how I think everything in life is just a video game. And so I'm like, how do I make a game that's kind of like an MMORPG, but also like a strategy game as well? Because that's what business is. So business is the ultimate game, right? So it's like your business is a living town. So this is, I'm onboarding with you guys right now for this little thing. Every building is a business function: finance, sales, demand, product, customer health, market scouting. Their health is computed from your operating data mocked in this demo. So imagine if you can connect— and don't worry, I'm not going to sell you this right now. I'm working on the couch on this yesterday, right? So their health is computed from your operating data mocked in this demo. So imagine you can connect your Ahrefs, like SEO tools, analytics, CRM, revenue intelligence, finance data. The more connectors you have, the more valuable this becomes. And I think this is helpful because I've always enjoyed— the reason why I was addicted to games is because I always looked forward to the next thing to do. I was always active, right? So the banner tells you where to look. So green is thriving, amber is constrained, red is burning. The daily empire briefing turns the map into one verdict and 3 quests for the next 7 days. And fog means missing data. So buildings with no connected systems stay fogged. Connect systems. To lift the fog and raise diagnosis confidence. Connectors buy visibility, never vanity. So we're gonna enter the town over here and you can see this. We have a finance base, a sales engine. We have a demand tower. Our marketing scouting tower is fogged right now 'cause we haven't connected that. I'm trying to make myself a little not as dark, but anyway, customer health hall and we have product factory. Okay, so you can see what's yellow, what's red, what's green. And you can see daily empire briefing over here. Activation leak at the product factory. So only 34% of new signups reach first value versus your 45% target and it's slipping. So what do we do? We view it on the map. High confidence that this is correct. This is burning right now. Your target, 45%. So we have a target that maybe we set internally as a leadership team. 4-week retention, week 4 retention, 71%. PQL to SQL is 11%. This is pulled from Mixpanel. So what do we do here? And maybe we can go a little deeper into this, but you might be wondering, why am I even showing you this right now? I'm going to tell you where I'm going with this, because the way I see this is this is where you use the real power of a Fable 5. Because I think I might've used like 1 or 2% of my allocation, I guess. But you can see here, oh, this is, by the way, this is all dummy data, but let's say your company's doing $4.2 million ARR. Here's your pipeline over here. Here's your runway, 20 months. Right now you have a city. You went from a town to a city and you have $1.9 million cash in the bank. So you can decide how you want to play this game because business is just a game of resource allocation. It's no different than playing a StarCraft or a Warcraft. And maybe I'm dating myself from these games, but it's no different from these strategy games. Oh, quest board over here. You can repair the activation funnel, accept the quest. Okay, so this is high impact. You have 5 to 7 days, high confidence. So guess what? Someone on your team can choose to accept this quest and maybe you put a bonus in this quest too. So maybe you tie a smart contract into this, you have stablecoins involved. And so let's say maybe Max wants to do a lot more work right now. Maybe he has a lot of free time. Maybe he wants to do all the quests. Maybe each quest has a bonus of, I don't know, $50, $100, maybe some are $500, maybe some are $1,000 or so. Oh, he can decide to take that and he can take as many as he wants. And guess what? Guess what? He's going to show up on the leaderboard. Everyone's going to see it. You're also going to see who's not taking as many quests too. Maybe someone else from another town, another business, they can see available quests and maybe they want to take on some of your quests too. And so, oh, what's the use of crypto? Stablecoins. Okay? Stablecoins, smart contracts. And then so you can escrow all that money and then reward it. And then you can celebrate in front of everyone and stuff. And maybe some quests are repeatable. But do you guys see where this is all going? So this is a game for us internally. So maybe we're thinking of Stripe, Gong, all this other stuff over here, but your entire company becomes a game, right? So, and you can see who's the most engaged in your company. You can see who's not engaged. Guess what? The not engaged people, I'm sure they can do their best work somewhere else. Maybe it's just not the right fit. Maybe they don't like working at your company, right? Or maybe you need to have a conversation with them to coach them up. Maybe they have something going on in their personal life. I have empathy for that. But my whole thing is, I like to say that, my friend likes to say AI is intelligence amplifier. AI is actually IA, intelligence amplifier. But I also think it's a truth revealer too, because as more things go towards this, when you make things experiential, you make things a game. And I truly think that this is the way that a lot of companies are going to work just to make things a lot more fun because we live in a remote environment now. But here's the thing, if this works for us, what's gonna happen? Maybe I'm just gonna give this away as a lead magnet. Maybe I want