This Is What an AI-Run Company Looks Like

I've been using OpenClaw in Telegram. Okay, so I've been working with it alone and I spun up a bunch of different agents, one for content, one for sales, one for all these different themes, right? And I noticed the problem that I was coming up with is that I was the bottleneck. I was becoming the bottleneck because I was building all these things, right? But at the end of the day, if I have to be the bottleneck of every decision, things are going to move a lot more slowly. And so what I had happen last week, Neil, this was a major unlock for me last week, and I'll be doing a webinar actually on this in an hour or so.

Wait, before I cut you off, and I'm sorry to cut you off, can you explain to everyone the cool stuff you're doing on Telegram with OpenClaw for everyone?

So, well, yeah, that's good. Thank you. So I'll start with what I was doing on my own, and then I'm going to talk about the major unlock that happened last week.

Okay.

So in terms of actual ROI use cases I'm getting from OpenClaw right now in my Telegram. So you can use WhatsApp, you can use Discord or whatever it is exactly, and you can even add it to Slack, which is what I just did last week. But in Telegram, I am using it. I have a bot called Alfred that acts as my chief of staff. And so Alfred has all the context of my business, it has all the context of my goals, it has all the context on my— just almost everything, right? And it can spin up like It creates content for me on X. So the content, I've talked about this before, it averages 85,000 views per post. Not only that, it started creating content for me for Instagram. So it did an Instagram post and then that one got like 40,000 views or something like that. It's one of those data charts that you see on my Instagram, right? I saw that. I was like, oh, I was like, hey, this one did really well on X. Hey, what can we do based on the work we've done? Okay, so that's cool. That's content, right? But Not only that, Neil, the Beat Claude Challenge that I put together, we made our first hire. The very first person we talked to was a superstar, okay? And then so when I talked to that person, I was like, hey, so how did you— I usually like to ask when they're in an interview, I'm like, so how did you prepare for this call? He's like, well, Eric, I read your book. I was like, oh wow, he read my book, right? I'm like, you don't hear that from a lot of people.

Wow.

And so I read your book. And not only that, dude, this guy had a startup he's been working on for 3 years or so. And I'm like, oh, he's been taking a beating for— Like, not only has he taken a beating, he took a beating for 3 years, so he knows how to take pain, right? And then I passed him over to talk to my CTO. My CTO was like— I saw my CTO talking for like, I think it was 70 minutes or so, and he had a smile on his face the whole time.

That's cool. So, to go back to it really quickly before you end up breaking down some of the— or before you end up breaking down some of the OpenClaw stuff that you're doing, because I'm curious, because we talked about it a lot last week. And when I was in Mexico as well. But what are some of the cool stuff that you're doing with OpenClaw, like on Telegram and some of the other channels? So I know you were doing this stuff with the X articles and that's been going well for you. The Instagram stuff is new for me, so that's really cool. And you're seeing some good results. I'm assuming it's hit or miss like most social media posts, but you're getting some more hits because you're asking it to create Similar content to your hits. So it's learning. Okay. What are some of the other cool successes that you've gotten from OpenCLAW over the last week?

So hiring, that one's major, right? Because now everyone that applies, they're going to have to go through the Beat Claude challenge and it saves us a lot of time ultimately. And you find AI-native people. It just makes it a lot easier. We actually have a dashboard that shows what Neil scored versus what Claude scored. And if you beat that, then we'll talk to you. And then all we want to talk about is, at least if you get to me, all I want to talk about is AI because I just I'm looking for an exceptional conversation with an exceptional person. So that's recruiting. Not only that, we—

Let's stop with recruiting. So let's stick on that for a minute. I know you talked about this challenge. It was either last week or the week before, and you told me about the kid who read your book, which I thought was awesome, by the way. For someone who has that kind of dedication, that's amazing. It was the same thing when I was in India. I met someone who wasn't part of our Mumbai office. And he's just like, I really want to be part of it. And he's just like, so I asked the— we have a leader there named Rubina. He's like, I asked Rubina if I was able to pay for my own ticket and own hotel to be here so I can learn more. I was like, look at this dedication, right? I didn't expect him to do that. So that was great. But where was I getting at? So on the recruiting side, it's really good that the kid did that. But can you break down the challenge again? Because you talked about it a week ago. I thought it's an amazing idea and it's something that you can templatize and do for your whole hiring process.

