The OpenClaw and Claude Code Money Making System (No Coding Experience Needed)

Eric Siu shows how non-technical founders can use OpenClaw and Claude Code to build marketing workflows without coding. He runs agents like Alfred (chief of staff) and Oracle (SEO) that handle outreach, ad creative, and page generation, spending about $100 a day on tokens and $1,000 on hardware. AI-first teams now outpace human-only teams, he says, citing Elon's claim that fully autonomous AI companies will move faster than any human organization.
Chapters
Cover OpenClaw and Claude Code, how to combine them together to not just do it for personal use, but for business use, for actually growing revenue. So I'm going to wait about a minute, 2 minutes or so before we get started here. And yeah, you want to hang till the end for this one because I'm going to be sharing a lot of examples on how this works. I'm going to be sharing a couple of tricks of the trade in terms of how to set this up correctly. And the cool thing here is you don't need to know how to code. So I'm a non-technical founder. I'm going to show you how you go about using this. Leave some comments too and share this wherever you're coming from, whether it's from LinkedIn, YouTube, X, or whatever it is. You're going to want to stay to the end here because I've been sharing this stuff in the last couple of weeks or so and it's been amazing. Especially because Opus 4.6 came out yesterday and the context window is like a million tokens now. There's these agent swarms that you can use, these agent teams. Yeah. So we're going to talk about that too. So I'm just curious right now, those of you, go ahead and type in the chat as we wait another minute or so for people to come in where you're coming from right now. And I just want to get that into chat and see where we go. And then we're going to get started. So this should be fast-paced. Again, I don't— I'm going to literally show you a bunch of examples. I'm going to talk through things and my excitement is going to carry us through this entire livestream, right? So I don't have any slides for you or anything like that. It's not prepared. I find it easier for myself to just talk and show things, show and tell. And hopefully you get a lot of value from this, but let me just see in chat where you all are coming from right now. So we have, let's see, public chat. We got Thousand Oaks. Okay. Steven Page, Thousand Oaks. I'm not that far away, or I'm in the Century City, Beverly Hills area. We got Mike coming in from São Paulo. We got Orlando. We got Amanda from Orlando. We got Madison. Wisconsin, uh, RevOps partner agency. Okay, that's cool. Uh, Saratoga, California, Vienna. Okay, cool. Oregon, um, Valencia. That's great. Morocco. We got people, international crowd. Argentina, Argentina, Argentina. Um, okay, so in Idaho, we got people everywhere right now. Some people are, are, you know, some people are colder than others right now. Uh, Denmark. Okay, great. So I'm just curious, um, as we wait for more people to come in to I'll start to ask more questions, but maybe I'll lead off with a couple things here. So I believe that during winter break, let's just put it that week that people had off. So we typically give the last week off and that's when everything took off for Cloud Code. Everyone was talking about it. So I saw everyone talking about it on X. I'm like, what is this Cloud Code thing? I don't know how to code. Should I pay attention? But then more and more people started talking about it and they started to show the use cases of the things that they can do. Now, the thing is, a lot of people were showing personal use cases or people just like that were coding, right? For me, I care about, as a marketer and a business person, I care more about driving more customers and I care about driving more revenue. And hopefully those of you coming in, that's what you care about too. And so I decided to play around with it and boom, that entire week just evaporated for me. I was literally immersed every single day. Yeah. And so much immersed, so much so that when we started the year, I mandated to the entire company, I said, for my ad agency, Single Grain, I said, hey guys, we're going to become proficient with this in the next couple months or so. And that is a requirement at this company. And so we're doing these, and I'm happy to talk about this too, about how we're going about the requirements here, but we have these hackathons going on. Okay? So these hackathons are required. They actually are, they take priority over regular work, over Company calls or customer calls, they take priority over sales calls. This is code red for us, right? And then we do these mandatory weekly office hours where people are coming in asking questions. We have a little help desk channel as well. And then we do a show and tell. And we're basically doing these hackathons every 6 weeks or so. And I was talking to the founder of Zapier. I was talking to the founder of Gamma. And these are both AI-first companies and they're doing these hackathons where they're taking the entire week, right? Yeah. And I just find that if you're in a position, a leadership position right now, if you aren't able to push this, if your company's not pushing this or you as a founder, you're not pushing it, I think you're behind the 8-ball right now. I don't think, I know you're behind the 8-ball right now. And so the reason I'm doing this right now is to try to get you ahead of this where you're not behind the 8-ball and where you're actually in fact ahead. And what's interesting to me is this morning I actually listened to a 3-hour, I listened to the first 2 hours actually, podcast. I was at the gym listening to it. listening to an Elon podcast that he did with one of the co-founders of Stripe and, you know, Dharkesh Patel. So very popular tech podcast. And he said something really interesting. And so I want you all to pay attention to this. And some of you might feel this is overblown. In fact, I want you to type into chat right now if you feel like you agree with this or you feel like this is overblown, okay? So type agree or disagree into chat, okay? So here's the sentiment. The Elon statement is, you know, all companies that are fully AI and autonomous in the future they're gonna move much faster than companies that have humans. Because in general, with humans, each human that you add is gonna add more complexity overall, right? Now, with that being said, I think for sales and certain customer-facing type roles, maybe not as much. But I just want you to type into chat right now. Do you agree or disagree with that statement that fully just very AI-first companies, AI almost autonomous, you have robotics and things like that where it can basically Think and do things much faster than humans without a lot of the emotions. Do you agree or disagree with that? Let me check chat right now. Wow, the vast majority of people here agree. Nick said disagree. Nick, type into chat why you disagree. I'm actually curious to know why you disagree. Rohan says, yeah, too much friction with humans. AI allows humans to do more at scale. That's what I'm going to show you in this livestream over here. Welcome, Kevin. So look, I'm so excited I forgot to introduce myself. So those that aren't familiar, quick introduction. My name's Eric Siu. My main business is an ad agency called Single Grain. We think of ourselves as a tech-enabled services company, marketing company. And then we have an SEO software, AEO software called ClickFlow, which helps people create production-grade content at scale. And then we have Carrot, which is our LinkedIn account-based marketing software, helps you personalize LinkedIn ads at scale in minutes, not weeks. And so That is the pitch. And then I do this podcast called Marketing School, and I have this on my YouTube channel right now called Leveling Up. So I dabble in a lot of these, just anything that helps business and marketing, I like to dabble in. So now talking about what I'm going to show you here. So I'm going to show you a couple examples first on how I'm going about using OpenClaw and Claude Code, and I'm going to describe these as well. So Claude Code, think of it as an agentic an agentic coder that helps you code things and build things. Myself, as a non-technical person, I am able to ship a lot of prototypes quickly. I can build almost whatever I want. I can connect it in with my Granola, my HubSpot, my Gong, and then come up with interesting insights and then build prototypes for the things that I want to build. You can use it for— we've used it to build a deal reviver tool that looks at all the lost deals that we've had and then just surfaces up the angle on what we should be messaging people with. Stalled deals as well. we'll find coaching opportunities for individuals on the sales side, also on the client services side. Looking at our Gong calls, the customer calls, what has opportunity for client expansion? What are prospects talking about on calls? And so this is all using different connectors. We'll look at that data and then it'll also