everyone to come into this world, world of business, WOB, not WOW anymore, WOB. So all that to say, look, you can run a stalled deal revival sprint, all these things over here. Yes, there should be more than 9 people. We're gonna do a better job of promoting next time, don't worry. And I'll tell you what, we have enough spikes from this content today that it'll probably do decently on short form, I would imagine, or maybe even mid form. So I'm confident this will be fine. But yes, Ayram, this is why we need skills for everything, right? Skills for promotion, skills for getting webinar attendees, skills for, I don't know, skills for AEO, skills for social media. But this makes it a lot easier to onboard someone to your team. That's why we have that Skills Dojo, right? That's something we're giving away for free. So some of you might not have seen this last time. I'm gonna show this to you as well. But the whole idea of having our Skills Dojo here is so people can pick up skills and you can see who's engaged and who isn't engaged. And our clients can use this if they want. There's a lot of interchangeable skills that we have that maybe we want our clients to use, or maybe we wanna use those for clients when we put 'em into like a private repo when we're helping them. But then, you know, when people come into our world, Maybe they can access the skills dojo we have, or maybe, you know what, maybe if they achieve a certain level, they have a certain amount of skills that it unlocks more quests. And so I'm constantly thinking about that. Maybe it's a certain amount of quests they unlock, or maybe like a certain amount of experience points that they have, and maybe it'll unlock the ability to go to the next town or trade with other towns, other businesses that are connected to this thing. There's so many things that you can do. But I was showing this to my team earlier today, this morning. I was like, guys, Just because I'm messing around with this doesn't mean this is a priority right now. Everything's a crawl, walk, run. The Skills Dojo we have right now, the API gateway that we have, everything that we're doing is leading us up to this point. And that means that we have to continue to prioritize correctly. We have to continue to talk to customers. We have to continue to be locked in when it comes to recruiting, and then we'll be able to do everything that we want. And then we're at a certain point now, when I think about this, I'm like, Oh, shit. We might actually have to raise money. And so that's why I'm spending a good amount of time talking to a chunk of VCs right now. So we'll see what happens there. But anyway, all that to say, one, you play this for— you use Sable 5 to build stuff like this. I saw a friend yesterday, he built— actually, he built a Super Smash Brothers game and it looked pretty good. But how do you apply Maybe just think about your childhood. What are all the fun things you did in your childhood that you actually miss, that you wish that you could apply to today? So for example, Max is watching this right now. Max likes playing basketball. How do you just make a simple app that coaches you on your basketball stroke? Or for me, maybe I want something that coaches me on my tennis stroke and watches how I get better over time and makes that a game, makes that a fun social thing that maybe you can open source. A lot of things that you can do. So you just got to think about one, what are you passionate about? And then 2, what do you do for work? And then what's the intersection there, right? So that's what I would say. I was the first comment giving that for free on YouTube channel. You can check on giving that for free. Giving what for free? Oh, I think it was eval skill. Okay. Have no idea what you're talking about, but sounds good. So by the way, we do have the Skills Dojo piece, what Max just linked in the chat. You can check it out. You can opt in for that. We're gonna give it away for free. It's gonna be public. And our vision with that is that everyone is able to upload their skills. The community can upvote, they can fork your skills as well. People can maybe comment on skills, they can review skills. We want everyone to have a profile there, right? And maybe that profile carries over to this. And guess what? Maybe Max's character as he walks around in this thing, holy shit, I didn't notice there's actually little characters walking around. walking around in this thing right now. I didn't notice that. That's cool. I didn't see that. That's cool. But anyway, all that to say is imagine Max's character. He has the best weapons, the best armor. Well, he got that all from doing damn good work, right? So we can animate all that stuff. There's maybe a place where he can trade, but you can see even the way I'm talking about this right now, this makes work worth— this makes work worth going to or more fun. So that's what I would say. Anyway, so I will open it up for questions. Maybe you guys can drop questions on maybe other stuff that you might want me to show. And in fact, I might show you a couple things just so we can get in for short form and mid-form as well, Max. But I'm going to answer some of the questions in a moment here, but let me get a sip of water. But hopefully you guys are finding this helpful with the 10 of you watching on this webinar right now. Okay. So let's see. I'm going to move over here. I'm going to show you some skills first. So the way we think about this Skills Dojo is we have these skill packs. And so when people join the company, we're going to have a lot more of