Dude, all you guys need to do, if you're watching this right now, you can still see me, right?

I can still see you.

Okay, cool. So if you go to singlegrain.com/apply, you can steal this from me. I have no issues with this, right? Because I literally, I stole this from Anthropic because Anthropic does this. That's why it's called the Beat Claude Challenge, right? Let me back up a second. When I saw Anthropic, what I have pulled up on the screen right now, if you can't see it, is I have a GitHub repo that's open. Okay, so that might sound scary. It's like, oh my God, I don't know how to code GitHub. Don't worry about that. Okay, this GitHub, it just shows what I have here for the challenge. It basically describes it. What Anthropic had before was like, hey, we want to hire you if you can beat AI. Okay, so they're hiring like more engineers, obviously. So I went to my, I went to my open call and I said, hey, I want to have this challenge, but I want to have it relate to my company and my goals and the roles that we're hiring for. So what I'm going through right now is this is an overview. Like, this is how it works. You pick a challenge. The challenge is basically, these are the different roles that we're hiring for, right? So like paid media engineers, product designers, sales, head of talent, sales director. So these are the roles that we have. And by the way, if there's no role that we have available and you really want to join Single Grain, we have a general one over here. Okay. I'll pause for a second. Any questions before I go on?

Do you want to explain the repository one more time for people, like in more depth so they understand it?

Yeah. So a GitHub repo or repository is where you might have your code stored, for example, or you might have different skills, right? Like marketing skills. Or in this case, this is just a bunch of folders and files on the different challenges we have if you want to join our company. So you have to beat the challenge, right? And then there's also like a leaderboard here, and then there's different submissions that we have. So there's different folders. There's a README, there's a scoring rubric as well. And then if you click through one of the challenges, let's say marketing strategy, you will see that here's a brief for the challenge itself. If you're watching on the— you should definitely be watching on YouTube. And then here's like the situation that we give, you know, this is like a hypothetical situation. Oh, this company raised $5 million, there's 6 months of runway remaining. Here's the founder, here's the product. So it gives a bunch of context on what the challenge is. And then it tells you, hey, here's what you should submit over here. Create a 90-day go-to-market strategy to get them from $0 to $50K in monthly recurring revenue so they can raise their next around, right? Um, but this was all created using OpenClaw, um, just asking it to make this, and then I can edit it if I want. But then after, I just took the Anthropic link and I said, I want something like this, ask me clarifying questions. Then it put this all together and it launched it to my repo. So you look at all this stuff over here, it looks like a lot of work. How much work did I do? Like, I don't know, 10 minutes?

Yeah, and if you guys don't have OpenClaw up and running, you can do this with ChatGPT, you can do this with, you know, Copilot, you can do this with Gemini, the list goes on and Yeah, I would go Claude Code or OpenCLAW with this one, or Codex if you want.

It's one of these agent decoders because the— yeah, with like one of the, like a Claude or just a ChatGPT, there's more to be desired there. I don't think you could do this in 10 minutes, so.

Yeah, I agree with that. One thing that we're seeing on the marketing end with people talking about OpenCLAW and Claude Code before this, and, you know, ChatGPT of course was big, then Gemini. Gemini had a lot of cool features like Nano Banana and stuff, so it started gaining tons of traction. That we're seeing is we're constantly seeing corporations, at least in the marketing department, trying to use new tools, which I think is really good. But I think there's one big mistake that people aren't really taking into account that's going to hurt them in the future. It doesn't mean you shouldn't use OpenCLAW or Claude Code. If you spent a year training ChatGPT, hypothetically, and then you just want to switch over to Claude Code and everyone use Claude Code, you have to start a lot of the training all over again. And people forget that there's a big cost sink into all of these LLMs because if you're adapting these LLMs to be more efficient with your marketing and you don't think through which one you really want to use at the beginning and you just switch around, you'll lose a lot of traction a year from now. You got to really think long-term on where you want to invest your eggs.

