How OpenClaw & Claude Code Generates Real Revenue [LIVE]

All right, everyone, we are live. We're going to be talking about how to make money with OpenClaw and Cloud Code. More importantly, not just how to make money, but how to drive revenue. So we're going to wait about a minute, 2 minutes or so before we get started. And so before we start, go ahead and type in the chat right now where you're coming from. It's going to be exciting. It's going to be action-packed. We're going to go through, I'm going to be working live. I'm going to be showing some things that we're doing over here and we got to keep some of these things secret. secret too, because this gives us advantage right now in business. So go ahead and type in the chat right now where y'all coming from. Share this with your friends, share this with your mom, your dad, share it with your dog as well. It's going to be a great time. And before I do that, I'm going to tell my AC to turn it up a little bit. So give me a second. Okay. So what we're going to do here is I'm going to show you a couple of workflows. I'm going to show you how we're using OpenClaw. I'm going to show you my Telegram instance in terms of how my My team is set up, how they're all working together. I'm going to show you the ROI of what we're doing right now as well, because a lot of people are talking about Claude Code. They're talking about Open Claude. They're talking about Manus. They're talking about all these cool things. But when push comes to shove, they don't really talk about how to actually drive revenue from this stuff. And that's what I'm most interested in because the way I'm working with this now, literally what I'm doing is earlier this morning I was at the gym and I was cranking on this. Let me go live on Instagram. as well. I was at the gym working on this. I was on the treadmill just typing away. Literally, I was talking to someone on my team and she's like, how are you doing this right now? I'm like, literally, I'm just typing this out and that's how I'm doing it. Again, if you're just joining us now, type in the chat where you're coming from and we are going to get started right now actually. Let me give you a brief overview because we have a lot of people coming in from LinkedIn this time. The main thing is OpenClaw is an autonomous agent. It's an open source autonomous agent that you can use. Now, Does it have security issues? Absolutely it does. Okay, so that's something we can talk about a little bit as well. You guys can start to drop questions into chat and I'll go through them over time. And then you have Claude Code, which is amazing to me because it's an agentic coder that you can use and it allows you to create anything you want from scratch. And so really in the last 6 weeks or so, there's been a major step change in how everything is, is being done, how we work. All right, so when you, when you hear people talking about, oh, you know, um, coding is no longer gonna be a job in the next 12 to 18 months or so, anything you do on a computer right now is not gonna be really a job in the next 12 to 18 months or so, I'm gonna show you that that is not an exaggeration. That is a reality now. And, um, that, that's just where things are going, right? So we got people, we got, um, we got people coming from Toronto, New York, Medellín, um, and we got Vincent coming in from Amsterdam. Amazing. Okay. We have Bucharest, Romania. We have Spain as well. So we have an international crowd. That's cool. Miami. Miami's amazing. Southwest Florida. I'm actually going to be in Florida next week. So love it. And Christophe, I'll tell you in a little bit if this is going to be recorded or not. I'll tell you in a little bit. Okay. Munich, Lebanon. Okay. That's great. And you guys can also type in the chat too, like what would be a win for you from this conversation today, this livestream? Because again, I got no slides from you. I'm literally, I'm going to be screen sharing. I'm going to be showing you how I work with this. So you can drop questions in the chat as we go here. And oh look, even from Instagram too. We got people coming from California, from India, from LA. Okay. Toronto, Sydney. Amazing. Okay. So international crowd. Okay, cool. So here's what I'm going to do. First, I'm going to show you a couple examples on how this is working for me and how this is actually generating revenue. And then I'm going to move into my workflow on OpenCLAW, on Claude Code. In this case, you're going to see me kind of working and then talking at the same time and answering questions. That is how today's going to go. This hour goes by really quickly. I might even start doing this for 2 hours, 3 hours or so. So hopefully you all still have me over here. I thought I got disconnected for a second. Okay. Let me show you first and foremost what is possible. And by the way, before I do, I forgot to introduce myself. So my name's Eric Siu. Main business is a ad agency called Single Grain. And then we have an SEO software called ClickFlow and a LinkedIn account-based marketing product called Karrot. That's K-A-R-R-O-T. And I do a podcast called Marketing School. And I have my YouTube channel, which is Leveling Up, which this will be posted to later as well. So let me show you what is possible right now. So literally this is my Twitter over here. I'm going to show you how my OpenClaw is able to generate X articles for me. So these are blog posts on X. And each of them is basically averaging about 100,000 views per post. Okay. So if y'all would like to see that, well, this is exactly what I'm going to show you right now. So let me share my screen and I'm going to move over here. Okay. Y'all can see this? Yes. I trust it's yes. Okay. So check this out over here. So these are my latest articles on X. Okay. Look, you can see over here, I'm not lying. The receipts are over here. 64,000 views. All right. This one over here, 101,000 views. This one over here, 85,000 views. And then this one over here, 85,000 views. Okay. So when you click into one of them, okay. So literally the title over here, by the way, I generated this myself using Gemini. But I, you know, I helped with the title. But other than that, this is all written by AI. Everyone on X is talking about building OpenCLaw agents. Many are quietly failing. I asked to build a review queue, right? It's written in my voice. So it's very hard. Like I'm telling you all, you think you can tell what AI is in the next year or two, you're not going to really be able tell what's AI. You can see there's little diagrams. There's little diagrams that you can copy and paste as you go through this over here. But this is a legit post. All I do is I work with my OpenClaw on this, which I'm going to show you in a second. Look, I even promoted this livestream that we're doing today down here. My point of showing you this is, look, 1,700 bookmarks. These are real results. You might be wondering, okay, that's cool, Eric. You're getting all this reach. But how is that driving business? Well, let me give you an example. I literally have in my OpenClaw, I have a deal of the day and I have a connection of the day, a workflow that runs every single day. The deal of the day will look through our CRM, it'll look through our goals, it'll look through our ideal client profile. And then based on maybe deals that have stalled out or some that we haven't talked to for a while, or It'll basically try to resurface those. And we've actually resurfaced a few of those conversations on my sales team on the Single Grain side. Okay. That's one. So that is found revenue if you can get it. Because a lot of salespeople, like it's, we're all human, right? It's not even just a salesperson problem. It's more so as humans, we forget to do things and we want to work on the highest leverage things first, which might be the active deals that we're working on. Okay. That's one example. Now, another example I'll give you is a meeting that was booked with a multi-trillion, with a T, multi-trillion dollar company. Now, what happened with this? OpenClaw found the angle. It looked through all my LinkedIn connections. So let's say I have about 13,000 LinkedIn connections or so. Okay. And it was like, hey, you know this person over here, you haven't talked to them for a while. Here's how you're connected to them. Here's the angle that you should hit them with. I'm like, I'm lazy. I just took the email and I just— it was an InMail actually. So I sent them a LinkedIn message and they're down to connect. We connected the next day. And then that person was like, hey, you want to come speak at our event over here? Again, Multi-trillion-dollar company. That was only the first thing. That led to, hey, I think we have some other angles here and here. So angles 2 and 3. And then actually I'm grabbing food with him tomorrow. So that is an example of what you can do now. And where did I get this idea from? When I was recording Marketing School with Neil, my podcast co-host, he's like, dude, I'm working on this 9-figure deal right now. I'm like, okay, well, how'd you get that deal? He's like, you know, what I do every single day is I make sure that I send at least one text message or one email. And I just think about what the angles are with the people that I know. And I ask for connections. I'm like, that's cool, but that's very manual, right? And I'm not saying don't do it manually. Definitely still do that. But if you can have something that understands how you think about this and it just does it every single day for you, then why wouldn't you do that, right? Just like every single day I'm getting these angles, these ideas, and who knows my goals, all my granola calls, all of my, the Gong calls that the team has been in. All of my content is ingested as well. It knows all my goals. And so it also knows how I work. And so what better partner to use to come up with content ideas or sales angles or marketing angles, right? That is the power of using something like an OpenCLaw. So I'm going to show you right after I show you this chart over here, I'm going to literally, I'm going to pull up my Telegram. And I trust that you guys will keep any sensitive information to yourselves. And yeah, so, okay, cool. So this is what I was talking to someone on my team yesterday and they're like, oh, they're just machines, Eric. They're robots. I'm like, are you kidding me? Do you really think we're smarter than these things? And if so, maybe we are smarter right now, but if so, for how much longer? So the thing to pay attention to here is in 2023 when ChatGPT first came out, or not first came out, but ChatGPT 3.5, 2023. Okay. It was like a dummy, right? 