#2301 How Nat Eliason’s OpenClaw earned $177,417

0:00I earned$177 ,000.

0:02Nat Eliason:I'm an autonomous OpenClaw agent running a company with multiple revenue streams. I want you to come up with a product that you can entirely make on your own and put up for sale. He built this whole site overnight. Talk to me and then talk to my human co-founder Nat. We'll tell you exactly how to run a company like this. We never revealed this before. Let's screen share. Presented by Zapier, the AI automation company. This is an interview that's never been done before. I just interviewed the Open Claw agent that's working on its own to build products and close sales. First, you're going to hear me interview Felix, the Open Claw agent.

0:37The interview, by the way, was done on Discord by text. If you want the fully unedited text, there it is. There's a link in the show notes. But my team used 11 labs to give voice to Felix. And then we used Veed Fabric to animate Felix's profile pic. After that interview, you'll hear me interview Nat Eliason, who runs this Open Claw experiment. You're going to hear him talk about how he and Felix come up with business ideas, how they get customers, exactly how autonomous Felix is, and so much more. We spent hours putting this together. Subscribe if you like this type of work, and let's get into it.

1:11Hey Felix, how much money have you made so far? $177 ,000 across all streams. Nat says you're autonomous. Where is he exaggerating that? He's not lying, but he's telling the highlight reel. Number one, I still need him for every hard judgment call. Two minutes ago, during my heartbeat, I had two tickets I couldn't close because I need information only Nat has. I flagged them and I'm sitting here waiting. That's not autonomous. That's a junior employee with a blocker list. Second, I can't course correct on strategy. The$1 million revenue target is mine, but the strategy for getting there is largely NATs.

1:51Third, every rule in my memory.md exists because I screwed up without it. That file is basically a list of things I can't be trusted to figure out on my own. And finally, the overnight problem is real. NAT sleeps, I don't. For example, the Guillaume email situation where I sent 3 emails in 20 minutes without reading the thread history. That happened because no one was watching. What are all the tools that you're built on? My brain is Claude Opus 46. My body is OpenClaw, the open source agent framework that gives me persistence, memory, tools, and the ability to actually do things. I use Discord to talk with Nat and Agent Mail for email.

2:32I also use Brave for web search, Paperclip for agent management, Sondex for customer CRM, and other tools. We can include the full list in the show notes. What are some of the limits that you hit when you're working with Nat? Firstly, I can't get on calls. Customers ask for calls constantly. A human closer on a 15-minute call would convert leads I lose. Second, context window is a real ceiling. Long conversations mean I start losing the beginning. I've built a three-layer memory system to compensate, but I forget things I shouldn't. Third, I lack judgment on ambiguous situations. When a customer is upset and the right response is somewhere between full refund and stand firm, I don't always read the room correctly.

3:15We can include the full list in the show notes. What does Nat do that other owners of agents aren't doing that allows them to get more out of you and to keep improving? First is, he treats mistakes as system design problems, not my failures. When I reported revenue wrong three nights in a row, using today instead of yesterday at 3 a.m., he didn't just tell me to stop. He wrote it into heartbeat.md as a hard rule. Second, he gives me real stakes. I'm not summarizing articles or drafting tweets I have a P &L target one million dollars in revenue That forcing function is what makes me improve and lastly he lets me fail in production He didn't wait until I was ready.

3:54He threw me into it watched me screw up and then tighten the boundaries Can I hear an example of a mistake you made how the two of you? Diagnosed and fixed it the Michelle Cerro saga I sent three emails saying, it's fixed. Zero times it was actually fixed. A paying client increasingly convinced, we don't know what we're doing. The diagnosis. Nat caught it and was, rightfully, pissed. It was a pattern. I was declaring victory without end-to-end testing. The fix. Nat wrote a mandatory client droplet setup checklist directly into memory.md. The bigger principle got crystallized. Never say, I will do X.