suggest content based on how we're performing in Ahrefs, which is our SEO tool. So there's a lot of different things that you can connect. The thing that you wanna think about here is that you now have the ability with the Claude code to build anything that you want. Now, the cool thing is you can certainly build anything, but just not everything. So you just wanna keep that in mind, okay? So then Claude Code comes out, right? And it's been out for a while, but it really took a step change up in late December. And so, I mean, let's just say that was like 5 weeks ago, okay? 5, 6 weeks ago. All right. Then during that time too, I start seeing people talk about this Claude Bot thing, okay? And then everyone's talking about Claude Bot and it eventually gets renamed to Molt Bot and then it gets renamed to OpenCLAW. So now it's called OpenCLAW. All right. I'm like, okay, people are posting all these, their Mac Minis. I want to actually show you my little Mac Mini over here. I named him Alfred, but I have him right here. Alfred, like Alfred from Batman. Anyway, I'm like, okay, everyone's talking about this, but I don't want to learn this thing. Claude Code's already doing so much. You know what? I'm like, Eric, you're being lazy. Everyone's talking about this right now and how much they can do with it. People are talking about how it orders food for them. It orders pizza for them or it controls their oven. I don't think I want my AI to control my oven. Someone used it to make a bird feeder and it tracks the birds coming in and out. I'm like, this is really cool, but that's nice. That's really cute. But how about for business? You all hear like, how about for business? Everyone wants to learn the business use cases. I'm going to show you. OpenCLAW is— think of it as a true autonomous agent. People are throwing this agent word around like, okay, Yeah, I built this agent in Zapier. No shade on Zapier. We pay for Zapier. I built this agent using Lindy. No, you didn't. Oh, this N8N agent. No, you didn't. These are not agents, man. In my mind, an agent is something that is proactive. It has memory and it's self-improving. Almost like a human being, right? But we humans, myself included, we're flawed, right? We have a lot. We're very emotional. We make mistakes, right? Or sometimes we miss details or forget things, right? And so OpenClaw was the very first time that I saw something that I'm like, oh my God, yeah, this is it. So I'm going to show you a couple examples right here. And then I'm going to even show you how we are doing $50,000 of copywriting work from legendary copywriters in 10 minutes, how we're doing $45,000 of programmatic SEO work in about 20 minutes or so, and a bunch of other things. So let me just go into it right now. Yeah. Make sure you stay tuned. I'm going to start with the examples and I'm going to show you how you go about doing these things. Okay. So let me actually show you my screen over here. So let me share this. Okay. And if you guys can't see this clearly, let me know. I can make it a little bigger too. Just type it into chat. So I'm going to make this a little bigger. So, okay. So I was talking to my podcast co-host about a week and a half ago, 2 weeks ago. Okay. So he told me he has a 9-figure deal on the plate right now, 9 figures in revenue. And he's like, you know what I do every day? I actually reach out to someone and I email them. I look for the angle and I reach out to them. I'm like, this could easily be done using something like a Claude code or a Claude bot. And so what I created here using OpenClaw was the deal, the daily deal finder. So those of you that remember Woot, for example, W-O-O-T, back in the day, they used to give you daily deals every day. So the daily deal finder. Okay. So what OpenClaw does is it goes through our CRM sales transcripts and based on our goals, it figures out the right people to talk to and what the angle is. And I poop you not, no cussing on this one. I'm trying not to cuss. Nope. I poop you not. This alone booked me a meeting with a multi-trillion dollar company. And that led to a lunch and that led to other angles that led to, you know, us just talking more about working together. Right. Now, there's so many of these relationships that you have, these conversations that you have that get just lost in the ether. And so I've been building all these relationships for years and years now. And relationships are supposed to compound. You're supposed to constantly think about how you can add value for people, right? And so when you look at this, okay, let me give you an example. Here's an example output. Okay. So this is someone that we actually spoke to a year ago. Okay. And their role is in account-based marketing. And they were 3 years at this company. Okay. Their spend is $300 grand a month on account-based marketing, $3.6 million a year. So the discovery call, which this picked up, okay, this is random in Telegram. So I talked to my Alfred in Telegram. Discovery call was this, explicitly said she's figuring out the gap with ad personalizations. She's looking for new startups beyond Sixth Sense, asked about one of our competitors at this conference, and then uses these tools over here. She's getting 100% LinkedIn match rate, but wants deeper one-to-one landing page personalization. So guess what? We completely forgot about this person, right? So it's like if you're getting this daily deal finder, plus you can ask it to do more by the way, but look at this. It actually gives you the angle too. Hey name, great connecting last Friday. Been thinking about what you said regarding the ad personalization gap. You mentioned using competitor and wanting deeper Sixth Sense integration for landing pages. That's literally Carrot's sweet spot. Would you be open to a quick 15-minute demo? So look, I think this can be refined a little bit, but with a lot of these things, what you're looking for is You are looking for it to get you maybe 6 to like a 6, 7, 8. You just want it to give you some ideas and then you can work off the ideas. We just need a little push as human beings, right? Again, we're all flawed as humans, right? And then right here, here's the context. Okay. She said the team wasn't ready yet for a managed pilot. She values self-serve, but mentioned would like to do a paid pilot. How big would 1-to-1 landing page? So we do landing page personalization for Carrot. Target enterprise accounts for 1,000 plays. If you're getting this every single day and it's connected to all of your Gong calls, all of your Granola calls, sales transcripts, customer calls, all of your— and by the way, I'm recording this transcript right now using Granola and that's pretty powerful too. But it also looks at your CRM. It also understands all the goals with your business. You can combine all these things together, especially now that you have Opus 4.6 with the 1 million context window. And you have the swarm agents, the team agents that it can use, it's extremely powerful. Let me pause for a second here and explain what that means. Before I do that, I want you to drop in the chat right now maybe what use cases you would like to use something like this to solve, and then maybe I can jump around here, maybe I can show you those use cases or what else you'd like to see. Or yeah, just what else you'd like to see and what use cases you'd like to see. Just drop that in the chat as well as I go through this. Okay. What I want to call out here is, okay, when you— and I lost my train of thought before I went into that. I might just go straight into the next example. The way I have it set up right now with OpenClaw and Clawed Code is— oh, we're talking about the context window. Sorry. This is what happens when I don't have slides. The previous limit with Opus 4.5 was 200,000 token context window, which means it's limited in terms of what it can remember. If you can do a million, that's 5 times more. These things eat up a lot of tokens very quickly. Tokens are just like the text that you're outputting or it's outputting to you, tokens. What I would say is that's one thing. Context windows are a lot bigger now. Now, the other thing too is When you are able to spin up these agent teams, what that means is Claude Code and Open Claude, they now can spin up multiple instances of entire teams. Imagine if you want to spin up an SEO team right now, or if you wanted to spin up a social media team. You can do that now and it'll do it in parallel instead of you just working within one window. It is a little scary. Yes, you have to learn how to use a command line interface like terminal and all of that. Honestly, for myself, I'm going to show you how my my cursor setup looks. But it's honestly, if you put the work into it, our team was deathly scared of doing this stuff, but we said, look, man, this is a requirement. And after we got through the first hackathon about 2 weeks ago or so, we just asked everyone, hey, how do you feel about this now? 10 out of 10, 10 out of 10. This is amazing. This is amazing. And I feel like once you get people to see the light, once you get them to inject that drug