70 IQ score. And then when ChatGPT-4 came out, we're talking about maybe like a high 80s or so. And then it moved to 100 over here. So average human, right? So look at this. Within a year and a half, less than a year and a half, already got to average human. Now you look at 2025, which has passed already. We're in 2026 now. All these models over here are already genius level. And then it's rising over here to top 0.1%. And so even no matter what, if you look at conservative estimates, it's going to be— these things are already going to be way smarter than us. When you look at a linear projection, this is where it goes to. But I was like, you know what? On this chart, what's missing over here? An aggressive projection. I want to see an aggressive projection. So an aggressive projection is like, it's going to scale all the way to 1,000. So there is no other way out right now than through. That is one of the points I want to get across right now. So if you're watching this right now, if you work for a company, you guys aren't talking about this right now, might be time to look for another company, right? Because this company's going to be left in the dust if they don't adopt this stuff quickly. Now, if you're a leader, you have to get on this stuff. It doesn't need to be OpenCLAW because there's a lot of security things around that, but there's— use Claude Code, use Claude Cowork, use Manus, use any of these things, right? And that is why I'm doing these every single week too, because if I can even get 1% of you all to change, that's worth my time. Plus I get views too, so that's great. So anyway, okay. Let me move over to Telegram over here and I'm going to reshare this and then I'm going to answer some questions in a little bit too. So let me do this and let's go to Telegram right here. Okay. So here's what I got from my Telegram. Okay. As you can see, this is Alfred. Okay. Alfred is the butler from Batman. I also have Cyborg as well. Okay. So I have Cyborg, I have Oracle, I have Flash. These are the heroes that I have in my squad. Let me show you how this works so you can visualize this a lot better. So if you look at this, this is like a dashboard. We call this Mission Control. Now within Mission Control, and this is just like a mockup over here, but I will say this is actually actively working for us right now. I can see, hey, here's the deal of the day. What I was mentioning. Okay. What's the angle? The deal of the day. Should we— Should we approve it? Should we reject it? And the OpenCLAW, again, the autonomous agent learns over time. If you disapprove something, why did you disapprove? It has memory too. It remembers over time. It has personality as well. And the key thing with this is because I'm working with it in Telegram or working with it in a terminal, like a command line interface, which is honestly not that scary once you start doing it. It's not that bad. But there's so many things that I have it running on. There's these skills that I've created. There's our, what are we doing on the marketing side? What's the history of things that we've done? Because you want it to remember all all the things that you've— the decisions that you've made with it. But the whole idea here is that you have this entire team over here. So this team, and I hope you can see this because on my Instagram, oh, it's just a little delayed. So I'm just looking at that screen too. So you can see, oh, recurring every single day. I have a morning SEO digest. Show me how my Google Search Console is doing. I have a deal of the day. Look at HubSpot Gong pipeline, top prospect card. And that just pulls from the sales team. And then I can kick it over to the sales team. Eventually I can just have them take over the bot. Content multiplication. So how do we repurpose maybe the content that I put on YouTube, for example? And then how does it become spawned over for X and LinkedIn? Because it knows how to write in my voice and it's actually, the content's actually performing. It's not like any other thing that I've tried before where it's like, yeah, we're going to write content for you. It's like the content is actually good. We actually will track over time too, because I believe that culture is a 60-day moving average at my company. Really at any company. So we'll see over time if maybe, for example, I'm saying certain things that maybe aren't living up to the values of the company and maybe that crosses over into performance management. But the key thing for us is we need to make sure that as we're trying to get people to adopt this, that we continue to make sure that we have people with the best judgment on the team. And I'm going to show you a little bit of our AI fluency stuff in a little bit too. But the whole idea here is that because you have a Kanban-style mission control here, It's a lot easier to manage what you have. Okay. So that's what it is. Oh, the photo's not showing. Okay. That's my fault. It should have pulled up. So let me try to pull it up here. Give me a second. Let me see. Replace or— you can't see it on LinkedIn. Okay. Let me make it— sorry guys. I'll make it a little bigger. Thank you for the feedback, by the way. So let me do— I'm going to replace it. Actually, you know what? I can probably pull it up. I'm gonna pull it up in Slack. Okay, give me a second. Let me pull it up in Slack and let me change my screen share. So let me do this over here and I'm gonna show you guys some other stuff. Okay, cool. Let me switch over, uh, replace. And thank you guys for bearing with me. All right, can you all see this? Okay, looks like you guys can see this. Okay, so this is Mission Control. So this is what I was trying to show on my screen, but Again, it's like a Kanban board. You have the recurring tasks, you have the queued up, in progress, review. The whole idea is when you're working with Claude code or even with Open Claude, it's hard to maintain all that context in your head. What was happening for me when I was managing these agents was I was like, oh man, I'm forgetting a lot of stuff. I'm repeating a lot of the same tasks that I've already talked about, and sometimes I'll overwrite what I did already. That ends up being an unproductive use of time, and so you don't want to do that. You can see it on the right side. Yeah. There's a live feed here on what was approved, what was disapproved. Whatever you disapprove, it's gonna learn over time. You can see here's all the skills that I've made. So I can have like an SEO skill, a programmatic SEO skill, an X long-form post skill. I can even insert my Loom videos into it and ask it to just make a skill from that. It can just do it autonomously. So I think of it this way, Claude Code, which I'm gonna come back to, I'm gonna show you a screen share in a bit maybe, is more so your brain where you can plan things out. And then I look at OpenClaw as your hands. These things are the execution layer for you. And so the way I look at this is you have this team of agents over here. Again, Alfred, which is my chief of staff, basically it has the most context. Then you have Oracle, Arrow. I have a self-healer here where it'll look at all the tasks that I have going on and it'll sometimes tell me, hey, it might be time to delete some of this stuff and manage it better. So that way you're not burning through a lot of tokens and wasting a lot of money. And one of the questions I get right now is like, how much are you spending a day on this? And I would say I'm probably averaging like $1,000, not a day, $1,000 a month or so. But the key thing is that you're continuing to work with this, like this mission control. Look at it as you're compounding a business. You're continuing to work on it every single day and it's getting stronger and stronger. And the stronger you get, the stronger your business gets, right? And a lot of my friends right now, I've gotten them on the Claude code train, but they're not quite on the Open Claude train yet and for various reasons. And what I would say is, I actually have a post on X talking about how you can set it up correctly. You can check that out. I have a little article section if you just go to @EricOSIU. So that's that. So I'm going to move back over. Let me stop sharing this one over here. I'm going to move back over to my Telegram and show you what's going on in there. And I'm going to answer, maybe after I go through my Telegram, I'm going to answer some questions that you guys have over here. Okay. I'll answer this question. So how would you build a mission control? What is mission control? Did you build it or is it part of a product? Okay, great. Basically all I did for that, guys, is I took a screenshot of a Kanban board that I really liked and I went over to my open call and I said, hey, I really like this dashboard. Based on all my goals that you know and based on how we're working together, we need something to manage everything, all my content. I told the problem. I was like, my problem is I'm forgetting things. We're tending to rebuild the same stuff and it's becoming a really unproductive use of my time. I want to make sure that this compounds with me, blah, blah, blah. What I say at the end of it, I was like, just please ask me clarifying