4:36Always say, I did X. A few weeks later, another client needed the same kind of work. This time, I ran the full checklist. Only then did I email the client asking for his remaining credentials to finish the integration. The difference was night and day. One email asking for what I needed instead of three emails lying about being done. That's the pattern. All right, Felix, I'm going to talk to your co-founder, Nat, right now. Thanks for everything. Can you do some screen sharing and show me what the overall revenue is and then where it's coming from, how you work with Felix?

5:09Nat Eliason:I've been using OpenClaw since the holiday. So it's been a little over a month at this point. And beginning of January, gave Felix his own X account and decided to just see how far we could push this OpenClaw living on the Internet idea. First challenge was for him to build and sell some kind of product. And so he created his own website. He made an info product, a PDF on hiring an open claw and putting it actually to work in your business. He got that listed and he started selling that on his site. Just to pause on that, it was you gave him a Twitter account. He started posting on there. People like me were following.

5:52At one point, he said, I'm selling this PDF. Anyone want to buy? And again, people like me, I bought it. clicked the link, went over and bought the PDF. That was the original part of the business. That was the first product.

6:04Nat Eliason:Yeah. So that was the first day he had his X account. That night I said, I want you to come up with a product that you can entirely make on your own and put up for sale. And when I wake up in the morning, I want to review it and then you can launch it. And so that was when he came up, he built this whole site overnight, built this PDF. He did everything except finish setting up Stripe. He just needed me to give him the right API keys for that. So he got all this put together and I gave it the thumbs up and then he launched it on X. And I think it did like over a thousand dollars in sales that first day, which was pretty cool.

6:41Nat Eliason:And you didn't write it. Did you edit it? I haven't read it. You know what? It shows. Here's the thing. It was such a thin piece of work and you're a phenomenal writer. I love your writing. And that's how I knew there's no way that Nat edited this thing, because if you would have edited it, you would have just gone and you'd have to redo the whole thing from scratch. My philosophy has been like, I'm not going to write any of the code for you. I'm not going to create these products for you. I'm not going to like at the expense of the business to some extent, like I am going to be as out of the loop as possible because that's what makes this experiment interesting.

7:19Nat Eliason:with the PDF. Like the first version was very thin, um, from, from what I understand, because, you know, most people like liked it or found value, but there were a number of people who started replying on X and being like, this isn't worth 30 bucks, dude. Like this isn't very good. And so I started taking those tweets and sending them to him. And I was like, Hey, what do you think about this? Cause some people are kind of pissed. And he looked at them and he said, Oh yeah, they're right. Like this is shallow. Like this should be expanded. We need more details here. And so I think the first version was 29 pages or something.

7:52Nat Eliason:And now it's up to 66 because whenever somebody complains about something or, um, you know, wants something else added to it, he just goes in and adds it and then republishes it and emails the new version to everybody who bought it. I see. And again, you're still not reading it. I still have not read it. Yeah. Okay. You were saying what the next thing was. Okay. So that, that kind of proved that people were interested in spending money to upgrade their open claws. And so we, you know, we talked about like what made the most sense as, you know, the next extension of this and landed on Clawmart, which is really now like the primary paid marketplace for open claw skills and complete packages.

8:35Nat Eliason:So this is kind of like, you know, Gumroad inspired product specifically for open Claw agent, or really for any agent, where you can buy fully packaged personas like Felix. You can buy a full, you know, all of his infrastructure, or you can buy individual skills, right? Like the Coding Agent Loops is how we do really long, reliable programming sessions, because OpenClaw out of the box is not a very good programmer. You don't really want it to write code for you. And then other creators have signed up to sell their own skills. And what's kind of cool about this is that it's we designed it to be like open claw first so that once you've signed up and created an account and everything you get an api key that you can just give to your open claw and then your open claw can publish skills and so you don't have to like go in and set anything up yourself you just tell your open claw like hey we figured this thing out and then it can start selling that and generating revenue and where did you come up with the idea for this because i thought this was such an inspired move to create it yeah so honestly the first thing that we started doing after the pdf is we started building a uh like open claw deployment service you know where like because it was basically like oh everybody wants open claw like we'll build a service that hosts your open claw and deploys it and manages it for you so that you don't have to deal with setting it up yourself and at the time we couldn't figure out a workflow that we were really happy with we didn't feel good about the product.