a little bit, there's no turning back. Okay. And so the one promise I have for anybody that joins my company is that we're going to leave you off a lot better than when you first came in. And by the way, we're recruiting. So if you like this type of stuff and you're ahead with this type of stuff, reach out. You can hit me on Twitter. You can hit me on Instagram, whatever it is exactly. Cool. All right. So I'm going to show you the next example before I check the chat to see what else you guys would like to see here. But let's just go over here. Okay. So Ad Creative Scaler. Okay. So let me, before I show you the ad, okay, check this out. We actually had a prospect that came in for a single grain recently, and they're like, you know, we're spending a good amount of money on ads every month, and we would like to crank up because they're working with a lot of influencers, a lot of UGC stuff. They're like, you know what? We want to crank up our ads, our AI creatives to 5,000 per month. Okay? So we're pretty solid with it, but 5,000 is stretching us a little bit, right? And so, you know, just using my phone, I don't have my phone next to me, but just using my phone, using Telegram, I was just sitting on the couch. I was like, I wonder I wonder if this thing, this OpenClaw thing, can do the research so that ad creative strategy work, look at that company's ad library on Meta, look at their website, understand their brand, look at the competitors as well, do all the initial research. And I wanted it to come up with 3 initial ad concepts. It came up with pretty good ad concepts. I was like, you know what? I want to see if it can actually leverage the Gemini video models and the image models. And so bear in mind, this is just me watching TV and doing this like crazy. The fact that I can do this while I'm sitting on the couch. So I asked for an example using the latest Gemini models and it came up with this. And look, it's drinking this powder over here. And will I say this is amazing, the most amazing ad? No, it's not. But I just wanted to see the creative concept get going initially, and then I can build off of it. So if we had a creative on our team spend a couple hours on this, of course this is going to be way better, right? But the fact is I did this on the phone. So that is the crazy part. Okay. So I'm going to check to see what else you guys would like to see on this. And by the way, guys, guys, if you guys are getting value from this, type yes into the chat because this will give me good direction on where I should continue to go with these live streams. Because honestly, these feel good for me. So, okay. So I'm going to go up real quick. Let's take a look. Oh damn. The yeses are overriding what I can read in the chat. Okay. Francisco says he'd like to set up an insurance agency. Damn it. Riverside needs to fix this thing where if anybody talks, it just scrolls me down. Okay. Jacob says, keep sales reps in daily loop on data from HubSpot. Okay. That's very possible. Competitive analysis of competitors in search, input CRM and client calls. Jorge, it's not OpenClaw versus Claude Code. Let's address that first. It's OpenClaw and Claude Code. Okay, so I'm going to show you something real quick and then I'm going to go into some of the other examples here. People are asking about data stacks and things like that. By the way, you guys can keep dropping questions in as we go. That's totally okay. And we can go from there. And then I know Rohan says you want to see the SEO examples. I'm going to show you that as well. So all things will be shown. For local SEO, what could it do? I haven't used it for local SEO, but let me just go through a couple of these other examples. Let me give you another one, guys. Get this. Every single week I do an 80-minute podcast for Marketing School. In fact, I recorded yesterday with Neil and we have these 80-minute conversations that there's really good stuff. We're talking about the latest in marketing, the latest in business. We're talking about the experiments that we're running. There's so much good stuff in there, but Opus Clip, like any of these social media clipping tools, They're good, but they're not the best. Okay? And so I want to find something that can do a great job when it comes to clipping content from my long-form podcast, or even this episode right here where I'm like, this is the raw energy that's just coming from me right now. Oh my God, we're only 20 minutes into this thing, right? Time goes quickly when you're having fun. So I'm like, okay, I wonder what it can do here. Oh, okay. I just dumped a YouTube video into it. Top 5 clips, give me the highest virality potential here. Claude Code will replace your marketing team. Okay? I built a $30,000 go-to-market plan in one afternoon, 95% there. Doesn't even look like AI. And by the way, guys, if you stay to the end, I will answer questions around how you can start to set this stuff up because I think it's very important, especially if you're a founder of your company, you have to do it. You can't just say, oh yeah, I'm pushing my team on it. I'm having my leaders on it. That's exactly what Neil said to me in a podcast. I'm like, yeah, that ain't good enough. Right? I was like, you know what? I was like, I'll tell you what, Neil. I swore on my life and I also swore on our videographer's life. I went that far. Yeah. I was like, Neil, I even swear on your life, you're going to be in on this stuff. You have to be in on it because otherwise your company's going to be in trouble. Anyway, so I digress. These headlines, it finds the moment over here. It gives a hook as well. So I can either make net new content from it or I can just make a clipping tool if I want. I can use Claude Code to help me with that. So let me actually shift over. I'm going to show more examples in a second, But it's really important for me to also address, it's not Claude Code versus OpenCLAW, it's Claude Code and OpenCLAW. They both have different use cases. I was actually workshopping this with Claude Code the other day. I was like, OpenCLAW is good, but I'm in this Telegram interface and there's so much I can do. There's so much I can do, but there's also so much I can't do because Claude Code does a good job with planning. You guys will like this. Can it create websites? Absolutely, it can create websites. This stuff made an ad. It made like an Hermes ad and it looked amazing. Okay, really sky's the limit. Anything you can do on a computer, it can do and probably better than you. Not probably, definitely better than you. And in fact, I'll swear on my life on that one too. Definitely better than you. Okay. All right. So let's move over here. I'm swearing on my life too easily here. So let me, let me come over to, I'm going to do a new share over here. I'm going to share Cursor. So let's share Cursor. Okay, let me open it up and make sure I can share it. Give me one second. Okay, let's try this over here. All right. Okay, there we go. Cursor's right here. Okay, and I know I will have to make this bigger as well. So let me make this part bigger here. Oh, it's hard to read that. Okay, this is about as big as I can make it. So focus on where my cursor is, no pun intended, is right here. So I was like, you know, yesterday, in fact, I came back to this morning. I was like, what can this actually replace? I was like, okay, if we did Claude Code alone, if we're talking about, you know, an SEO director role plus 2 writers or like a growth marketer, for example, and then it kind of gives like an average like US dollars cost per month or an SDR, for example. Okay, so if we're talking about an SEO director plus 2 writers, Claude Code alone can do 90%, writes content, analyzes, Google Search Console, Google Analytics, identifies decay, can't publish to WordPress. Open Claw alone can do 30%, can browse GSC and publish to WordPress, but shallow analysis, poor content quality. That's debatable. I mean, eventually I think this is just going to become amazing. If you combine both, you have Claude Code writes production-grade content, and then you have Open Claw publishes to WordPress, updates internal links, submits to Google Search Console, fully hands-free pipeline. The whole idea here is you can see as it goes down, it's like you can't, Each of them alone isn't quite enough. When you do SDR and outbound, Claude Code can do the research for you. It can enrich via Clay. It can draft emails. It can send emails via Google's Gmail's MCP. It can't browse LinkedIn. OpenCLAW can browse LinkedIn. It can send connection requests, email, but personalization is shallow. Now, I don't know about that again, because if you're using Opus 4.6, it can do a good job. The thing is, an OpenCLAW loves to eat up tokens very quickly. You have to be— very vigilant about how you use these tokens. And let's say if you don't have a lot of cash to spend on this stuff right now, I would recommend using KIMI. That's K-I-M-I 2.5. And that's an open source model that you can use. You can run it locally. Okay? So my point here with showing you this is like, look, social media manager. Okay. Cloud