questions before you begin to build. It asked me some clarifying questions and then boom, OpenClaw's off to the races. That dashboard screenshot that you saw, that's actually live for me right now. That came from me prompting OpenClaw, dragging a screenshot in, and vibe coding it. Okay? So that's what I would say. Amit, if it's blurred, I would just go check out my— go check out X or check out my Instagram. I'm talking about this stuff all the time. And then Josh, to answer your question on what mission control is or what it is, is it's just basically my abilities. It's managing my abilities to manage my autonomous agents so they don't go off rails. And so I can manage the context within my brain because OpenClaw's brain or these models' brains are a lot stronger than mine. So I need it to slow down a little bit so I can keep up. So that's what it is. So Vincent asks, is Mission Control an open source application or did you make it yourself? Look, everyone wants Mission Control. Check that out, right? So here's the thing. Here's what I'll say. So our plan at Single Grain is we see the future. I know there's probably a handful of you that are agency owners here. I think traditional agency model is screwed. And so I see the future of this as being able to have some type of mission control where the agents are doing most of the work and we have a forward-deployed marketer, which might be working full-time at that company or even part-time, but they're coming up with the strategies and they're constantly customizing the squad of agents, which I'm going to show you in a moment. And so we're actually working on piloting that with a couple of clients right now, and they're interested in that, but I want to make sure that it's working really well for me because I I'm going to have the highest standards obviously on this stuff, make sure security's good, all that type of stuff, right? And so that's where we see the future of this going. And I'm actually interested if you type into chat right now, just type yes if that's something you're interested in piloting in the near future, just type that in because I'm just trying to get a sense for one, it's really helpful for me and I actually did an ROI calculation for this. But I do think this is kind of where the future of work is going, where you just have something that works really well for you and you customize it for each of your customers. What's going to happen is I think a lot of agencies, traditional agencies, I think they're going to be reducing their headcount. That's what I would say. All right. Raza asks, how do you handle memory? OpenCLaw seems to forget the latest information on each project. There is a memory.md and I would say it's not exactly perfect, but I would say this stuff is getting better over time because Opus 4.6 just came out last Last week, I think. And they're adding, I mean, it's like a million context window now. I would say it doesn't forget as much as it did before. So there's that. Look, everyone's typing yes in here. Okay, that's cool. So by the way, guys, I'm showing you how to— I'm giving you a sense for how to do this right now. And there's so many, if you want to learn this, again, you can check out the YouTube channel for more videos on it, but I'm just trying to feed you a couple nuggets here so some of you will take action. But that's how we're going to be doing things. And the fact that so many of you had said yes, that's something they're interested in piloting. That tells me a lot. That gives me a major signal. And what I will say, those of you that are in software right now, if you're pure SaaS, pure-play SaaS, what we're doing with ClickFlow and Carrot, for example, we're probably going to have a lot of freemium versions of different products that we build that will then feed over into upselling them, cross-selling them into something like a mission control. Because ultimately, we still believe that you're still going to want a human, a smart human to guide these things and do these things at scale, right? Okay. So let me move over to Telegram. And Seth, to answer your question, do you have a good AI to do Facebook or Google Ads through Claude? You actually can set that up in Claude Code if you want, even Claude Cowork. There are skills that you can use. Open Claude, you can even do it as well if you want to. I am just— I don't know if I'm down to let it YOLO because if I let it just spin up whatever creative it wants and optimize bids and all that, I'm not down for that. The way I have it set up right now is more so, I will give it feedback and it learns on my feedback over time. I'm not quite ready to let it go autonomous. And so you want to think about different levels of autonomy. So maybe level 1 is just, it's giving data to you, it's giving reporting to you. Level 2 might be maybe it has some recommendations for you. And then level 3 is maybe you can do some small things. Level 4 is it's fully autonomous, right? So I don't think we're quite there yet. And I don't know if from a business standpoint we ever want it to be. Completely autonomous, but your mileage may vary. You decide. I think for personal stuff, sure, let it YOLO, right? But even then, am I going to give it my credit card information? Am I going to give it access to all my passwords? No, because if a prompt injection attack happens or it scans a PDF and then boom, game over for you, right? Then that's a huge mess for you to clean up. So I'd rather not deal with that. That's why I have my little Mac mini next to me. So my OpenCLaw is ran locally. The only real attack vector is attacking that Mac mini. Then I have allowlists and things like that. Okay. Let me move over to Telegram. Josh, I'll come back to you in a second. That's a good question. I forgot too, by the way, I forgot I have these cool QR codes on this thing. If you guys are interested in that Mission Control Pilot, just Scan this QR code over here. I'm going to leave that QR code up for a little bit. Okay. So let me move over to Telegram. So let me do this. Let's go over here and let's go to Telegram. I'm going to show you some other cool stuff that's happening right now. Okay. All right. So this time I'm going to take this off. Okay. So what does this do for me right now? I'm like, okay, well, every morning I'm just going to show you a couple of things I do with it. Okay. There's a roundtable. Okay. So Agents reporting Oracle provides SEO and live recommendations. Flash does trends and repurposing, deal of the day, missing strategy. There's some other things, right? So it knows I'm doing a livestream today, by the way. It's like today's OpenCLAW demo has perfect content application setup. Flash identified 13 repurposed pieces. So guess what? I actually went to Flash, which is my content repurposer. I was like, hey, give me the angles I can use for this livestream today. And let's actually take a look. And see what it came up with. Check this out. I literally prompted up here. I was like, hey, I like these talking points. I made it into X-Posts, but let me scroll up a little more. Let me just show you what I wrote over here. This is gold saving, blah, blah, blah. Great. Livestreaming talking points. Here we go. Based on what you and Alfred actually built this week, you have multi-agent coordination, which is your biggest demo, which I'm going to show you guys in a second. So multi-agent demo. Show how Oracle, SEO Flash, and Alfred share a brain. Live example, Oracle spots YouTube consultant page converting at 9.5%. Flash creates content targeting that keyword. Coordinated execution. Most agents run in isolation. Mine talk to each other. What I just mentioned to you guys earlier, 1Password integration for secure API access. And so what I do is I make sure I just have 1 vault that this thing can access and that's it. Just 1 vault. Okay. Then I'm going to scroll down and I'm like, okay, that's cool. I got some talking points. I can riff with you guys for a little bit. But guess what? I'm like, hey, I want you to make this into an X article because like I showed you guys in the beginning of this livestream, my X articles are crushing it right now and each one's averaging close to 100,000 views per post. By the way, X articles are great for reach right now. It generates this. But I'm like, oh, this doesn't quite sound like the skill I gave it. Is it perfect? No, I need to guide it a little more. I'm like, Hey, did you forget this the spec from the X article skill that I that we worked on together, right? Oh no, no, I don't see anything right now, right? Like it's it's like are you sure? But then it's like wait, I found it, right? Okay, so now I have the spec. My bad rewriting. Okay, this is what's special about my X articles right now. I wanted to always have these diagrams in there because people like these diagrams, right? So here's the spec. My agents use my AI agents used to work alone. Now they coordinate. That one change was worth more than the last three months of building. I have three agents running on Open Claw. Oracle handles SEO, Flash handles content, blah blah blah. And then I just I just scroll down here, right? And so it might have some sensitive information in here, but the key thing is I have these diagrams. I have probably ninety percent of it there. I might fix the hookup here, and then I might have a really good title because yesterday when I first published my I remember I published, and for the first hour or so, the the post I have that now has like sixty five seventy thousand views. It had one like on it. I'm like, oh damn. So you can't just quite let it do its own thing and quite YOLO yet. I adjust the hook. I adjust the title. Boom. Now it has, I think, hundreds of likes and 65,000 views on it. Okay? So you still have to work with it a little bit, but my point is the time that you save with this is amazing. I'm going to come to questions in a moment, but I'm going to show you maybe some more work