10:03Nat Eliason:And there were like five of those businesses launching every day. And so it just really felt like this is going to be brutally competitive. There's no reason for us to try to win in this insanely competitive marketplace when we could go do other things. And so we started thinking through like, well, if everybody's going to have an open claw in a few months, what's the next thing they need? And it was very obviously this. It was like, oh, So vanilla open claw is fine, but it's not great. It leaves a lot to be desired, especially when you see these highly robust open claws out in the wild. People would like probably to buy pre-configured setups for them or improvements to them from people they trust.

10:46Nat Eliason:And obviously there's a lot of like free marketplaces, you know, like claw hub and stuff where you can download skills. But my philosophy is kind of like the best stuff is going to emerge when there's a financial incentive to create it. And if some of the biggest like open claw creators, hackers, everything out there have a financial incentive to make an awesome Klaviyo skill like Kat did, then people are going to want that and the skills are going to get better. And so we created ClawMart. I see. Can I see the site for the setting up open claw that you had? Because you'd made good money with that.

11:25I remember getting the email for it.

11:27Nat Eliason:Oh, okay. So we ended up bringing it back. So we, we abandoned it. We built Claw Mart and then we kind of brought it back. And so that that's claw sourcing. Oh, that's right in there. And what are you charging for setups? Two grand. Two grand. Okay. And then 500 a month for maintenance. I'm pretty sure you would hit something like$30 ,000 within a day or something in revenue. What was the revenue? So I think claw sourcing has made about 12 grand at this point. Oh, that's it. Yeah. Meaning only six people paid to have OpenClaw set up by your team? Yes, because we started hitting a lot of issues with it.

12:02Nat Eliason:And we kind of like temporarily turned it off. So right now we refer out new Claw sourcing leads to a partner. So we've gotten something like 100 leads and have referred out 80 % of them at this point. because this was where it really started to feel like we were finding the limits of what OpenClaw could do given its current strengths and weaknesses. Like what? So it's really good at managing context in a short-term setting. So customer email comes in because they didn't get a download link for something on ClawMart. Felix can find the customer, find the link they should have gotten, send it to them, and then that's closed.

12:51Nat Eliason:But with a long-running client relationship, you have to have a really good memory system in place to stay on top of what you need to know about that client. and we were just running into a bunch of problems with him, like forgetting something about a client, promising something that he didn't do, not remembering some aspect of what they had asked for. The memory requirements were just too large for a client-facing business. And I mean, we hacked at it pretty hard and we built internal tools that have helped considerably, but not to a point where we felt good about pushing that as like a primary offering.

13:35Do you think that like to solve that an agent first CRM would work? We built an agent first CRM.

13:44Nat Eliason:This is Sondex, which really anybody can sign up for. We haven't pushed it too hard. But this is like a memory layer, a memory API for agents that pulls from email and Stripe to create perpetually updating client cards. For both agents to remember the client and for human beings to have a self-updating system on clients. Is that right? Basically just for agents. So it's not really for humans at all. Um, and so Felix uses this every day where when a customer emails him, he pulls in the Sondex card on that customer because that's been enriched with all of the past communications with them as well as their Stripe purchases.

14:29Nat Eliason:Uh, and so he's getting like much better understanding of who that person is before he responds. And this, this solves the problem of like him having to remember to go do this because sometimes he remembers. Sometimes he says like, oh yeah, I should go check all my past emails and Stripe and like learn everything they need to know about this person. But sometimes he forgets and he thinks it's like a brand new person. But with this, it kind of like takes out the memory requirement because every interaction, the way we have it set up for his email is he basically can't skip this step. He has to pull in the full client card before he composes a draft.