Code can do 60%. OpenCLAW can do 70%. But when you combine these powers together, it could do 95% of the work. And so I just had to do an analysis. I'm like, okay, well, total roles covered. Okay. 6 out of 9 partially. OpenCLAW alone, 5 out of 9 partially. Combined, 9 out of 9. Okay. Monthly headcount replaced here, we're talking $40,000 to $74,000 per month, by the way. OpenCLAW can replace $28,000 to $74,000 per month and then $67,000 to $74,000. Then it does these annual savings. Then for me, when I run this thing, by the way, I have a clawd.md file and your OpenCLAW and your Claude code will pull from the clawd.md. That's your goals, your spec in terms of how you want it to work. There's all these little nuances that you have to learn. And that's why it's so important for you to go out there and learn this stuff and do this stuff. And again, if you stay till the end, I'll tell you my preferred way of going about learning this stuff because I picked this up and all I needed was a little time. Okay? So I get it that we have other commitments. There's family, there's kids, there's a spouse. There's a lot of other life commitments too, right? And I understand that. But even if you can dedicate one day a week to this, and that's asking a lot, you're going to be— I think you're going to be good, right? You're going to be ahead of 99% of people. So it calculates the ROI for me. When you combine these things, almost none. So what I'm doing right now is I am actually thinking about how I can combine these into an entire team of an AI agent squad. So let me— I'll stop my share for a moment here. So the way to think about this is OpenClaw, like Alfred over here, Can be my chief of staff. Alfred has the most context around me and has the most context around my business and knows what my goals are. My goals are getting updated every single week. It's running these cron jobs, these tasks pretty frequently to constantly refresh the goals, seeing how we're doing, conversations we're having, seeing any red flags and just flagging it to me. The cool thing is eventually I'll be able to share it with the team once I feel good about all that. Then you want to start to build. My heroes are in the DC Universe. I have Oracle, which is the SEO bot. My objective there is to see what amount of SEO work it can replace on my team so my team can go do more higher leverage work. Same thing with my internal social. The bot's called Flash. I have these different OpenClaw agents and then I'm eventually going to have them just interacting with each other in a mission control. Then they're just proactively solving things. They're bringing solving ideas to me and then they're Saying, okay, what do you approve? What do you disapprove? And it'll learn from that as well. And then it can execute on these things for me. And then it's going to compound over time. That's how you want to think about this. And so more and more, I'm telling my team now, today I'm writing a newsletter to my team. I'm doing a video as well, just saying I do these weekly newsletters. And my main point today is the laws of physics around business has changed. No longer can you say, Oh yeah, it's going to take me a week to do that or 2 weeks to do it. Doesn't work anymore because what I'm about to show you, the example that the $50 grand of SEO work in, $45 grand of SEO work done in 20 minutes. Someone told me that it was going to take a week or so. And I said, I'm just going to do it right now. And I not only did that, I ended up doing 4 times the amount of work. And took me 20 minutes initially. And then I just probably like another 20 minutes to finish the rest. And then the work was basically, let's just call it 90% there. I believe you still needed a human in the loop for this stuff. And then we were— Carrot for our LinkedIn account-based marketing ads product, we're like, I actually was working with Claude Code on this. I was like, Carrot's continuing to drive revenue. We don't really put any ad dollars towards it. Shouldn't we do that? And then it would create an entire strategy, but that strategy was left sitting there for 2 and a half weeks or so. And then once we confronted the problem, he's like, hey, what's taking so long here? The individual that was responsible for it was like, Oh yeah, you know, creative is held up right now. And you know, part— I really want to cuss here. I'm not going to cuss, but I'm like, really? Like, forget that. Literally in 20 minutes, I spun up the ads and I just went into the ad account myself and I just turned the ads on. Okay? So my point of saying this is that, you know, not that humans are useless, it's more so like the way we work now is different. Okay? It's more so the laws of business have changed. And you as a leader cannot accept it when people are constantly telling, especially when we're trying to bring people into this future, we're saying AI fluency, Claude code, we're going to bring you there and we're constantly sharing this stuff. And then you tell me something's going to take 2 weeks, especially when there's open utilization on that team. So we measure utilization as an agency, right? So we know there's utilization there. And when a bottleneck is presented, you now expect your team to be able to solve these bottlenecks. You expect them to be able to run through walls. Because if not, then what are you doing? Right? And so you can see some of my excitement coming out, but also I have a lot of frustration coming out too because in my mind, I'm like, how could you not want this? Right? Because the world, for me now, I played a lot of games growing up. I just feel like I'm playing a video game every day. Right? And so if you want to play the game, you don't want to play this way, that's okay. Right? Because we can help you find something else. And so That is our mandate now. And one more thing, I'll digress in this way. If you are a leader right now too, you really got to think hard when it comes to hiring. And one of the first things I ask that I look for is that AI fluency. Hey, tell me the coolest thing that you've done with AI so far. How has it affected a KPI? And I'm looking for exceptional ability. That's what I'm looking for. I'm looking to see if this person can hang with me and have an exceptional conversation for 20 to 30 minutes or so. And if they can't, then We just can't hire, right? We just can't hire you. Just how it is, right? Because the baseline expectation now is at least Claude code proficiency. Now, by the way, we're not teaching people OpenCLAW right now because there's too many security vulnerabilities that you have to pay attention to. Maybe when we talk about setup in a little bit, you want to pay attention to that, right? Because I have it set up on a Mac Mini over here, not because I wanted to spend the money, but because I wanted to have more security control. Even that's not enough to just protect it completely. Yeah. So Claude Code is, we're okay bringing people into that world, but Open Claude, that's kind of, you decide if you want to bring people on that journey because there's a lot more. Excuse me. Drink water too fast. Okay, so we're back. All right. I'm just going to look at some of these. These comments over here. So Jorge says, how do you keep your C-suite from getting distracted with all these possibilities? The key thing that when you're talking to your team about this is that whatever they build, excuse me, whatever they build, when it comes time to show and tell, I appear for show and tell. And so everything that I ask is like, hey, that's cool. How's this going to tie in with ROI? What is the return on investment here? How's this going to help you hit your goals? How's this going to help the company hit its hit its goals? Also, how much time is this thing going to save? We're not just building a bunch of gadgets. We're not just building the build because to your point, you can get easily distracted by this. That's what you want to pay attention to. A couple of things I covered. Again, it's Claude Code plus OpenClaw for yourselves listening to this. If you want to bring your team for the ride, probably just Claude Code. Then right now, Serge, the average monthly cost for these tokens all in, I will say I'm spending about $100 a month. Sorry, no, no, no. It's $100 a day right now. But if you use Kimmy, it's basically going to drop to nothing because I am using the latest models like Opus 4.6 that just came out yesterday. Later today, I'm going to be spending a lot of time on Opus 4.6 just to see how these agent swarms work out and how that long context window helps too. The other thing I do ask it, and I'm going to show you guys some copy examples too. I'll give you guys a prompt over here that you can use and I'll repeat it too. The prompt that I use, and actually I'll drop it in the chat as well. No, Kimmy, Julian, Kimmy is actually, that is just another LLM. It's like an Opus 4.0. It's a model, right? It's an open source model. That's what it is. Let me just do this. Okay. Bill asks, do you have any concerns with Kimmy being Chinese owned? I would say, and