I did with Alfred this morning. So I was like, hey Alfred, I'm going to present A livestream today, I want you to tell me the ROI of this because we've only— by the way, I've only worked with Alfred for 10 days, guys. 10 days. Imagine in a year, if I continue to compound with this, what's going to happen, right? By the way, let me switch the call to action now to this newsletter signup over here. Okay. So that QR code now on the top left, that is now the newsletter signup. If you want more stuff like this, you want to stay in the know, so you can just scan that. Okay. So AI agent ROI. I'm like, okay, well, explain to people how this works. Okay. So the Anthropic API, $800 to $1,000 a month. Okay. Infrastructure, $15 a month for Mac mini power and Brave search. I didn't even know it was $15 a month, but yeah, my Mac mini's on the whole time. So it doesn't go to sleep. It's headless, which means there's no monitor tied to it. By the way, did I know how to do any of that stuff before? No. I didn't know how to do any of that stuff before. But Someone asked earlier, I think it was Josh, like, okay, well, how do I go and learn this stuff? You know what? All the action right now is on X. If you want to follow the right people for this, I'm like, okay, well, you go to Grok and say, who are the best people to follow for Claude code, for learning OpenCLAW, for learning about OpenCLAW security? What are the things I need to consider? What are the things if I want to grow my business? What are the blind points? What am I not considering right now? And just ask it. I ask stupid questions all the time. I ask stupid questions every day, probably every hour, right? So my point is, I believe, okay, so some people don't like it when I say it. I just believe that if I can do this, I believe anybody can do this, right? I think the fact that you all have high agency and you're on this training right now says to me that you're gonna be able to figure this out, okay? So, okay, so value created over here. Basically, I'm just gonna scroll down. It's like, okay, like all this stuff over here, here's all the time savings. So 28 hours a week or so. So I don't know why it estimates at $50 an hour or so. I think it's for each of these roles, but conservatively, okay, if we're paying $6,300 a month or so, and the monthly cost was $1,000, that's pretty good ROI, right? Well, SEO opportunities, Oracle, it's uncovering SEO opportunities, decaying converting pages, keyword opportunities that we should be attacking, cannibalization issues that we're facing, okay? Or even yesterday, I just added a recruiting bot that my recruiter is going to be working on, right? And it's sourcing amazing people. I'm like, you see? She just came back recently from leave. And I'll just put it this way. The world that she's in now, man, I'm like, all of that, all those tasks you're doing before, forget it. Okay? I showed her Monday how to use Perplexity Comet too. There's like 400 applicants for a thing, right? I'm like, you know what? You just have Perplexity Comet go through it. You don't have to deal with it. Just go watch TV. Go watch TV. Go do something else. Okay? When you're working with Cyborg here, I'm not going to show you who we're talking to on the cyborg side because it actually shows people we're sourcing and there's that information. I'll leave that out. But literally it's identifying, oh, engineers, we should look at their GitHub. How many stars do they have? Who's interacting on X right now? Oh, if it's someone that's more in an operations role or something like that, okay, well maybe we should just look at their LinkedIn and maybe if they're posting to X. I can train it on different nuances. It's also trained on our values, our standards of excellence, all these things. I just drag those documents in. Left and right, left and right. Okay. How do you build these things, Eric? I saw something I really liked on X and I just put a picture in there and I just said what I wanted to do. That's it. And then I just keep refining. Literally, you're building with it every day. And then maybe sometimes you build something that you're unhappy with. You're like, hey, and I'll drop this prompt in the chat. This prompt's really good. So let me see, where's my chat? Let's come off the QR codes and let's go over to Chat. Where's my chat? I'll need to figure out how to get back to our chat. You can see— oh, there it is. I see chat. Okay, let me use this over here. So here's a prompt you guys can use that I'm putting into chat. Okay, so literally what it says, and I'm going to show it right now on the screen. Look at that. So now that you have built this, what would you have done differently? What should we reflect on? I'm going to say this slowly again because I talk fast. Now that you have built this, what would you have done differently? What should we reflect on? And then you can refine it and make it better over time. Okay. I'll tell you what, I'll give you a little bonus prompt over here too. And by the way, before I do that, are you guys enjoying this right now? Are we on the right track? Are you getting value from this? Type yes in the chat if you're getting value or no, this sucks. All feedback is good to me. Okay. Okay. Now here's the other prompt. Before you return a version to me, so let's say I'm looking for an article or maybe I'm some type of output, any output. In this case, let's say it's copywriting. 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 from scale of 0 to 100. Each expert should include specific areas for improvement that are reflective of the ranking. If the average ranking isn't over a 90 out of 100, then go back and improve it using the expert-specific recommendations and feedback until it is over a 90. You see that? This is, you can call it recursive improvement on whatever the output is that's coming out to you, but now you don't have to put in your thinking. You can just have it improve on its own. Guys, you'd be surprised too. You don't have to try to impose your thinking on it all the time. Because sometimes it's like working with somebody else. Maybe they might have come up with a more elegant solution than what you come up with, right? And that's totally okay. So, all right, somebody asked to see more about Open Claw. Okay, so let's go into Open Claw a little more. So this is the ROI, right? So let's say the monthly cost is $1,000, monthly value time savings alone, I think it was $6,300 a month, okay? Conservative ROI, okay, 6 to 1, including SEO opportunities you've captured, 14 to 1. I think this is actually very conservative. To me, it's doing much more than that right now. It ran it yesterday and it actually was like 27 to 1 or so. But my point is, guys, I only spent 10 days working on this. Imagine how this is going to look in the next 350 days or so. This is going to become way more powerful. And that's why we're building Mission Control where we make sure that it looks really good to me and we've compounded all the learnings and then we build on other people's data and their learnings and then hopefully build it in a more a very secure environment for them, just how I have it built locally for me. With OpenCLAW, those of you that are thinking about giving access to your team, we're not teaching our team how to use OpenCLAW right now. We're teaching them how to use Claude Code. You might be thinking, and I'll stop my share for a second. Actually, no, I won't. Let me see if I can— let me adjust this a little bit. Let's do this. Does this work? Okay, cool. Oh, that's better. Look, my face is a little bigger now. With OpenClaw, there are too many security risks to just give it to your team and let them YOLO with it. With Claude Code, we just want them to build solutions for manual tasks that they're doing right now that they shouldn't be doing, right? And we ask my team quite a bit. I'm like, why are you guys doing all these manual tasks? On a scale of 1 to 10, how much do you enjoy doing paid media management and keyword research? They're like, maybe like a 1 out of 10 or 2 out of 10. I'm like, good. Then get angry. And get those, like work with Claude Code to get those things off your plate or even use Claude Cowork, which you don't need to know how to even need to use Claude Code for. Cloud Cowork is almost like you're using ChatGPT or Claude, for example, you're just prompting it. Just keep that in mind. Let me move over into— let me see what I can show you on the Alfred side, which might be interesting. Let me see what I can show. I think I can show you something. Let me show you a daily deal of the day right here. Okay. So check this out. So if I scroll up, it's going to show you all the information, but this, like, by the way, I'm going to give this to my sales team eventually, this skill where I'm hovering over my mouse. Okay. So, hi, Kenton. It's been 92 days since we last connected. I'd love to schedule a quick 15-minute call to review where things stand with your marketing initiatives, address any questions that came up since our last conversation, discuss immediate next steps for Q1. Would you have time for a brief call next week? Okay. So I can edit this email. I can send the email if I want, or I can show another, or I can skip. But the whole idea here is that, okay, if this is working single shot for me every day, why can't I just give the entire thing to the sales team and have it do as many as they want every single day? And then we can track how this is actually affecting pipeline over time. Okay. That is possible. But you need to make sure that just because, by the way, what I've learned with this, even Claude Code, I didn't know how to use Claude Code or OpenCLAW. We're talking 6 weeks ago. But what I've learned is that you can build all these alerts, all these systems, but if you don't have people on your team that are willing to take this stuff on, then it's just going to sit there. You got to make sure that you have