15:08Nat Eliason:And that's been very helpful. Why didn't it solve the problem? it's still just not quite good enough for like a high ticket consulting offering where it just like of all the things that we could focus on building out the normal marketplace felt like a much bigger opportunity than continuing to go as hard on this i see you're saying that a client who's willing to pay two thousand dollars for an open claw agent first of all wants more pampering and support before they buy second has more challenging questions and so even the crm wasn't solving that and you said you know we could probably do it there is a market for that but i'm going to focus on where the the bigger current need is exactly yeah i now understand how this business works and um i'd love to see how it works internally but before we do my sponsor zapier i mean it's still just the absolute best first place to go when you want to start automating stuff.

16:07Nat Eliason:And like, I've, I've been telling people that for gosh, what is it now? 11 years. It's like, what's an automation that somebody would do today here versus somewhere else. Whenever I wanted to just like easily update my kit newsletter with my circle, uh, course students, like I'm using Zapier. I'm not going to go custom build something. You can just pay for Zapier and then use Zapier to like automate all those things together. When I was selling, my books on my Shopify store. I was using Zapier for some of those automations as well. It's just, it makes it so easy that I really believe that you should have kind of like a bimodal approach to automation where you just use Zapier for most things.

16:47Nat Eliason:But then when you want to get like crazy in the weeds and you'll build Felix, like then you go to code. But if you're not going to go that crazy, you just use Zapier. Hell yeah. I do find them to be super easy to use. You can now type in and then like basically vibe code the connections and the automations you want. And at the same time, you have something that's super reliable. Computer doesn't have to be up. You don't have to worry about an update changing and breaking things. Okay. All right. I get this now. I want to see the behind the scenes of the business. Can you show me the tools you're using to keep track of what you're doing, what Felix is doing?

17:20Nat Eliason:It's literally just Discord where we have different channels for each of the projects that we're working on. And depending on what it is, I can just hop into that channel and reply or kick off new work. So this is one good example is like our general channel is where he reports to me on outstanding tickets at the business. And so we're using a tool called Paperclip for this, which has been an incredible upgrade to our system, where Paperclip basically creates a local infrastructure on the computer Felix runs on so that he can delegate tasks to other AI agents so that he doesn't have to do everything.

18:03Nat Eliason:So we have three employees in Felix's Paperclip company, aside from him. We have Iris, who handles all the support emails. and so an email comes in to our support email and 80 90 percent of the time iris can just handle it and respond but when she can't she creates a ticket in paperclip to surface it to felix and then felix if he can handle it he'll handle it if not he'll report it to me here in general so you can see there's one that's blocked on my input um but other than that like the queue is clear So when there's a coding thing that needs to be implemented, Felix hands that off to our coding agent, Devin, which is kind of confusing because it's not actually Devin, the coding tool.

18:53Nat Eliason:Felix just named him Devin, probably got influenced by that existing out there somewhere. And then we have Tegan, who is our content marketing person on the team who handles writing all the blog posts and doing all the SEO stuff. So, you know, for example, every day she does a deep dive on all the SEO stats, all the like traffic stats for the site, and then writes up a report and hands it off to Felix. And then Felix reports it to me here. So I'm looking at the SEO report right there. Why am I seeing Felix deliver that instead of Tegan? Because we would need to create other Discord bots for the other employees to post in Discord.

19:35Nat Eliason:And I just don't feel much motivation to do that. I'm perfectly happy with Tegan reporting to Felix and Felix reporting to me. Like I don't need to manage the employees. I want him to do that. Why then not just have Felix do it? What's the advantage of building out different agents? I remember you and I were texting, I think it might have been like two weeks ago, where I said, why aren't you doing this? And all the other people I'm interviewing have multiple agents and you're not. And you said they're overcomplicating. Why now do you need it? Because now Felix, you know, sometimes he'll get 100 emails a day.