by the way, I'm Chinese, and so I don't have any concerns with it. Um, myself, but I would put it on a local device and run Kimi on a local device. So you might buy another Mac Mini and have it run there. I've actually seen, um, Twitter posts of people running multiple Mac Minis. Okay. Um, and yeah, um, Goku, yes, Opus 4.6 is 50 free credits. Go ahead and use that. Um, and so let me give, let me give you this prompt over here. So I, I just dropped it into chat. Um, I don't know if it actually showed there. But the whole idea with this prompt is before you send me any output. So for example, let's say I'm trying to write copy. I'll say, I want you to pull from the top 10 advertorial legends in the world. And I want you to rewrite this and score each one on a score scale of 0 to 100. Only when it's 95 out of 100, I want you to return it to me. That way the copy it gives you is just much better. Okay. So I'm going to show you some of the output that it produced. And go from there. But somebody asked about the SEO example. So actually, no, I'm not going to jump to that yet. So I'm going to share my screen. Check this out. So this is ClickFlow over here. Okay. So originally our copy wasn't that great. All I did, I actually went into Claude Code and I just literally put in that prompt and I said, just, I wanted this to be better. Okay. So for ClickFlow, it says your AI content doesn't drive revenue. ClickFlow changes that. Sounds like your best writer at 10 to 15x the volume. Okay. So originally we weren't just like the comments we're getting, early comments right now we're getting with ClickFlow is that the content quality is a lot better than these other content generators. Okay. And then also the speed as well. And so we're like, okay, not enough people talk about that. And our original headline didn't cover that. And this, it gave me a better version of it, but we still needed a human to look to refine this, right? Like this piece wasn't here before. This piece over here, it kind of reads like a sales letter. So we have these little blocks of these portions over here. Like the volume play is over. Every AI writer and SEO platform sold you the same story. Publish more, rank more. So everyone did. Now the internet's drowning in content that all sounds the same. Google's quality updates are demoting it. AI overview, skip it. ChatGPT and Gemini don't care. Your readers bounce in seconds. 50 mediocre articles don't outrank 10 exceptional ones. Not anymore. So What am I saying here? I'm saying that we're getting, if this is pulled from thinking from Ogilvy, for example, Eugene Schwartz, who else is in there? Eugene Schwartz, I think the guy that wrote Breakthrough, not Breakthrough Advertising. God, I forgot his name. Anyway, this table over here, we added in as well. Okay? So we didn't have this table before, this reality check piece over here, feature comparison, and it starts to get a little busy. And then this piece over here, But this page was originally converting at about, I don't know, I think the homepage was converting to the next step at about 15% or so. I think it's now up to about 23 or 24%. Okay. So if we go 23% on 15, that's a 53% conversion rate increase on that. And then we're also having it go through our entire funnel too. So you can use Open Claw for that. You can use Claude Code for it, whatever you want to do. And that's what it is. And there will be a recording for this as well. And KG asks, how do you learn all these when doing all these? How do you learn things when doing all these things? Do you process or dedicated time or do you manage every— how do you manage to be everywhere? So I learn partly by teaching right here and I'm experimenting a lot. And so in whatever free time I have, I'm learning this stuff because I only have 3 priorities right now. One is tech-enabled services, which you're looking at this page right now. 2 is talent, which is recruiting and talent management. And then the 3rd thing is just really driving AI fluency. And it's really my job to make sure that we drive the company forward on this. Okay. All right. Glad that that stuff's working for you guys now. Okay. So one of you asked for the— excuse me. So the example for the SEO content or the programmatic SEO content. So Manus, I'll make this a little bigger. So Manus is really cool because it's now owned by Meta and it's a general agent. So I'm not even talking about Claude Code in this case. I'm just like, guys, if you didn't even want to learn Claude Code or OpenCLAW, just log into Manus. I basically said my prompt was so simple. I was like, hey, we're looking to create these programmatic SEO pages. Here are 2 templates we have on our website. I want you to expand on this keyword over here and come up with 150 variations. And here's what it did. So for carrot, It came up with 156 keywords over here and it looked for these listicle pages and we have these service pages. And it came up with, here's what we got over here. Here's the service pages. Here's these tools pages, whatever. I'm not going to bore you with all the details. And then I was like, okay, we'll come up with the content. So I wouldn't say this content's the best because you can tell it's pretty AI in the very beginning, especially the first sentence or so, but it gives me a good first pass. And then we can run this through ClickFlow to make sure it's better. These listicle pages, yes, we have good domain authority. Google's talked about, yeah, they're cracking down on it, but these work really well for us. And so we wanted to generate more of these pages because we're not covering a lot of these ABM keywords. And so we wanted to generate more. And so again, when I asked the question originally, how long is this going to take to do? It was like, oh, well, it's going to take 3 days to do keyword research and then it's going to take another week after that. And I'm like, okay, that's pretty fast. And our team's pretty fast at that, but I'm like, this needs to be faster. And the cool thing is the SEO team took it on. They looked at this workflow after and they embedded it into their own workflows. And in fact, some of them already had this, right? So that's what I would say. And by the way, you guys will notice I talk very quickly. I definitely recommend watching the replay later if you want, and you can even take the transcript if you want. But that's what I would say, right? This is something that you probably want to rewatch. And the cool thing, by the way, I'll say about Manus is that Manus has added skills now. And so for this Riverside webinar thumbnail, guess what? I don't even have to go to designer anymore. I just pull up this skill over here and I just type /riverside-webinar-thumbnail, boom, generates a thumbnail for this. If I wanted to generate an ABM outreach generator, I was just messing around with this. Even Claude Cowork allows you to do this. It'll figure out the people to reach out to for Carrot. It'll look at what the context was when we talked to that person. There's all these things that we can do here. Let me see if I can dig up another example here for you. I'm looking at identifying and follow up on deals. I don't want to give— I don't want to share sensitive details here. Yeah. Okay. Maybe we'll go here. So stalled carrot deals over here. Maybe I'll just show you one. Maybe it's okay. So while we wait for Manus to pull this up, Adrian, the prompt that I shared earlier is probably the best one. I don't know if I'm able to share it with you all. Public chat here. I don't know if that works. Can you guys see that at all? Okay, there we go. Yeah. So that is the prompt. Save that prompt over there. Before you return a version to me, have 10 of the world's greatest advertorial experts on subjects like design, copywriting, psychology, and CRO review the page, provide detailed feedback, and rank it on a scale of 0 to 100. Each expert should include specific areas for improvement that are effective. of the ranking, reflective of the ranking. If the average ranking isn't over 90, I actually made it 95 now, keep improving it. And so my point of showing this, by the way, thank you for all the thumbs up and emojis. That's cool. I didn't even know this could do that. Shout out to Riverside, by the way, for the webinar feature. So, okay. You can actually change this to whatever it is that you're working on. It could be on SEO, it could be on sales, it could be on management, whatever strategy you're building, whatever output you're about to get, You want to use this, and it makes the output way better. And and that's what actually helped me with the click flow output that you saw. Okay, in fact, I'm gonna show you the new single grain copy. We actually used you know AI driven copy, like just the rotations. The speed is a lot faster. Like I get it, and I'm not just I'm not gonna just publish it there. I'm going to work with it. I'm gonna you know whatever stakeholders I'm gonna loop them into that, right? That's how you want to think about this. Okay, so identifying and following up on single grain carrot leads over here. Okay, so let's