these, but because this is still very new, but if people join the paid media team, there's a paid media pack. If people join the revenue team, there's a revenue pack. If people join the SEO team, there's an SEO pack. And so you can add these skills in there, but also guess what? All these skills have evals, which defines what success looks like for the skill. And every person's going to have to make a Loom video that the agent will then evaluate to make sure that you pass. And if you pass, then you've acquired these skills. And so let's say someone on my team, they have 100 skills. Well, you can see how many skills they have, right? Versus somebody else, maybe they don't have any skills. Well, that is— That's a coaching opportunity or that's a performance management opportunity. Either or, you get to the point a lot faster. And so you should definitely have these skill packs and then you can discover. And then so one of the ones that we have that we use quite a bit that I think we've since forked a few times is this X Long Form Post Writer. So this has helped us get to, let's say going from 87,000 impressions a month to maybe 3 million plus. And I think this is going to continue to grow because we're going to continue to make this X Long Form Post Writer better and better over time. The cool stuff that we're sharing here, we're going to share more numbers around it. Usually when you have numbers, case studies, things like that, it usually performs a lot better. You can copy these prompts and you can see how this is just the skill itself. Do you want to copy it? We have ASCII diagrams inside of it and this is going to get better. Maybe every quarter it gets significantly better. The whole idea is that we're building on the shoulders of titans. If Shubham put together this skill, I decide to fork it and make it my own and then someone else decides to fork it and then maybe that one gets That version gets a bunch of upvotes, that becomes a new canonical version. Show them can say, the original person can say, hey, that's the new version. We're just continuing to build on each other and this is all just becoming a game. It's all connecting to each other. This is one of the examples of these skills. One other thing I'll show you, I'm going to show you a couple of loops here and I'm going to answer some questions. Let me switch over. Let's share my window over here with Codex. So we have, let's see, we have a LinkedIn loop creator. Okay, this one's pretty cool. So you know what I did? I just said I found a post where someone wrote about how they get 5 million impressions on LinkedIn per month. And I'm just like, hey, how do we build this into a real loop? And then it starts going through it. Here's a spec, here's the post itself. It reads through it and I think it gives me an example too. So Let's see. It gives me an example. Let's see, where is it? Maybe this one? This one? Okay. Yeah, this is good. Okay. So this example over here, if you're looking at the right side of the screen, this is the template of what I would be posting. So let's say, imagine it pulls from my content ingester. So this is the LinkedIn loop creator. Okay. So the whole idea here is how do you go from 0 to 5 million impressions a month on LinkedIn? So you can see the format here. I built a simple AI revenue loop scorecard. It helps you find one AI workflow in your business. So giving away for free. And then here's some meat in the post. And then it's like, hey, comment loop if you want it. And this is literally the format that that guy was posting about in his X post. And then you can have different variants in here as well. And so this makes it easier for my team where maybe I get this 70, 80% there, I prototype it. And then I hand it off. And then we're just testing this over time, but we're testing the entire premise of the original post, which is, hey, post a resource that people will typically pay for, give it away for free, try to get 20 people to comment on it, and then you can use their software, which is what they're promoting. Use their software to reach out to ICP fit, ideal client fit people that have commented inside of LinkedIn. And this actually worked for me a long time ago, back when I was, Whatever I post on LinkedIn would get 50,000 views. Now I'm not so much in LinkedIn's favor yet, and we're going to get that back. But this is the whole idea. We use the Loops Creator skill and we use the repo template, and then we're able to create stuff like this. People are sharing a lot of smart stuff on X all the time. I'm just constantly dropping an Instinct and I'm just using the Loops Creator. How do we make more? How do we make more? How do we make more? By the way, if you want the Loops Creator skill, Comment loops and we'll send it over to you. See, look at that. That's a short-form video. That's a mid-form video for you right there. So that's one example, okay? I'm gonna give you one more. So let's look at the, what is this overnight repo over here, okay? So I literally did it again here. I think I was on my phone and I took 2 screenshots. What are you running overnight? This is what my friend who runs a company called 10X, He asked this during interviews, right? And I'm like, oh shit, I don't ask that to people when we interview them. And in fact, we should. So now not only did I learn during the interview process, it's not just show me what you built and let's nerd out on it. Show me your Codex, show me your Claude code, show me your Hermes, show me your OpenCLaw. Now it's what are you running overnight? And then it's