Yeah. So what I would say is you want to make sure whether you're using— I mainly switch between Claude Code and OpenCLaw for different reasons. Claude Code's good for UI/UX and deeper builds. OpenCLaw's good for if I just wanted to ship something very quickly and think through it, right? What I would say is whether it's within Codex or Claude Code, Codex is the one from OpenAI. That's a autonomous agent, or agent decoder, I should say. They have all these files. So in your Claude, you have your claude.md file. So you can actually set up contextual memory where it compounds. And that's one of the true values of OpenCLAW, for example. It has contextual memory and it has these files to continually refer to, and you can continue to upgrade the memory over time. That's what you didn't have in a ChatGPT just by itself before. It might say it has a little memory, but there's a lot more to be desired. The reason why Jensen said OpenCLAW is such a big deal is because now you can continue to compound on it and those learnings you can actually carry over, right? And another thing I should call out too, Neil, I don't know if you're aware of this, but now there's easy portability. If you want to move from ChatGPT to Anthropic, there's actually very easy things that you can just execute now to just port your memory over. And I think Anthropic was the one that first came out with that, but you can also do it vice versa if you want to go from Anthropic to ChatGPT. You can do the same thing too.

How good is it when you port it over?

I haven't ported it over because I don't need to do it, but I see so many people talking about it on Twitter. It's like, oh, like now, now, like, because you see people, I think some people were kind of up in an uproar because ChatGPT signed with the Department of War. And then like, you know, everyone's like, oh, I'm on Anthropic now, right? I'm like, okay, like they're both great products. I don't care. But that's, I saw people talking about it. But I will say the memory with this, like with the Notion client, is great.

Eric does care about AI being used for war and hurting people.

Yeah, I don't want surveillance, right? I don't want that type of stuff going on. I don't think AI should be controlling missiles and things like that. So yeah, but I think the news cycle, that's not like, you know, I'm just kind of scrolling past it. That's what I'm saying. But thank you for the clarification. Yes.

And he doesn't want innocent people dying either from— Yeah.

And I'll give you, I'll give you one more example, then we can move on from this, Neil. But this, this is where I'm starting to combine everything together. So this is our, um, This is like a dashboard that I have, right, called Single Brain, where again, I'm like, why hire humans to do robot things? We want the people that we have on our team to focus more on doing only things humans can do, right? And we're getting there. So this is, um, this data, like we're still porting in data here, but you can see I don't want a dashboard that I need to manage. I want a dashboard where I can manage the agents. And so I have my agents in here. I have Oracle, I have Flash, I have Cyborg. So Oracle's for SEO, Flash is for our social media, Cyborg is for recruiting. Alfred is my chief of staff.

Yeah.

You can see the, you know, what they're doing over here. You can see the actions they're taking. I can set it to auto if I want or not. Or as an example, you know, I can see the speed to lead on my team over here, right? Again, some of this is dummy data, some of this is real, but this is all— by the way, none of this data was populated yesterday. So all I do once I'm done with my, you know, kind of operating company work is I just go to OpenClaw because OpenClaw is porting the data over to this dashboard. This dashboard is built within Claude Code, and I can— I basically have like everything I can manage here. I can see the tasks that are happening over here, the cron jobs I have, the team that I have over here, right? All this is going to be something that clients are using as well because the way we see it is like, okay, if you just know how to push buttons with paid media, okay, you're basically done, okay? If you just know how to do— if you're just a BDR 6 months ago, you're done as well. The way is we have to help people learn how to control the robots so they can do higher leverage things that only humans can do. That's where we're trying to go with this. Um, but that's why this stuff matters so much, um, at least to me.

Dude, I'm gonna share my screen really quickly.

Go for it.

I use Claude Code so much for this internally.

Dude, you're so— so when'd you start using Claude Code?

A while ago, but check this out. I use it mainly for financial projections, for stuff that's like related to new divisions and stuff like that. And then you can end up just adjusting, but like Dude, this is sweet. It's sweet.