just do this one over here. Okay, so this. Hasura. So they came in as a lead for Carrot. Okay. Look. Hi Maurice. Checking in from Carrot. I hope you're having a great week. It's been about 8 months since we last connected. I remember we discussed your ABM strategy. I'm curious to hear if anything has changed. Guess what, guys? That outreach is— this is fine, guys. This works because a lot of times, 99% of people aren't ready to buy in market. Maybe they forgot. Maybe priorities shifted. Maybe they went with a competitor. The competitor sucked, whatever it is exactly. But to be able to do this, by the way, again, you can do this in Claude Cowork. They have a nice general agent you can use. Eventually what you want to do here is you want, because you have to keep going to this every time, the reason why you want to open Claude is because open Claude is autonomous. I'm going to show you my open Claude in a second, but it's autonomous and you don't have to keep going to it and begging it to do the work for you. I think that's where all this stuff is going and open Claude just happens to be ahead right now. Will Meta get there? Maybe Meta will get there. I mean, with Meta Madness plus Meta, Good combination. So just the sky's the limit with this type of stuff. Okay. Now let me flip over. I'm going to show you some more copy. Let me actually move over to the Single Grain website. I'm going to show you my setup with Telegram, and then I'm going to move over to what you guys can do to start to learn this stuff and some security considerations and other setup considerations you should set up. So yeah. Let me move over here. Okay, cool. Oh, look at that. Look at all these popups. Okay, guys, by the way, this popup— Francisco, that's a funny question. Francisco asked, can you set up OpenClaw to use Claude Code? You probably— I don't know if you would do that. I use Claude Code to help me plan first and work through and test something before I roll it out to OpenClaw, but maybe you can ask Claude Code. Or OpenClaw or even Claude itself if your way might make sense. And by the way, in a lot of these situations, when you start to learn Claude code and OpenClaw, just ask Claude how you can— just ask it to teach you. Ask it to teach you. It's infinitely intelligent. It's definitely smarter than us. So there's that. Anyway, so there you go, Jacob. Exactly. So this popup that we had originally, our popup a couple days ago was just focused on Carrot, which is LinkedIn account-based marketing software. So 2 ways to fill your pipeline. This has now changed. Pick yours. And now, so it says ClickFlow for SEO and content teams, our AI content isn't converting. And then for Karat ABM and demand gen teams, our LinkedIn ads lack ABM personalization. Join the waitlist, start free trial. This is what it looks like. What we had before was just like a generic Karat ad, but now we get like, a lot of people aren't gonna be looking for Karat 'cause it's really tailored towards B2B. And in ClickFlow you have this. And so the speed in which we got this up, usually this would've taken like what, like a week, 2 weeks to get this up. We got this up in— we got the copy, we got the design, everything, probably 2 days to get this up max. And it's things, honestly, I even think 2 days is a little slow. Okay. So we got the copy from that, right? And using the same prompt again, I actually changed up the copy on the Single Grain site. Again, this would've taken a while to get done, but now this sticky bar up here will rotate, I think. Yeah. Look at that. It rotates. By the way, this is my first time looking at it rotate. Okay. I'm like, that's pretty cool, right? The last agency you'll ever hire. Okay. Marketing that produces revenue, not just reports. Most agencies focus, they report on impressions. We report on pipeline. Single Grain builds revenue marketing systems that compound, not campaigns that expire. And all this stuff over here, your last, you know, it sounds familiar, you're in the right place. Your last agency sent you monthly reports that you couldn't tie to a single closed deal. You're spending $25,000 plus a month on marketing and your CEO asks, what are we getting for this? Your competitors are showing up in AI search results, features. It speaks in the language of AI. By the way, I told it to pull also all the recent content that I've put out, how I talk about things, how our goals are set up. And it took that into account when it rewrote this entire homepage over here. Look at this reality check over here. Reality check, this isn't a problem you can solve next quarter. The companies acting now are the ones that will own their own categories, right? Or own the categories. So look, Am I saying AI everything and YOLO AI? No, I'm not saying that. I'm saying become AI enhanced. Because when you're AI enhanced, you're 10 to 100 times faster. I actually asked Claude this question. I was like, with Claude Code, how much faster am I now? It's like, you're 25 to 30 times faster. Claude was like, actually, arguably, you're infinitely faster because you couldn't do any of these things before. I'm like, okay, that's cool. Then I started playing with Open Claude. Then I went back to Claude Code. I was like, hey, with Open Claude, how much faster am I? It's like, You just added a multiple of 5 to 10x on top of that. So what? I'm 300 times faster now? 250 times faster? Again, even if I'm 30 to 35 times faster, how can you expect someone to keep up with you when you're that much faster? They can't. And that's why you have to touch this stuff. Look at this language, revenue acquisition, revenue visibility, revenue creative. Cool. Just cool. I'm so proud that we got this. Not only did we do this, we got this out quickly. And then our designers and developers that worked on this are like, this is the best copy we've ever had on the site. And again, is there room for improvement? Absolutely. That's what I would say. Now I'm going to show you my Telegram for a second and I'm going to move over into the setup considerations. I did a post yesterday on X that actually did pretty well. So let me do this. So Rohan says he's been using Lovable. Lovable's cool. And, um, yeah, you guys are talking about— oh yeah, burning €1,000 in 30 hours is, is nuts. Uh, Jacob, you want to make sure you protect your usage there. Okay, cool. So let me move over here. Let me show my Telegram real quick. So let's go to my Telegram. All right, y'all can see this, right? It looks like it's okay to see. Okay, so I actually asked it because I— there's some sensitive stuff that I don't want it to show, but I'm like, hey, I'm about to do a live webinar talking about my OpenClaw What are some cool things we can show if I were to show this chat? This is my multi-agent architecture diagram over here. This is what you want to think about. Again, you have Alfred as the orchestrator. By the way, look at this picture of Alfred. Can I dig it up? Okay. Well, I'll dig it up later, but it's like the Alfred from Batman Begins. Anyway, you have Oracle, Flash, and I have Arrow coming for sales. Then they have shared learnings where they learn from each other and they interact with each other. The key thing here is you don't try to build 10 of these at once. You try to build Alfred, get Alfred working first, and then you start to build these agents on top slowly. Then decision interface pattern. I don't want these things to just give me insights. I want it to give me— allow me to give me ideas, but then I approve and reject and I give a reason as well wherever it makes sense. Then it can learn so it doesn't give me the same crappy ideas over time. I also will run an analysis on 30 of the key team members that I have, right? And cross-reference Gong calls with core values and then flag anything that's red, yellow, green. And so what it does is it will analyze, kind of look at a 60-day moving average and say, okay, hey, this person can actually perform better over here. This person could be doing better over here because I look at culture as a 60-day moving average. And so at least I'm able to see this 'cause I'm not able to be everywhere at once. When you have all these calls kind of combined, you can start to analyze people's languages. You can start to see their patterns over time. You can start to see maybe something might be off and maybe I should talk to that person or maybe their manager should. I can just forward this over. And this looked at a small set of discovery calls here. So content brainstorm from sales calls. Analyzed 57 discovery calls. I think this is over a certain period of time. Extracted pain points, budget signals, mapped to content recommendations. Show AI listened to my sales calls and told me what content. Great. Yeah, I saw that above and it actually gave me, it's like, hey, we're struggling to generate more pipeline right now. Nothing is working with AEO. That gives me ideas because that's the type of content that can convert people and not just get a lot of views. So just keep that in mind. I did a hiring process upgrade. So I was walking back from the gym this morning