like, okay, I thought this was insane. What do you do over here? And so I was like, Hey, I'm lazy. How do we just apply this to our brands? Tuesday, 6:25 PM. Oftentimes I'm just on the couch here. Here's— we have an overnight eval loop here for every serious AI-assisted build. What is the pattern? It builds with the best model first. It captures 100 to 200 real examples and runs overnight sweeps, tests cheaper models, prompt variants, routing logic, guardrails against the dataset, score on business outcomes, not just in accuracy. Track correctness, cost per run, monthly cost, latency, failure modes. Human review required regression risk. Okay, so so I think this ran yesterday. Did it run? Okay, set report back to me at 8 a.m. Okay, 8 a.m. report July second one time report. Okay, so what happened? So this was supposed to run, and then it was supposed to note to me like I haven't checked my other chats yet, but the whole idea here. is that you're constantly having this thing work even when you're not around. And so now I can confidently answer that question and maybe I'm going to go deeper on this, but I got 2 things from it. Not only do I have a question I can ask in my interview questions now, now I have an overnight eval loop that I'm working on. And I have no idea what happened. Okay, let's see what happened. So we're doing this real time. So the overnight eval ran at July 1st, which is yesterday at 11:00 PM. And the 8:00 AM report back generated July 2nd, 2026. Okay, here's the status. Success. It did 4 daily eval runs, mornings, 4 summary runs, and the result was 100% exact and it's now mechanically healthy. Okay, great. So I have no idea what it's talking about. So what are the implications? See, don't have to be super smart to just ask these questions. And then we're gonna see what happens and then I'm gonna get to the questions here. So while I wait for it, I'm gonna start to answer some questions. We're gonna wait for it and okay, it's answering. So then I'll get to your questions. Okay. So the implication is now we have a working loop but not yet a useful intelligence loop. What we proved, the daily automation works. Okay. What we did not prove yet, it's actually good for your business that this works on real emails and leads. Okay, great. Look, watch me be lazy. What's the /goal for this? That's all I do. What's the /goal for this? Then I'll look at it real quick and I'll just run it. Then once it looks good, I'll just say, hey, is this loop ready? Then I'll use my Loops Creator skill to move it forward. Again, don't need to be a genius to do this stuff. You just need to be very curious and you just need to be willing to test. I'm going to come back to this later and I'm going to start answering some questions now. Let's stop sharing. Then Max, I don't know if you're able to highlight these questions, but I'm going to start from the top. If I hit question, Lawrence, 22 minutes ago. Lawrence says, I've been trying to figure out where these loops should be running. My obvious options are Claude code with all the tools attached plus some repos for each loop. N8n, throw away N8n. Don't need N8n. Garbage. In fact, I'll leave it. Someone showed me NNN immediately. It discounted what I thought, my impressions on them. Number 3, a new vibe-coded app. So how should I go about with that decision? So if your harness, your daily driver is Cloud Code with your tools and some repos, I would just go with that. Because Cloud Code, the app itself is great already. That's why my daily driver is Codex and I feel like I can get a lot done. They basically copied each other in terms of the UX. That's what I would choose. Now, Aram, next question here. With loops, what are your recommended platforms to build loops, learn the MVPs needed to start building loops effectively, and number 3, scaling loop building in an effective way? The best place to build these AI loops, I would say, is just use the AI Loops Creator skill that we have. And the repo template. Once you use that, the way I have just shown earlier is you can just take an X post that you like, an Instagram reel that you like. You can take screenshots of stuff and just send it to your harness. And once you have the skill in there, you just say, hey, can we make this into a loop? Does it make sense as it relates to our business? How will this help us make more money? And that's all I do. And then Aram asks, what is the simplest way to get started with Fable 5? What quick MVP project do you recommend to understand its limitations and strengths so I know when to not use it? This for me, because I try to apply it to business, I just say, hey, based on everything you know about me, my repos and everything, every setup that I have, what are the highest leverage use cases of this with only what you can do? Because you don't want to burn tokens that way. And I know that I gave this answer a little earlier, but I want to answer this to the crowd and maybe this is good short form as well. But in my opinion, Fable 5 should be, it's the bazooka that you don't want to bring to a knife fight. I would say that you can ask it to do a technical sweep, a security sweep. You can ask it to look at the technical limitations of what you've built already with all the projects that you have in tow and just say, hey, if you were to rebuild these using Fable 5 capabilities, how would you rebuild them? Oftentimes I like asking that question. That's one of my favorite prompts. After something is built, I say, Now that you built this, if you were to rebuild it, what would you do differently? And