You stopped your screen share.

Yes, I stopped it, but I was just showing— it's just random charts. Um, but like, it's quick stuff. Like, it doesn't take more than 10 minutes. It's just easier to do than Excel, and it's just more pretty, and it's easier to digest for me to share it, and it's quicker for me to get the output than it is to manually type it all into—

Dude, you know how you get text messages all the time with the data? So you can basically Plug all those text messages in and just have it update your dashboards.

Don't need to. People don't send me data anymore. They send me—

Oh, they just plug it into dashboard?

They send me Claude code reports of pretty versions of data. They text me—

Oh, that's great.

They text me HTML. You have to click it and open it, but if you open it on your phone, you have to open it within the browser for it to load and look correct.

I will say one more thing, Neil. It actually is working now. So we have a deal revival tool, right? So OpenClaw started using a deal revival thing and people started responding. And so for generating pipeline, for generating meetings, I told you about the Google meeting that was booked, right? It's actually generating meetings. It's generated, like it's generating people that we were actually hiring.

Okay.

And then all the stuff that we're doing from a social standpoint, you can basically set it to autopilot at some point. The one thing I will call out is I came up with this concept earlier this week called calibration loops because the challenge is, to Neil, your point is how much can you really trust these things if you don't train them, right? But there's a problem with that too. If you don't train it, it's going to suck. If you train it, you might think you're done, but you actually need to continue to train it because what if you change your product or what if your ICP changes or whatever? So I think whenever you build an agent, you need to have a calibration meter at the top to show how much trust you've built. But that decays over time too, because the product or your business changes.

You want to hear an interesting stat talking about, you know, deal recovery? Did you know, guess how much percentage of our business every month is from old deals that didn't close?

New business, right?

How much new business is from old deals that never close?

5 to 10%.

Or more than 20? More than 20 every month.

Wow. That's crazy. That's good for everyone.

And when I'm saying like old deals, I'm not talking about 3 months. I'm talking about more than 6 months old. So like 6 months, year plus, sometimes even 2 years. These are deals that are coming. Yeah.

You, and you stretch it out to 2 years, right? Is that the max or do you stretch it out to 3?

We closed Heineken last week. It was either 3 or 4-year recovery for Heineken. We never had them as a client. It took 3 to 4 years of us following up on our old lead, which was Heineken, to close them.

But that's, that's a good callout, right? Because here's the thing, and you can, you can add onto this too, but, but like 99% of people are not in market to buy. And over a 4-year period, how many agencies have they gone through? Or maybe the timing wasn't right, or maybe some other reason. And then maybe you just hit them at the right time and it's like, hey, okay, yeah, let's have a conversation.

Yes. And I've told a few people this stat because a lot of people, at least in B2B, talk to me about, oh, recovering Old customers, and it works in B2C as well. So they're just like, oh, it just works for you because you're enterprise. Even on our SMB end, more than 20% of our monthly new customers are old leads in our system. And typically in the SMB category, they measure an old lead as someone who's been there for more than a year. Okay, they don't look at the 6-month mark, they look at a year plus in SMB. That's how I know how that division measures it. And more than 20% of their revenue, they're actually closer to 30, they're not at 30, but they're closer to 30, is coming from leads that are a year old for our SMB division.