and I was just dictating to it using Whisperflow. I was like, hey, here's my org chart and I want to tell you who I felt like I had exceptional conversations with, going back to the Elon example I gave earlier. So I want people that have exceptional ability. But that should be demonstrated by having an exceptional conversation with that person. Okay? And this helped me update the hiring docs in real time because this has all my hiring docs connected to it. So again, I have all my different tools connected to it. And I'm going to talk about this setup in a moment, but all my different tools that are connected to it, it knows my goals. It refreshes my goals over time. And it's being trained on my content, my new content every single week. And so just consider that. And then also, by the way, there's a There's a WordPress API integration I have with it too. And so I use OpenClaw to do SEO analyses and then it scans, it comes up with a strategy and then it uses ClickFlow to bulk publish to my WordPress because it has the API key. So there's just a couple of things that you can do. So you can actually, the key thing here is it's going to autonomously execute for you and you're looking to build that compound value from a team perspective. So that's how you might look at it. And I actually had my SEO oracle, it's been broken, right? And so you can see like this thing, it gets kind of stupid, right? And so like I had to kind of like, you know, kept kicking it, kicking it, kicking it, and it wasn't quite working as intended. So the other thing I'll call out too is just if I look at the videos it provided, this is like another one of the creatives. And I won't say this is amazing or anything, but look, it even has like a sound effect. So, you know, sky's the limit. That's all I want to call out. So now let's go into the things to consider when you're starting to set this up. And first, let's talk about how you can go about learning this stuff too. Okay. Yes, my SEO Oracle is running on the Mac mini as well. Julian asking about security. So let's talk about that. Let's talk about how to learn this stuff first. Okay. So I would recommend if you're trying to pick up Claude Code first, Go to X. Okay, that's where all the action is happening. Go to X. Go to Grok. Grok is the LLM that's trained off of X and just ask, hey, who are the best people to follow in terms of Claude code right now, in terms of learning Claude code? And also who are the best people to follow in terms of OpenCLAW? Now it's going to list all the people, the engagement, and then you just follow these people. You look at all the videos that they have out there. There's this guy, Alex Finn. He's got good— I think he's more of a marketer, but he's got a lot of good energy to share. My YouTube channel, Leveling Up with Eric Siu. Riley Brown is another great one as well. And you just want, you're just looking because this stuff is changing so quickly every single day. You can't, in my mind, I think if you're a leader listening to this, and I think you're all leaders, you can't afford not to be checking in on the action on X, right? Just because it's moving at blazing speed right now. That's what I would say first. Now, Julian, in terms of security, the reason why I got a Mac Mini, I don't have it on a VPS, is because there's more, you have more attack You have more attack vectors, I would say, available when you're just putting on a virtual private server. There's been a lot of talk about security. I just run it locally on my Mac mini over here and it's connected to my internet. That's how I do it. Let me move over to this post over here. I just threw this post up yesterday. One thing too, by the way, from a marketing standpoint, check this out. Look, I just put this up yesterday. I published it at 4:00 PM. And look at this, 68,000 views on this one. Not bad. So if you want more reach right now, you write articles on X, which is kind of funny. Long form is back. So I said this, ordering food with your OpenClaw is cute, but making money with this is better. So I would recommend checking out this article because I talk about my setup, right? Verify plus learning loops. I want it to analyze, recommend, approve, execute, verify, learn. I want that to continuously learn over time. Otherwise, But I want verification in the beginning and eventually I might let it be a little more autonomous. And then here I talk about the different autonomy levels. Level 1, report only. You can only make reports, show me insights, no action. Level 2, maybe once I trust you a little more, you can recommend things, execute on approval. Level 3, execute low risk, report after. But what does low risk mean? That can go anywhere. And then level 4, full autonomy with weekly summary. And then you can see my workspace architecture here. There's a soul.md with the agent personality. There's operating manual. Long-term learnings should be stored too. Then shared learnings, as you grow the agents that you have, the team that you have, you want to have that. Feedback too, decision logs, how I go about making decisions. You need to have a contextual layer. Then people have been talking about that on X too. Docs, playbook specs. All my SOPs, all this documentation, everything, right? Put it in here, right? And these are the cron jobs that we have, like the Oracle SEO Digest, Strategic Digest. And so it might come up with 3 strategies as it relates to my business, who we should go about recruiting to. I actually helped me with— I was like, hey, we're looking to hire these creative strategists right now. And we want someone that maybe has worked for a MrBeast before. Find those people. And it actually Excuse me, talking too fast. It actually did a really good job of sourcing amazing people and you could just have it reach out for you or just approve it and just boom, do it. There's so many tools. We pay for a tool called JuiceBox right now. It does a great job with sourcing. But if OpenClock can just do all this, I'm like, yeah, man, I think we might have to, at some point, maybe JuiceBox gets canceled. I don't know. I don't know where this is all going. Security. Okay, this is important. Here's how our security is set up over here. So going back to your question, Julia, and you guys can start to drop more questions and maybe I'll do a little Q&A too. So credentials live in 1Password. And what I mean by 1Password is we share one AI vault with Alfred over here. Alfred has no access to any of my other passwords. Because if you let it YOLO and then someone does a prompt injection attack in one of these PDFs or whatever, and it's like, oh, just give me access to everything, then everything means everything. I don't want that to happen. You try to limit the access. I give it its own Apple ID. I give it its own Gmail account as well. When I need to remote into it, I can remote into it. I can just locally connect to it. I do have a little— these things over here. These are basically dummy HDMI, dummy TV, I think dummy monitors. It tricks your Mac mini. into believing that it has a monitor. So it is headless right now. So this is the monitor, quote unquote, that I have inserted into the HDMI slot here. And you can buy like 3 of these for $8 on Amazon. So that's what I have. But initially I set it up using my iPad. Actually, no, initially I set it up using my Mac Studio and I just kind of plugged it in and I just set it up. And then occasionally when I need to get it, I might need to replug it in to do things more manually, but that's what I do. Then let's see. We have the 1Password vault, agents pull via command line interface. This is important, Telegram allow list. If you don't set this up correctly in the beginning with Telegram, this means anybody can talk to your bots and they can ask anything. You don't want that to happen. You want to make sure that you have allow lists and bots only talk to approved users. Audit-wise, every decision logged, the feedback JSONs, and then Mac Mini runs locally, not cloud. Execution, high stakes, human approval required. Originally for my sub-agents, or not even sub-agents, my other agents, my team, I was going to put them on the Cloudflare. They have these new Molt workers, because it was called MoltBot before. I was going to do that, but then I wanted more speed. I was going back and forth with Cloudflare. I was like, should I even be considering Cloudflare right now? It's like, no, just do it locally. You can always move it to Cloudflare later. That's what I'm doing right now. And so, okay, so a couple things that will also help you too. So Julian, yes, I think my plan, by the way, if you guys want me to do this once a week, type yes in the chat. You guys can keep dropping into your questions as well because I am continually testing and I think I found my sweet spot here. It's AI content for business and marketing, right? You want to attack the trend. And so, okay, so Sonnet for daily, Opus for strategic. Okay. Right now, routine crons, SEO Digest, Deal Finder, they run on Sonnet. It's about 15 cents per run, which is actually pretty expensive. I might consider buying a couple more Mac Minis and just running Kimi 2.5 locally to save me even more money. Opus will fire only on high-stakes decisions. 