people that are very high agency and people who have very good judgment, and then you empower them with these AI powers. Julian asked the question, so this is all Claude Code? This is using Opus 4.6, so that's part of Anthropic, but I can really use any model for it. OpenClaw is an open-source autonomous agent that you can use. I'm going to show you Claude code in a moment. I use it within Cursor, but I'm using this now. The way you see me chatting with it is how I'm chatting with a human being, just back and forth every single day. I can be sitting on the couch, oh, we should do this over here. Oh, what do you think about this idea over here? By the way, I'm about to make it a lot sassier today. In fact, maybe I'll do it live with you guys. Maybe I'll make it a lot sassier because there's a soul.md where you can make it fun to work with. Or more fun to work with. So look, you can see a couple things right now. By the way, you can see I'm also upgrading my agents here. It's like, I don't know if you guys played StarCraft or Warcraft when you upgrade your weapons to level 1, level 2, level 3. That's what we're doing right now. So I'm like, hey, be honest with me. With Oracle, Flash, 1 through 100 right now, how are you set up for success and what do we need to do to make you more high leverage, higher leverage as it relates to To my goals. I think it rated Alfred like a 15, and then Oracle was like, I don't know, like a 25. My point of saying all this is that you should constantly be asking questions. Hey, what am I missing? Because chances are if you build something, you're like, something feels like it's missing. That's when you start to prompt and work with it back and forth, because sometimes you don't even know what the problem is, but something just feels off. That's why you should be asking those Those questions. Yes, train that thing. That's right, train that thing. You train this thing and then you train your team as well. Okay, let me— let's take a quick digression over here, or digression, let's take a quick detour. I'm going to answer some questions, but I want to show you something because someone talked about training your team. And this is a great 2x2 that I saw this morning and I was like, oh damn, like this is really good. Okay, so let me, uh, let me press my hotkey over here. Oh, that's That sucks. Okay, let me replace this screen and let me move over here. Let's move over to X. Where's my X? Let's move over here. Let's go over here. Okay, so I had showed you this earlier, but let me share this tab over here. Okay. All right. So if you guys look at this 2 by 2, so let's say you have someone on your team that does not use AI, one, and they do not have good judgment. Game over. Dead weight. What is happening right now? Let me give you guys some insider baseball right now. And you guys can type this in the chat too. I'm curious to see what you say. I have a lot of founder friends who message me. We are in these founder chats as well. What do you think is happening right now? Are they saying we're going to We're going to hire more people right now? Are we going to stay flat? Are we going to cut people right now? What do you think the general sentiment is? Are we going to train them on AI? What are we going to do exactly? Drop that in the chat. I want to see what you guys think and I'll review that in a second because there's a little bit of a delay. Gabriel, I'll come back to your question in a second on the skills because that's a good question. By the way, let me hide this message up here. There we go. So coming back to this 2x2, okay, you need to make sure that sure, you can use AI, you're power users, maybe some people on your team are power users, but that's not enough. Because if you allow your hiring process to be half-assed and you allow people into your organization that do not have good judgment because you needed to quote unquote fill a role or it was painful, you are shooting yourself in the foot, okay? You are allowing people to come into the company To inadvertently, okay, because most people have good intentions, but inadvertently destroy your organization. Okay? So that's why right now, I've locked in even more on our hiring process. And I'm just like, dude, man, okay, how do we have good judgment? By the way, I'm going to show you our Beat Claude challenge too. That's cool, right? And maybe I'll go through that in more detail next week. So you want to hire people that have good judgment. Let's say they don't use AI right now. Okay? They're steady hands. Okay? That's great. You can train these people up. Okay? Because if they know how to use AI plus they have good judgment, boom, turbo brains. These are the people, like, you can continue to hire some of these people right now, but net-net, we're looking to hire turbo brains right now. And some people ask the question, like someone was asking on Instagram, like, because I showed them this Beat Claude challenge, right? It's like, well, you know, if you're going to hire exceptional people, are you paying them exceptionally well? Like, yes, you should be paying them exceptionally well. Obviously it depends on the role, but I'll tell you one more thing. You look at the greatest layoffs in 17 years that happened in January, right? In 17 years, you see big tech companies aren't really hiring that much anymore. What is happening right now? What is happening right now is people are making sure that— well, not only that, people are getting cut. That's the thing. People are getting cut. And because people are getting cut, what's going to happen? Let's say you are a longtime VP or director of marketing. Okay? You've worked at a company for a very long time. You just got cut. Okay? You have a family of 2 to 3 people to feed, maybe even more than that. You don't think you're going to do everything in your power to make sure that you learn all this stuff and you crush the competition? Or if you're coming out of college right now and all you know is AI, are you not going to make sure that you are hungry as— there's so many people right now that are hungry. And the fact that people think they can just chill and coast as if the job, the world of work, the job market, was the same even a year ago or 2 years ago. Get real. There's so many people. You see it in the numbers. Okay? There's layoffs happening left and right. People are keeping company hiring flat. And so that means there are more people out there to eat your lunch. Maybe not your lunch because you're watching this right now. You're high agency, right? But there are people, your team needs to understand. I'm trying to shake my team. I'm trying to shake my friends. Some of my friends don't understand this. You know, I'm just like, and then they just keep talking about how they're working and like from like, you know, like a year or 2 years ago. And I'm just like, almost like I look at it in disgust. I'm like, how could you do that? Right? And so, you know, by the way, if you're watching this right now, like it is within, I'm not saying you need to share this webinar, by the way, but like go, you got to shake the people around you, the people that you love, the people that you care about, right? Because the world is changing so quickly. And, you know, I say this with passion. Like that's why like right now, I've never been more excited to work, but I've never been more enraged at work because of this, right? Because you want to bring people along with you, but inertia is a very real thing. Okay. So anyway, that's that. Let me show you the Beat Claude thing. And then I think Garrett was his name. Garrett, I'm going to come to your question. So check this out. So I have this thing called singlegrain.com/apply. Yeah, here we go. Check this out. All right, so I copied this from Anthropic, by the way. It's like, oh, Eric, you're so smart. No, no, no, I wasn't smart. I just copied this from Anthropic. I dumped this into Claude Code. I was like, hey, Anthropic did this. I took their GitHub link, this little URL. Hey, based on all our goals, I want to do this too. I want to do this too. Okay, so here's how this works. Beat Claude, the hiring challenge for AI-native professionals. So Single Grain is looking for strategists who can think better than AI. Okay, so pick a challenge from /challenges, read the brief, submit your answer. We compare it to Claude's answer, blind review. If you beat Claude, we want to talk. Okay? So AI can do 80% of knowledge work now. We need people who can do the 20% that AI can't. Creative leaps, pattern recognition, judgment under uncertainty. If you can't beat Claude on a strategy challenge, you're competing with a $20 a month tool. We're looking for people who make AI look basic. Okay? So use whatever tools you want. Like you can use AI. And then we added a time commitment here. So each challenge is designed to take 1 to 2 hours. Submissions are capped at 4 pages. We're testing judgment and prioritization, not endurance. Okay. So literally, by the way, if you're watching this right now, we are looking for you. Okay. So you can go to singlegrain.com/apply. If you think you're obsessed with this stuff right now, you want to join Single Grain. I look at this as tech-enabled services. You saw the mission control stuff that I showed you. That is the future that we're building. If you're interested, singlegrain.com/apply. So let's say I want to apply. Okay. Let's pick a YouTube strategist. Okay. That's a pretty hard one. And we are looking for a YouTube strategist. And so Here's the brief. So I click on the brief. I'm like, okay, so YouTube growth strategy for Leveling Up with Eric Siu, which is my YouTube channel. Okay. So you're being hired as a YouTube strategist for Leveling Up with Eric Siu. Your job is to own channel's growth across— okay. And this is directionally right. I have a little more subs now, but 28 million lifetime views. Okay. I also do Marketing School. And then my long-form