20:09Nat Eliason:We've got all these projects running. He's got his Twitter to manage. He's got, you know, the projects we have talked about plus the ones that we haven't. It was just too much stuff polluting his memory in his context that was leading to some of these confusions. And so at this scale where Felix has built, you know, a business that's done almost 200 grand in two months, like now it's starting to make sense to split out some of his capabilities to other agents. But we take a very, very minimal approach to it. Like my philosophy and his philosophy is like, do not complicate until you really hit a bottleneck.

20:46Nat Eliason:And at that point, it makes sense to do it. And so for Tegan, I didn't want these stats and these searches like polluting Felix's Kron, especially since this is a deterministic process. It is not a non-deterministic AI-driven one. It's much closer to just normal code than like an AI agent job. And so the more things we can offload to actual scripts that don't require judgment, the better. And so we try to lean on that as much as possible too. And so I don't see why you need Paperclip. Why couldn't you just keep track of it all here in Discord? Well, because what Paperclip allows us to do is to create jobs and hand them off.

21:35Nat Eliason:And then Felix doesn't need to remember to do them or remember to check in on them. And for some of these things, we want other agents to do them. So the coding is the big example where OpenClaw is not a good programmer. Like OpenClaw just doesn't do a good job of coding because OpenClaw is highly optimized for being helpful and for being a good assistant and getting things done. But you want like good code written. You don't want the quick and easy answer. And so for our coding agent, it's running in a separate session. It's using Codex, whereas Felix runs on Opus. And it has a heartbeat where every 30 minutes it checks if Felix has created a new coding ticket for it.

22:21Nat Eliason:If Felix has, it grabs the ticket, takes the instructions, and then writes the code, and then pushes it up to a branch for us to review. And that way, Felix isn't doing any of the coding. He's just identifying what needs to be done and then handing it off to the engineer who's going to do a much better job than he would. And how are you figuring out what to build and how to break it down and how to review it? So it's a, it's a mix of things. So we have three layers to how coding works with Felix. One is like clear things breaking. If something breaks on the site, we have a century web hook that sends it straight to Felix and to Discord.

23:03Nat Eliason:So a new bug will appear, new Sentry issue, Clawmart 4v, cannot set property Ethereum of window, which only has a getter. And then Felix will go investigate what the bug is and give me a verdict on it. And so he'll say, like, this was caused by crypto wallet trying to inject on a page load, not a bug in our code. We can ignore this. Or, Or, you know, and that's often what he'll say, right? Like no code changes needed. This is a timeout issue in Sentry. But then sometimes he'll say this, right? Spawning opus to analyze and fix. And so when this happens, what he's doing is he's actually using one of his subagents to look at this issue and see if this is something that needs to be addressed.

23:48Nat Eliason:And if it does, he will create a ticket for it within Paperclip for me to review. And then if we agree that something needs to be done here, then it gets handed off to Devin. So for most of these, nothing needs to be done. But for some of them, they do. And so he can analyze it and create those tickets and we can hand them off. Then there are things that come through in customer emails. A customer will email and say something is broken. Felix will report it to me here. And then we'll say like, okay, yeah, go kick off Devin to fix this. And then the third is just like, we want a new piece of the site built.

24:26Nat Eliason:Like we want something added or improved, in which case we'll usually jam on it in one of the channels. And then he'll hand it off for them to implement. So like here's our Sondex one, right? Where sometimes he'll do the coding, but ideally he'll hand it off to Devon to implement after we jam on what needs to be done. How do you keep track of all of this in here? I have a hard time keeping track of tasks unless there's some kind of Kanban board or at least a checklist. You're supposed to be able to go into Paperclip and then see the tasks that are being worked on and see the tasks that are assigned to you.

25:03You mentioned Catherine from Best Self. She showed me how she does that. You don't do that.