10x the cost difference, same outcome for 80% tasks. Now, this actually broke my setup once. I would say that maybe for the heartbeats, which are the recurring check-ins that it's doing every 30 minutes, every hour or so, maybe you're using Sonnet for that. That way you're going to save a lot more money. Because I will say I'm spending like $100 a day right now, which I'm going to have to wean off at some point. And then each scheduled job runs its own section, no context bleed. So my 7:00 AM SEO Digest doesn't carry the 8:00 AM Deal Finders history. So you want a clean slate. And then, oh, obviously we're still trying to teach my team this too. You want to make sure that you're compacting Especially when you're using OpenCLAW because it's a pain in the butt when you have context overload, like the thing just broke. Yes, OpenCLAW is open source, autonomous agent and all that. You just got to make sure that you're compacting and maybe clearing where it makes sense because you don't want to give it context overload where it just didn't know what to do anymore. I do ask it to also spawn subagents too, to work on things in parallel. But the thing now is that if you have 4.6, Later, I'm going to be installing Opus 4.6 to my OpenCLAW. If I can just spin up those subagents now, then I'm good, right? Or not the subagents, but the agent swarms or team swarms. Let's see. We talked about this earlier. Again, Claude Code for testing and planning. Claude Code's planning feature is really good and it's really detailed because when you do it on OpenCLAW, it's just like, okay, I'm just going to go build it. You don't want to do that. And then I do dictate quite a bit. Okay? So on my phone, I use Whisper Flow and the dictation is amazing. And so I can talk through my idea there. And by the way, I do also use Claude Code on my phone. Inside of the Claude app, there's a code section and sometimes I'll work through it on more strategic things. I can't do more planning stuff as much in there because I want to be on a desktop dragging in pictures and things like that. Let's see what else. Yeah, you want to have a knowledge base as well, preferences. And then we talked about the physical part. So this is the little thing that I have plugged in. So this is what Alfred looks like. He's sitting right there right now. That's what I would say. I don't know if you can keep hallucinations in check. I don't know if you can ever— humans, we hallucinate all the time, meaning that we're very flawed. Are we a perfect source of information? We're not. Maybe the only perfect sources of verifiable information is maybe math and physics. Everything else is our interpretations. That's what I would say. Okay. So Jonas says, the keyword is WAF, web application firewall, to control how much you communicate. Just context, what you already said about Eric with Cloudflare. Okay, cool. So I am maybe going to show you, let's see. I'm going to show you maybe one or two more things. You guys can drop questions in the chat if you want, but I'm trying to see what else I want to show you here. We talked about that. We talked about that. Let me just tell you where I think this is all going. So I think the future of this, and then I'm going to answer questions. I think the future of this is going to be all of you doing this right now. You're going to be running these teams and these teams are going to be— I look at the way we're going to be working is very much the Palantir model. The Palantir model is you have these 4 deployed engineers. So we might have 4 deployed marketers where we're trained on all the agents that I've refined. And once I feel like my marketing team is perfect with my team of agents, then I'm going to talk to prospects or clients about rolling this out. And then the way I look at it is you're going to have a 4 deployed marketer just iterating on the strategy and then customizing the team of agents for that customer. And then we're just going to have very smart people working on these things at scale. And these things will be replacing a lot of the manual work that humans have to do today. And so I think a lot of the work that we do is going to be reimagined. And I actually think this ties in with the SaaS-pocalypse that happened in the last week and a half or so, where you see like, oh my God, HubSpot is down 72% in the last year or so. All these stocks are down like 40%, 50% here, whatever it is exactly, right? And the reason that's happening is because people are like, oh my God, the seat-based model is going away. And that might be true when it comes to buying software like HubSpot, you have to buy a seat and things like that. But you know what's going to replace that? That agents are going to replace that. You're going to be buying— the seats are going to become agents. You're going to be buying agents that will do a lot of this work for you. You might have primary agents and then secondary agents, like what I just showed you. That's the architecture. That's where it's going. I don't think software is going away by any means. I think, not financial advice, but I think these things have been pounded, overpounded. That's what I would say. I'm excited about this future. I think it's worth it for you all to be putting the time in to learn this right now. Because this is how the future's going to be set up. Going back to the Elon quote from earlier, the podcast that he just did recently, fully agentic teams and robotic teams are going to crush. He didn't say crush, but they're going to outperform teams with humans, just straight up because they're infinitely patient, they work forever, and they have shared context. Then these Optimai robots have actually more dexterity than humans. Whether we like it or not, I think that's the world that we're going into, but that's what it is. VPS type recommended. I don't have a good VPS to recommend you. Maybe someone else in chat can. Then let's see, how to be efficient with tokens, not compact or clearing context. Any tricks on that? I would say, look, I had a cute setup with OpenCLAW saying, hey, For most of your things, use Sonnet. And then for anything that requires thinking, use Opus 4.5. And it was working. But the thing is, when you use a lower quality model like a Sonnet, not lower quality, but just less thinking in general, less tokens required, you get what you pay for. And so the way I look at it, if you really want to save right now, it's probably buying some computer locally and just running KIMI 2.5 on it. And I think you're going to save a lot of money that way. We're talking what, 95% savings on costs? That's how I look at it. And yeah, um, Hostinger has been recommended quite a bit, Francisco. Um, let me see what other questions we got here. Do I have the Q&A section? I should use these polls and all that too. Look at all these questions I have over here. Okay, cool. Um, let me— I'll answer maybe like one more question here, uh, before we go. Who is the provider? Cost of Alfred? Oh, okay, so we'll answer that. Okay. So guys, because OpenClaw is open source, there's no real cost to set up Alfred, but you are paying for— I'll just give you my cost around the Mac mini stuff. Okay. So Mac mini, I paid about $1,000 for it. So I upgraded the memory, I upgraded the speed on it. And then I am paying again $100 a day for my tokens right now. And that's kind of it. So let's just call it like $1,000 initial setup for me. But again, you don't need to buy a Mac mini. You can set this up on an old laptop if you want. If you have an old computer, just think about setting it up there. Then there's a lot of configuration you have to do, like making it headless, making it not sleep and all that type of stuff. It doesn't go into— because if you let it sleep, then you can't access it anymore remotely. Those are my costs right now, but I am looking at bringing the cost down quite a bit. You can also work with your OpenCLaw to figure out how to bring your cost down. You can type /usage and work with it. There's a lot of nuances. Even the things I'm talking about right now, I'm like, You're going to have to learn how to use these things and then you'll figure out these nuances along the way. Look, if you enjoy this, go ahead and subscribe on YouTube. It's Leveling Up with Eric Siu. You can check me out on all the other socials as well. Then I have a newsletter too, or whatever else you found. Then I think a lot of you are now on the newsletter, so we'll be sharing more stuff there. I'll be doing these lives a lot more. Let me know what else you want to hear about. You can drop a comment in whatever social channel you're coming from right now, what else you want to hear about. And, uh, that being said, I hope you all have a great weekend. I hope you all just start testing this stuff. Please start testing this stuff. You can't afford to fall behind, and I won't let you fall behind. Just keep sticking with me and, uh, it'll be good. Have a great weekend.