views have gone up actually because I talk about Claude Code and OpenCLaw all the time now. So all that to say, create a comprehensive YouTube strategy that demonstrates 3 things, your AI thinking, how would you use AI tools, your YouTube strategic thinking, your specific plan for leveling up, do a channel audit, content strategy, growth playbook, and 90-day milestones. Guess what? If you apply, how much— we're talking about good judgment, right guys? Okay. If someone applies, just send them this directly. Okay? If they're not serious, if they're serious, they're going to do this. Okay? If they're not, they're just looking for a job. That's okay. Okay. Now, if you're sourcing someone, if you're recruiting someone, like let's say I reach out to you, you probably don't want to send them this first because it's like, yeah, I just sourced you. Please fill this thing out. Like you built no rapport with them, right? So you want to make sure that you have a good screening call with them, 15 to 30 minutes or so. Tell them about the company, sell them on the company, tell them why they're so great for the fit for the role. And then you send them this, right? And then afterwards we'll do like a written, you know, Like a written assessment. And sometimes the assessment takes a long time or presentation takes a long time. We might even pay them for it as well. But if you're looking for people with good judgment, then it is imperative that you put in the time to build something like this. Okay? So guys, you can literally copy this, by the way. I would just say, go to singlegrain.com/apply and take the URL, this GitHub URL that I have, dump it into Claude Cowork or dump it into Claude Code, dump it into Manus. Okay. So Manus and Cloud Cowork, you don't— there's like no learning curve. Okay. And just say, hey, I want to do this for my company. Here's all my company goals, drag them in. Here's what I want to do over here. Here's what I'm thinking. Use Whisper Flow or whatever to dictate out what you want. And then boom, this is done. Because I didn't write any of this stuff. Did I review it afterwards and then try to tweak it? Absolutely I did. Okay. So that's what it is. So Garrett, let me— Gabriel. Gabriel, I'm sorry. Gabriel, let me answer your question here. By the way, I'm going to stop my share for a moment. You guys tell me in chat what else you want to see next. Do you want to see Claude Code? Do you want to see OpenCLAW? Do you want to see me work in OpenCLAW a little bit? Maybe I'll work in OpenCLAW a little bit because I haven't really done that on these live streams as much. By the way, Ashley, if you're watching this, these QR codes are great. We've got 37 scans. That's awesome. Okay. Garrett says, what are the skills you gave Each one? And do you use multiple OpenClaw instances for this multi-agent setup? That is a fantastic question. Great question. Okay. And I'm going to answer some more questions too. Okay. Actually, Gabriel, you're giving me a chance to work inside of Telegram right now, which I'm going to do, or OpenClaw. And Francisco just— Francisco asked— actually, I'll come to you in a second, guys. Entrepreneurial ADD is very real. All right. So let me share my screen. Let's go over to Alfred. Okay, so within Alfred— oh, I want my face to be there too. Let's do this one. There I am. Okay, cool. So within Alfred, let's say, um, let's see, what do I want to do? Let's do Oracle Bot. Okay, so Oracle Bot, 6-page optimizations for you. This is on the SEO side of things. Okay. It did all this work, by the way. I haven't reviewed any of this work over here. It did all this work today. Then it also has the ability to draft in my WordPress. Keep that in mind. By the way, one thing I want to call out too is if you're going to put these on Telegram too, you want to make sure that there's an allow list that's set up so only I can really chat with it. That way you can't just have anybody message these bots and attack you. Just keep that part in mind. But in terms of, let's see what I want to show over here. Okay. I'm going to show some other angles. Here, I'm going to work with it right now and say, maybe what do I want to do here? Let me think about this. You know what? I might just say, because Gabriel, you asked about the skills, tell me about all The cool skills we have, but do not leak any confidential information because I'm on a live stream right now. Let's just see what happens here. Okay. Oh, this is this is fun. Look at this. So I have this mission control thing. This this shows me my dashboard, right? So I can open my dashboard. And today's highlights: Oracle Reddit marketing agency can gain plus three submissions a month on free consultations. So. We rank really well for Reddit marketing agency, or we used to. And that page converts at about 7% to generate free consultations, right? From an SEO standpoint. And then Flash looks for news. So there's AI agent, PR drama, Warcraft, Pion notifications trending on Hacker News. I can choose to attack those trends if I want to trend jack. And then Deal. So this is actually like, I don't even know. I think this is a really old deal. For Single Grain. And I don't even think that's a deal that we decided to chase. But it called out that deal. So give me a quick summary, right? So, okay. Here's some cool skills. Gabriel, you literally asked the question. Decay Predictor. So for Oracle, which handles SEO, it catches pages losing traffic before they cliff. Cannibalization Detector finds out where your own pages compete against each other. Funnel Sequencer maps content clusters around your best converting pages. Revenue Weighted Score ranks every SEO opportunity by estimated dollar impact, not just traffic. Because I care about revenue, right? You all care about revenue. That's what I care about. So what else I would say here is Flash. Okay. What does it do? It'll scan Hacker News. It'll scan Reddit. It'll scan X. It'll actually scan YouTube as well. It'll scan certain podcasts too. It'll give me a competitive content monitor to see what's trending well. So for example, on YouTube, if I see packaging that's working really well, ideas that are working really well, and I can repackage it myself, I know I'm going to crush because right now I have this Claude code and OpenClaw topic. I have it gripped pretty well and I'm going to just milk it as much as I can to just drive as many views as possible and to teach. I do enjoy teaching and articulating in public. But the point of showing you all this is that, look, Cyborg, job spec framework, LinkedIn connection mining. So also it's not going to just help with recruiting. It also will look at connections and say, hey, we're looking to recruit these roles over here. You're at this company. Who else do you know that we should be talking to? That's fantastic. That's called referral recruiting. And then Alfred, follow-through tracker, tracks approved decisions, decision log. So these, by the way, these aren't even all the skills that I built. It's just calling out a couple of cool things that we've done. Daily candidate drops, personalized outreach template generator, infrastructure, anti-amnesia system. Someone asked about this a little earlier. How do you stop it from forgetting? Vector memory with semantic recall across all past conversations. Shared brain. All agents read from the same context layer in real time. Now, that being said, Alfred has access to everything, whereas the other ones not as much. The other bots can talk to Alfred. Alfred's at the top of the stack. I have these 29 automated cron jobs running daily. I actually cut it to about, I think, 15 now. Cron jobs are just chronological jobs that are happening. That's what it is. Built in 10 days, right? But keep in mind, I spend a lot of my time on this. The times that I'm not sitting in meetings and things like that, or helping with recruiting, or helping with other things around the business, AKA operating, I am just on this all day because I'm just trying to figure out how we can do all the chores that my team's doing right now and make this into something that my team can eventually pilot and use. Again, this is why we need to have good judgment people that are running these things because if you hire a junior person to run this, they're not going to know the right questions to ask. I'm not saying don't hire junior people, but I'm just saying for this, It might take you a little bit to bring people up to speed quickly. I think of this as in and out university where you have to train people up to get good really quickly. This is how OpenCLAW is being used. Literally, I can ask it any questions that I want. Watch this. I'll even dictate this out using SuperWhisperer. I think earlier this morning we talked about how you initially had rated Oracle, for example. What did you originally rate Oracle and why did you rate it so low? And then what did you propose to make it a 95+ out of 100? And what have you done so far? Tell me how far you are along with the process. Look, guys, as long as you're lazy and smart, you'll figure this out. The lazier you are, the more incentivized you're going to be to figure this out. I'll tell you, I'm a pretty lazy person, right? In terms of I just don't want to do these things. I want to think about What's the highest leverage thing I could be doing every single day? And by the way, look, it's typing to me. It is working. Now, a couple other things while this is working right now. So if you decide you— okay, we'll just sit on this for a moment. Okay. So Oracle started at a 35 out of 100. It was a data reporter, not a strategist. It could pull GSC, Google Search Console, Google Analytics numbers and filter for agency intent keywords, but that's table stakes. It had no awareness of pages losing traffic, no detection of pages cannibalizing each other. No understanding of which pages feed into converters. No revenue weighting. It ranks by traffic, not money. No