oftentimes we'll get a better output. I also like asking, hey, I want you to put an expert panel together as it relates to whatever we're talking about and score this on a scale of 1 to 100. Only return it back to me when it's 95 plus and you'll get better recommendations that way. I like using the Grill Me skill from Matt Pocock. That's P-O-C-O-C-K. He has— so oftentimes I'll just say, hey, grill me on this, grill me on this. Because I know I haven't given enough context and so it'll do a good job of just constantly asking questions. It asks it one at a time, so it just continues to update its context as it's going. That's what I would say. Now let's see, Lawrence says, I think you're in ChatGPT as the interface, but where are the loops running? Sorry, redundant, but it's okay. If you want to get started with AI loops inside of ChatGPT, open up Codex. Codex is great for building stuff and I find it to be great once I have my repos connected, like my loops creator and my repo template, then I can just ask it to build stuff and I can see on the left side whatever it's building and then it'll just let me know when I need to come back to it and review it. And then I set the loop into a routine and that way it can loop every single day and remind me on revenue gaps, hiring gaps and things like that. So let's see, Lawrence says, One of the biggest challenges I've been seeing with Claude code, that's where I work, is it's all me. If I want other people to have anything to do with it, it's tough. That's fine. At least in the AI stage we're at. No, that is not the stage that we're at on the AI side. So what we do with Single Brain, which is our managed marketing agents, when we work with clients, we have a harness. We use Nemo Claude in this case, which is NVIDIA's runtime, and you can use Open Claude, or Hermes there, but it's connected to Nemo Cloud for security purposes. And then we have— we customize the agent for them and it's able to talk back and forth. It makes it multiplayer. And so you can build loops using Hermes, for example. It's a great harness. You can build skills. The skills are very— the reason I like skills so much is that they're portable, right? So I can take that same skill and then I can move over to Codex, I can move over to Claude Code and say, use that skill again. So that's why I think it's very important for you to make it multiplayer, but also have some type of skills dojo where you can share the skills with your team. Otherwise, I don't see a point where I become really strong and then everyone else stays at the same level. There's no point in that. So I will do anything it takes to help people on my team who want to get better. And I also will get very upset if people don't want to get better because it doesn't compute for me. It's like, how can you not want to get better? But that's a whole nother conversation. Let's see. Okay. I think that's mostly it. So for the 10 of you that joined this Riverside webinar, I appreciate it. And then Lawrence, let me answer your question here with Claude Tag. So, oh, this is great. This is great. This is great. Here's a great Midform clip for you, Max, and it may be a short form. Okay. Here's why Claude Tag is garbage for businesses. You might be thinking, oh, you can talk to your team in there and you're using their latest models. Here's the problem with this. One, you're stuck to using their models. Two, it's usage-based, so you have to pay them on tokens. You're not paying them per seat, so the cost can start to rise a lot and you can only use the Opus models, which cost more than the Sonnet models. The other piece is that, I think it's number 3. Number 3 is you are feeding all of your context to Claude, to Anthropic. And so if you want to come off of it, that context goes with it, or people leave your company, that context goes with it. You can't compound. You are stuck with their models. It's usage-based. And then the context piece is the biggest one where you're feeding it all to them. When we have single brain, which is what we do for our clients, one, they can use any model that they want to use. Two, we can optimize token costs. We can do that. The third piece, most important piece is that The context, their IP, what they're compounding on, that's for them. That's not for anybody else to own. If you want to have some sovereignty from these models, whether it's the big ones like an OpenAI or a Claude, you shouldn't be using something like a Claude tags. Maybe you can use it for the beginning and just to test it out so your team can get used to it, but you don't want to stay there forever. The other thing is you're going to be limited on connectors too when it comes to Claude tags. There's a whole host of things like we build API gateways for our customers. We think about the Skills Dojo. We think about all these other things. There's a lot of considerations for a company. And so you can certainly use it to try it out, but claw tags, it ain't good for business. Cool. All right. So Aram asks, I got Hermes up already. I've been using Telegram since 2014. What am I missing to have AI build a business command center like yours? Then create a skill loop eval process like you're teaching your team to bring my business to the next level. So one step at a time. So I started using OpenClaw in, I think late January. We're already in July. Holy crap. So 7 months, something like that. And so every single day I continued to compound on the scaffolding and I broke a lot of things along the way too. And I pushed it to limit. I found that I just couldn't