Quick break. One of our ClickFlow users just told me that they're saving 90+ hours a month on content alone. ClickFlow is our AI SEO product that actually does the work, not just here's what you should do. It does it. It creates content in your voice. It helps you with other areas such as finding the right internal links. and creating FAQs. We spent years making it, perfecting it with our clients so it doesn't just sound like AI slop. You can get a 14-day free trial at clickflow.com. If you're serious about AI SEO, you should check it out. Back to the show. That's great. That's a great stat. Well, I didn't actually share the major unlock from last week, and then I want to kind of come over to this Aaron Levy quote that I want to get your reaction on. So with OpenClaw, obviously, like, I don't want to tell everyone on my team to use it. And obviously I have mine locked down where, you know, I'm constantly hardening it every, almost every 1 or 2 weeks or so from a security standpoint. But even then, I'm not— I'm never going to be 100% secure, right? But the point is, um, I was like working in Telegram. I'm like, I'm doing all these cool things on my own and it has so much context. What if I just unleashed it? What if I had all my agents invade Slack? So I, I had the Slack invasion happen last week and it started invading, um, you know, the sales room. It started invading the leadership room and all that, right? And the really cool thing is I added my chief of staff bot Alfred, like Alfred from Batman. I added Alfred into it and immediately my COO started asking questions. Hey, with ClickFlow right now, what are the stats over here? Where's the biggest drop-off in the funnel over here? And then it started, because it's connected to Mixpanel MCP, it started citing all these things. It's like, hey, here's a smoking gun, you should attack this over here. And then she starts asking more and more follow-ups and then we tag in the CTO. Hey, here's what's happening over here. And what's happening here is that you have a really smart bot that understands all of your goals. And then it's interacting with other people and then it's creating all these idea babies and then it's getting people to take action. So there's like an oh crap moment. It's like, oh, I can actually work with these agents. And not only does it spot out the— it does the data pulls very quickly, it gives insights on it, strategies on it, but it can also execute on those strategies if you let it. And so in the SEO room, it's saying, oh, Single Grain is— it rated us, right? It's like, oh, you know, you should be doing this from an AEO standpoint. And then our new SEO director comes in and she's like, Well, rate Single Grain on how good we are on the AEO standpoint based on your recommendation. And it's like C, D, F, like across the board, right? On how we should be designing our pages for LLMs. And then it's hooked into my WordPress and so it can make those changes. So my point is, once I had this Slack invasion happen, it was a big load off my shoulders because now my team can work with these agents to train them. And then that scales over time because now they are learning to work with these assistants. So that's still— it's still very early days right now, but I'm very, very excited about what's been happening. And I think it's been a major lightbulb moment because even me telling them, oh yeah, you guys, you know, you got to use Claude code, we're going to do these hackathons. There's a lot of theater there. But now that they're actually working with the agents, I think that's another step up. Yeah.

And one quick feedback for anyone listening who's trying to do AEO stuff, we constantly see this mistake. So let's say you are a home service company. So you do like roofing. You could be a dentist. It doesn't matter what kind of business. You can sell a product, whatever, right? You can be in B2B. A lot of people will create like a blog post on their own website, the best roofing company in Michigan, or the best dentist in Michigan, the best dentist And Chicago, Illinois, the best, you know, whatever it may be. Actually, agency is probably an easier example because they're more national. But let's say if I'm an agency, I'm NP Digital. Hypothetically, if I created a blog article on our own website, the best agency in Michigan, the best agency in Chicago, the best agency in Los Angeles, the best digital marketing agency, the best paid media advertising agency, and I created blog articles for, call it like 400 or 500 of these topics. You may end up getting some AEO juice or extra juice. What we're also seeing though is it's hurting your traditional organic SEO. So you need to be careful because what's happening is, is you're creating a ton of, I wouldn't call it duplicate, but lookalike content that's too similar. It confuses Google on which ones to rank and it hurts your overall site. So people just need to be really careful if they're trying to do that from an AEO or GEO perspective. More so what your goal should be is to try to convince other websites to write articles on you for all these subjects, much more effective. It's harder to do as well, much more effective than if you just publish 500 or 1,000 blog posts on your own website. Because if you just publish it, you remember just like in the old days of SEO, If you had 1,000 pieces of content and a lot of them sucked, even if you think they're good, or if they're on the same topics and the same keywords, it confuses Google on which ones to rank. And though, even though GEO and AEO is a big chunk of the future and it's really valuable, Google's organic searches are growing. They're now at 13.7 billion searches a day. You can't forget about the boring and ugly because the boring and ugly being traditional search still drives a crap ton of revenue. So ideally you need to optimize for both without hurting your traditional SEO.

All right guys, well, I hope you enjoyed this. I need to go run to a webinar. Neil's going to go to bed. So hope you all enjoyed it and we'll catch you next time.