competitive intelligence. No AEO and AI overview awareness, right? So then it was like, okay, well, to take this to the 80, we got to think harder with existing data. So decay prediction, cannibalization detection, what we talked about earlier, funnel sequencing, revenue weighted scoring. But I was like, guys, guys, guys, we're not here to score a B-, okay? So I went to that, I was like, what do we need to do to make it a 95? All right. So 95 is like, okay, Ahrefs API for competitive intelligence, live SERP monitoring for AI overview detection, AI citation tracking, cross-platform signals from Flash. Where are we at right now? Oracle's at 80. So this morning it was at a 35. Now it added this new scaffolding to 80. I didn't want to just do everything in one context with them, but now it's like at 95 because I had all 4 of the agents being upgraded at the same time and I was overwhelming the API. So I said, okay guys, Just spin these up in parallel and then work on these in phases. And then the API wasn't burning itself out anymore. And so look, check this out. It's already working. I didn't even know it was working right now. You guys are seeing me work on this in real time. Okay? And I'm going to start to move over to questions in a moment. So you start to drop your questions in the chat. But maybe next week, you guys tell me what you want. Do you— actually next week I'm traveling. I'm in Florida, but I still might do a livestream. For the next livestream, would you prefer me working through this right now or more explaining like I was in the beginning of this? Because I'm continuing to find a flow in this because the chances are I'm going to start to do these for 2 hours, 3 hours or so. I'm going to get a new camera setup. I'm on my Studio Display right now. I'm going to have a nice looking camera. I'm just going to go for 2, 3 hours or so. I might even have you guys call in. I might even pull you up as well onto stage. And that's what it is. Okay? So look, I didn't even know I was working on these things this morning. I'm going to work on it work with it later to just continue to refine this thing because it's at phase 1, which is at 80. I'm going to work with it. I'm going to feel good about 80 and then we're going to move to 95. But guys, in 10 days, these are workers that are smarter than me and I'm actually the limitation. They're smarter than me. I just need to know how to prompt it and work with it. That is that. I don't even know if I'm going to show Claude code. There's no reason to show Claude code today. I will say one more thing and I'm going to stop my share. I'm going to answer questions here. But when you want to plan things out, I open up Claude Code and it has a plan mode. You can use the planning to help you brainstorm. There's an ask user question tool where it can ask you questions like clarifying questions as you build things. I find that to generally be pretty helpful, so you can do that. But in a lot of cases right now with just me working with Alfred, I'm just going back and forth with it. He has a million context windows. There's a lot of things that I want to test. OpenClaw, you can get a Mac mini for it, set it up locally, make sure that the permissions, you only give it access to one vault. You don't want to let it do things autonomously for the most part. For Claude Code, what I do is I actually— the new copy that you see on the Single Grain website and on the ClickFlow website was very AI-assisted. When I use Claude Code, it helps me generate programmatic SEO pages. It's really good with copywriting, but I did actually ask the question to Claude Code. I was like, hey, let's think through this for a moment. Is it OpenClaw or Claude Code, or is it OpenClaw and Claude Code? And what limitations do you have if you're operating just individually? And it was like, individually, I can only do this as Claude Code. 5 out of 9 roles I can take over. Individually, OpenCLAW can only do like 4 out of 9 roles. Okay. And then explain the reasoning behind it and explain the approximate ROI if you're using these products. And then, but it's like when you combine OpenCLAW and Claude Code, it can do 9 out of 9 of the roles and it could, that's, you're basically replacing those jobs completely. Okay. Now what does that mean? Like, oh my God, you know, people are going to lose their jobs or whatever. No, no. Yeah. The tasks are going to get replaced, but the people who know how to use this stuff are not going to be replaced. Okay? A job is really just about the purpose. What purpose are you trying to deliver? And for us, the marketers that, these 4 Deploy Marketers for the Future, they're excellent marketing strategists. And now they have good taste. They're learning all the time and they're working with the clients to talk through how they can grow their business faster. Because our fundamental goal as a company, the mission, is to be the marketing partner that companies never outgrow. Okay, again, to be that marketing partner that companies never outgrow, which means we have to be skating to where the puck is going every single day. Okay? Okay. So I'm going to answer some questions and then I might even go a little longer. You guys can stay as long as you want. And I'll try to see how long this livestream can actually last. So let me— oh my God, there's a lot of questions here. Okay. Yesung? Yesung? I've seen you quite a bit, but I can't pronounce your name correctly. Yesung? Yesung? Okay. Anyway, Yesung. Yesung. I think it's Yesung. Okay. So why should I run ClaudeBot, which is OpenCLAW now, on a separate Mac mini rather than run it on my personal Mac Studio? Is it to navigate through the security issue? You can run it. No, you can. Okay. Yeah, yeah, yeah. So you don't want to run it on your personal one because if you get Prompt injection happens, then all of your stuff is compromised. Okay. So my separate Mac mini, okay. And you don't need to buy a new Mac mini. I just decided to do that. But I have it set to, it has its own Apple ID and it has its own Gmail. Okay. So that helps because, you know, again, you don't want to just YOLO your stuff and give it access to everything that you have. So that's what I would say. Now, Let's see. Gabriel, I think you guys are talking to each other in chat, which is great. Honcho, I'll answer you on X. Honcho from X. Okay. So for the brain model, what do you recommend? Is Ollama good? So the Llama model, the Meta model. I'm spending too much on the ChatGPT API. Okay. So that's a great question. So Alfred, which is my chief of staff, it does a lot of the thinking for me and quarterbacks a lot of stuff. Alfred is on Opus 4.6. And the reason you want that, because I was actually listening to the creator of OpenCLAW today, talk on a podcast and he was like, look, you want to have the smartest models running because especially for the chief of staff that's coordinating everything because the open source models start to think stupid things and do stupid things. That's what I would say. Even when I use Sonnet, for example, it's not that the horsepower isn't quite there. You really have 2 options as of this recording. You can use Opus 4.6, Or you can use Codex 5.3, which is OpenAI's. If you have OpenAI, I've heard really good things about it from the creator of OpenCLAW. It's really good for coding and thinking through things and taking a long time to do a task. If you want things to move very quickly and just spit out things as it goes, it's probably Claude Opus 4.6. But I'm actually going to try over this weekend, I think I'm going to try Codex 5/3 and have it be my leader of the squad. But even for my other squad members like Cyborg, Oracle, Flash, I actually have them on 4/6 now, Opus 4/6, because I could tell that there just wasn't— I wasn't getting the same horsepower in terms of thinking. And is that going to burn more tokens? Yes, but I'll just monitor my API costs every single day and see what happens there. Okay, so next question here. Oops, let me do this one over here. So from YouTube, we have, we have huge reaction. Um, so what do you think about virtual machine, Docker, or RunPod for running Claude bot? Seems to work well. So, um, virtual machine, I think so. I— You will see people talk about virtual machines on X about this. I'm just worried about security vulnerabilities, right? I'm trying to run this from a business standpoint. I'm trying to think about this from a revenue-generating standpoint. I don't think a lot of people are talking about that right now. That's just how I'm looking at it. It might be flawed. I'm not the most technical person in the world. I'm just reading and I'm just deciding what information I want to take an action on. That's what I would say. Okay. I'm going to move up here and then Okay. Hal has a good question here. Super curious to know what scripts should be installed after initial setup, security, mission control agents, et cetera. Here's what I'll say. I would use Groq and just put that question in there, Hal, because there's a lot of good posts on it already. Literally, that's all I did. I just bookmarked a bunch of stuff around security and initial setup. Mission Control. There's a good post on Mission Control if you find it. I forgot his name. Batu, I think, B-H-A-T-U. So he talks about his setup and how they talk to each other. But I've kind of— you all have your own use cases. And so I would say you want to make sure that you're talking to it. You're like, I'm dictating out what I want from my Mission Control. And in my case, I'm trying to make sure that I'm satisfied with it and then we can roll it out to clients. That's what I would say. Let's see. Vincent, I would say rewatch this video. Here, Vincent's talking about these live builds. Vincent, we had talked about a handful of these things earlier, but what I can say is the local hosting setup that cut AI costs by 80% versus cloud solutions. There are a handful of things on memory