use my OpenCLaw anymore because it just wasn't as reliable. Then I moved over to Hermes and I would say the big unlock, I was using Hermes inside of Telegram and that's good. It's good for maybe private conversations that maybe I don't want my team to see if I want that added layer. But the big unlock was when I moved it over to Slack because Slack is where my team hangs out, number one. So you can make it multiplayer. That's a big piece of it, but also the threaded conversations that you have inside of Slack. Whatever conversation you have, one's on maybe optimizing conversion rates, one's on optimizing sales, one's on optimizing SEO, one's on maybe helping with maybe your content generation, whatever it is exactly, it maintains that context. The cool thing with Slack is I can open all these chats into new windows so I can, on my other screen over here, I can have all these windows where I'm continuing to work on it. Also, Hermes has a desktop agent now or desktop app now that you can use where you can maintain things a lot easier. Which I didn't show you guys. Maybe I'll show it next time. But that has been a big unlock too. I think having the Hermes desktop app, just constantly building things and asking it to help you reprioritize and help you continue to focus on what you should be working on. I even have a resolver tool that will keep me focused because it'll look at all my harnesses and it'll say, hey, I've looked at everything that you have going on with Claude Code, with Codex, with Hermes. These are the things that you need to consolidate and these are the things that you need to work on. That's helpful for me. And I think we made that into a skill as well. But my point is with Hermes, once I moved into Slack, that was a big unlock because it became multiplayer and because you can have the threaded conversations and you can open up into different windows. And then we started to move a lot faster because a thing I keep coming back to is a tweet that I saw maybe a couple months ago. It was humanity started to move a lot faster when we started to trade ideas When we started to trade ideas a lot faster, the velocity of ideas, that's what allowed us to move a lot faster. So my whole thing is every turn that I'm able to have with my team inside of Slack, which isn't as reliable as I'd like right now, but some people can, they're actually experiencing this. I see the light bulb moments. If someone on my team asks, like Max asks a really smart question and then 2 other people see it and then maybe with the work that they're seeing in public with the agent, they come up with something that could help them work faster. Well, I like to say that we're all making idea babies a lot faster. And the more idea babies we have, the faster we accelerate. So that's how I see it ultimately. I think it's starting small first, pushing the limit, and then eventually moving it over to Slack to work with your team on it. And that's how I think you can maximize the use of Hermes. Then you can build that business command center thing. You're gonna use Sable 5 for that. But I'm so used to— I use Whisper Flow when I'm, let's say I'm walking, maybe I'm walking to the gym or something like that, or I'm at the gym, maybe not at the gym, but when I'm walking to the gym, I use the Whisper Flow app on my iPhone and I'm just talking to it. And maybe I'll have like 4 or 5 different thoughts and then it would just create it into different threads and then I can come back to it later and then I can move it over to the Hermes desktop app to continue to work. And then that's what I would say. And then by the way, that Skills Dojo that I showed you guys, that was me working with Hermes on it. It didn't take me that much effort. I put that repo together, I got it 70, 80% there, and then I handed it off to the team and then they've made it a lot better. So I think it's just a matter of being insatiably curious and then testing these things. Because again, I don't think I'm the smartest person in the world. I just do this stuff. So good that you got Whisperflow. That's great. Lawrence will answer this last question here. For the Slack Hermes link, where were you hosting Hermes? Local machines? I guess I'm curious about the Slack Hermes integration. Happy to look at existing content. I think we might have some stuff on my YouTube channel, but if we don't, let me know. I put it on a DGX Spark that you can see over here. If you want to set up Hermes correctly, I would put it on a local machine and then I would tie it in with Slack if you want it to be multiplayer. If you want to be single player, Telegram is fine. But when you make it multiplayer, again, things just move a lot faster with your team. Because the flow of ideas compounds a lot faster because one idea people see, one chat might give them 2 or 3 other ideas. So that's been really good for us. But I have 2 of these DGX Sparks. I'm probably going to go end up buying more local devices to have more compute locally because these token costs are going to just jump up more and more. So that's what I would say. So anyway, I hope you guys enjoy this. Let me know what you think and yeah, let us know what other topics you'd like us to cover in the near future. But we will see you next week. Next week, by the way, I'm bringing on Cody Schneider, my friend Cody Schneider. We're probably going to talk about AI as it relates to how he does it from a paid media side. So a lot on paid media marketing. So we're going to have a guest on and yeah, hope you enjoyed it and have a great rest of your week.