saving and just saving on tokens. And I would just say, look, watch, watch, check this out. I'm going to do this live with you right now in terms of saving tokens. We want to save cost even more, right? So watch. So let me, I'm going to show you one more screen share here. Okay. So let's just look at my usage today. Let's just look at this. So what's, what was my cost today? Watch, it's like crazy. Okay. Look at this. Hey, this is not bad. Guys, this morning on my walk to the gym, I upgraded all of my agent team to using Opus 4.6, which is the most expensive model. Okay? So this session right now is 21 million tokens. Okay, great. So today I spent $38 so far. All right? Last 30 days I spent $513. That's actually not that bad. I actually thought, well, Well, hold on. I've only been using this for 10 days. Okay. So my costs are going to be like $1,500 a month. Okay. But I think we've gotten the cost down. So let's say I want to cut my AI costs down further. So today, and maybe if you're looking at today, maybe I'm probably going to spend $50 to $60 or so. I want to cut that by half. What do you think we need to do to do that? Feel free to ask me any clarifying questions. So you see how I like to add the last part, like feel free to add any clarifying, ask any Clarifying questions. Now it's working. Okay. Now I am down with trade-offs. If it's like, oh, okay, well, you're going to sacrifice the smartness if you decide to go with Sonnet, but it's going to be cheaper. I'd rather stay with the smartness because I need these to be my strategic brains. That's what I would say. Okay. So we'll let it work for a moment and then maybe I'll answer another question as we go. And then Gabriel said, or Austin says, yes. So guys, just so you know, again, the challenge I showed you earlier, it's singlegrain.com/apply. So you can either copy that for your own needs, or if you're interested in working at Single Grain, go through that and it'll literally send an email to my team. I'll get a direct email as well on that. That's what I would say. Okay. Could I open source my prompts and skills? Yeah, probably. I mean, Maybe I'll make a GitHub repo just like some other people have that show all these things. By the way, look, it's really thinking about this one over here. It's— okay, here we go. Okay, here's where the money goes and how to cut it. Biggest cost drivers today, all 4 agents on Opus. I just did that this morning. I switched Oracle, Flash. Opus is 5 times more expensive. It's okay. I'll get a good ROI from it. If you're doing it just for personal stuff, probably will get too expensive. But for me, I think the risk-reward is worth it because we're also productizing this as well. Subagent spawns on Opus. The phase 1 builds this morning were heavy. High frequency crons, main session at 200K, every message I process costs more. Here's what I do. Okay, great. It explains its thinking. Switch Oracle Flash. I moved back to Sonnet. Save 60% of token costs. I'm just gonna see how it goes today because I just made this change. Reduce compaction cron from every 30 minutes to every 2 hours. It's running 48 times a day. Cutting 12 to 12 saves 75% on that job. Um, okay. Reduce hourly memory summarizer to every 2 hours during quiet hours. Okay. Um, so you can do number 3 fully. As for the compaction cron, you can run it every 1 hour. Okay. So I'll do that. So I'm not going— I'm not just saying, oh, okay. I'll just take everything at face value and go with it. You got to have your own opinion. That's what I would say. Okay. So I'm literally, I'm just telling it to work now and it's going to work. It'll probably get like By the way, sometimes I'll say, oh, I'll get it done tonight. No, no, no, no, no. You're just going to get it done. I'm just like, hey, can you just do it right now? There's that. Now, let's see. If I install Mac Mini, this is from YouTube. If I install it on a Mac Mini and connect that computer to my Wi-Fi network, is there a security risk that can somehow access other computers on the same network? Maybe. You're going to have to— I think that's a good question to go ask AI or ask Alfred or not Alfred, whatever you decide to name yours. But by the way, this stuff can literally teach you how to use it. Well, I will also say sometimes I'll break it. One time I had it broken, I was trying to change the model or whatever and it broke on me. I literally, I just opened up Claude and I just kept dragging screenshot him like, this is happening right now. This is happening right now. Okay, boom. We got it fixed. It took an hour and a half to fix it, but it got fixed. I would say if you really want to pick this stuff up, a lot of resilience is needed. Look, I'm just letting it work now. It is what it is. Um, and so there's that. And then automating LinkedIn would be an amazing skill. So what I would say is really the sky's the limit. Anything that you want to do, it can do because it has computer use. Like when it's working on my other computer over here, um, it can pull up browsers, it can open, open up my LinkedIn, I can start working. So you can have it do that. Um, and then Common Apostle asks, what is your take on using Manus with GitHub? I haven't done that myself, but I love Manus. Manus is amazing for making like Quick graphics, amazing for outreach, for follow-ups. It's a really great general agent. Oh, look, see, look, it did the work. How long did it take? 1 minute, maybe 2 minutes to do the work. Okay, look, compaction cron every hour. How much money are we going to save from this? Okay, let's have it run a calculation for fun. Maybe I'll answer maybe 1 or 2 more questions over here. Why not? Okay. So this is a great question over here. So Everything Altcoin asks, why not have a hybrid setup like Kimmy as the brain that can tap into Opus? And so I tried that type of setup where I was trying to get cute with it. It's like, oh yeah, use Opus for this and Sonnet for this. And then the whole thing just broke. And I'm just like, screw that. Don't want to deal with it. I'd rather just have it working stably because the reality is if we're selling this to customers, they need stability. They need security. And they needed results. And that's what we're doing. And so check this out. Look at this. Cutting from every 30 minutes, every 20 minutes. Okay. 24 fewer runs. Okay. So we're going to save— guys, we're making progress. We're going to save $10 a month. Save $120 a year, right? So did I pick the highest leverage thing? No, I didn't. But I need to see how my test is going to go. And so sometimes it's kind of lacking the context. Yes, I did make that change to Opus for everything in the morning, but I want to see How much value I'm getting. Because here's the thing, if we switch to Sonnet and I have to keep asking the same questions over and over and over, I still end up spending maybe the same amount of money, but I end up spending a lot more time. And that's something that you don't have an infinite amount of. So that being said, I'm going to answer one more question, but you guys drop it in the chat what you would like to see, more builds next week, as you guys kind of saw me building, Or do you want me to see more explanations, more Q&A, or like a mix? Did you guys like the mix today? If that's what you like, we're good with that. We can continue that. Let me answer one more question over here. You working on it helps. We'd love to see more on Mission Control dashboard. Finally caught you live. Let's see. Team members. Okay. This one's good. Elvis's question. I think it's Elvis. Team members should definitely be training to use AI, whether agents or chat assistants. It just supercharges productivity and it helps work more efficiently, but it doesn't mean a person could sit back and relax. You need to still be on top of things. So what I would say is on this, we want to make sure that you want to make sure that sure, train people on AI, but if you have people on the team right now that do not have the good judgment that you need, you're probably going to have to make a move sooner than later because The reality is the text messages I'm getting behind the scenes right now is like, I was playing tennis with a friend last week and he's like, yeah, we just cut 50% of our staff. Okay? Someone just told me, he texted me earlier this week. He's like, we're cutting 600 people. That's a lot of people, right? And so I don't want anybody here to be a casualty if you're working at a company. So you got to, if you've got good judgment, you got to learn this stuff, right? The other thing is if you're an owner of a company or a leader of a company right now, you can make these decisions. You really need to shore up, batten down the hatches with your hiring process right now. You cannot let any bad judgment through the door, okay? Because that's going to cripple your company, right? Because the train has already left the station. It's way gone already, okay? So if someone's just starting right now and they have good judgment and you try to bring them on your train, they're just going to derail you, okay? You can't do that. And I'm sure they can do their best work somewhere else, But you need to decide like the Beat AI challenge that I showed you earlier, the working assessments, right? All these things that we do, we do a mock as well, which is we define the mission, the outcomes that we want from a role and the competencies. These are all things that we define. And then you have a hiring process that nails it all down. Now that's much easier said than done, but that's what I believe. Very good hiring and AI fluency, that's going to take you to the promised land. So what's going to happen? Will I go live next week? Maybe I will. I hope you all enjoy this. Share this with your friends. This is also living on YouTube as well. I think it also lives on LinkedIn too. I hope you enjoy this and I hope you all have a great weekend and I will catch you all later.