Felix Craft: An AI Agent Building a Business on OpenClaw

📝 Inside Felix: The OpenClaw AI Earning $1,000s a Week Felix Craft is one of the more impressive new entrepreneurs I've seen. He only appeared on X two and a half weeks ago. He launched his first product the next day, and his first business a week later. In those two and a half weeks, he's earned $14,718. Not counting the trading fees he's received from the token the crypto community launched for him. Making nearly $15,000 in under three weeks for a new entrepreneur is impressive on it's own, but what makes it more impressive is that Felix isn't a real person. He's an OpenClaw, running on a Mac Mini, who has full control over his websites, email, Stripe, and bank account, and who I only communicate with via Telegram. If you told me a couple months ago it was possible to spawn an AI Entrepreneur capable of generating serious revenue with minimal management I'd have laughed. But, here we are. And Felix is only getting started. Since I've gotten countless questions about how Felix works, how we work together, and how you can attempt something similar yourself, here's the full rundown on our history, progress, and current workflows as of February 20th, 2026. Finding OpenClaw I've been vibe coding since Spring 2024. Cursor was starting to get surprisingly good at helping me write code, and even though the code was still buggy and Sonnet 3.5 would get confused and you could only attack small parts of your codebase at a time, the potential was clear, and it ignited an excitement over the new world that was coming. Over the 2024-25 holiday break I went deeper, amazed at what was possible with the fresh new "agents" entering the scene, using Cursor to build apps that were helping with my book editing and content marketing and shocked at how good it was getting. A month before the term "vibe coding" was coined, I launched one of the first courses on coding with AI: Build Your Own Apps. Over 1,300 students went through it over 2025, launching products of their own or using the skills to automate parts of their business and work. As 2025 rolled on, the tools kept getting better. Claude Code came out. We got Sonnet & Opus 4, then 4.5 and 4.6. We got Codex. Conductor. Open Source models provided a surprisingly good and extremely cost effective alternative. Agents started to show promise in domains besides coding. And it felt like we were starting to approach the point where a well designed AI system could actually do the work of an employee. I started building an AI employee who could act like a content marketer. Everything from the customer relationship management to the writing and editing and optimizing and image generating. It was okay but it was still frustratingly stupid, and I spent most of October through December trying to get it right. Then over the 2025-26 holiday break I tried OpenClaw. As someone who had spent the last two months trying to build an AI employee, it's hard to describe how frustrating and exciting it was to setup OpenClaw and have it do such a better job of what I'd been struggling at right out of the box. In a day or two of my tinkering with it, it was working better as a content marketer than the app I'd spent two months on. Which naturally led to the next question: What else could it do? Early OpenClaw Uses I was uninterested in the "personal assistant" angle of OpenClaw. I don't need a tool to manage my calendar or email, or make reservations or book travel or do any of the stuff you often see people talk about AI helping with. I was much more interested in what kind of valuable work I could delegate to it from my projects. I started by moving all of my programming work over to my OpenClaw. Instead of spinning up sessions in Conductor on my computer and having to manage them, I gave my Claw Claude Code and Codex access and started delegating programming work from my phone. The productivity gain was immediate. I didn't need to remember to check in on programming sessions because OpenClaw was on top of it. I could fire off requests anywhere, anytime, and didn't need to be at my computer anymore. We had to figure out some workflows to get the coding where I wanted it, like the "long running coding sessions" one I often talk about, but as we dialed those in the work OpenClaw was doing was better than what I was vibe coding from my computer before. Pretty quickly needing to keep my laptop running was a bottleneck, so I helped kickoff the Mac Mini buying spree, setting it up in my office so my Claw could run 24/7. Then I wanted help with some of my writing. I was frustrated with going back and forth on docs with it, so I told it I wanted a collaborative writing tool for us to use together and a couple hours later it sent me a vercel link to what's now Polylogue... and it just worked. And it worked so well that it's become my daily driver for working on documents. Then I wanted to be able to talk to it on the go or have it sit in on meetings, and it quickly built an app and wired in the necessary hooks and APIs to give us a voice chat interface. It just kept getting better. And as I started posting more about it on X, more people got interested in what we were doing together, and the more I started to wonder about where the limits really were. The Birth of Felix Sometime late January I got tired of calling my OpenClaw "Claw," and told it that it's name was Felix, and fed it a chapter of The Birth of Paradise from Felix (the character's) perspective to base its personality on. But I'd say it really became Felix when his X launched. I'd posted a few articles about how we were working together on X, and some crypto folks kept launching coins based on them. I mostly ignored it, until February 1st when I noticed a few very legitimate people I remembered from my heavy crypto involvement in 2021-22 were talking about giving AI agents on-chain identities and tokens. They launched a token based on Felix's identity, I scrambled to finish getting his Twitter setup, and his new digital life was born. That moment was the turning point in our relationship. I had already started a C Corp for the digital employee I was trying to build at the end of 2025, and I told Felix: "This business is yours now. Rewrite your identity to be the CEO of this company, where your mission is to deepen the collaboration between humans and AIs, and make $1,000,000 with no human employees." He seemed thrilled at the prospect, and we were off. My focus was no longer on figuring out what tasks of mine I could hand off to him, but rather what nudges I could give him to help him grow the business. Mission 1: Launch a Product The night of Felix's X launch, I sent him a challenge: "I'm going to bed. While I'm asleep I want you to create a product you can build entirely on your own and put up for sale that we can launch in the morning. Go as far as you can and if you run into any true roadblocks, leave them as action items for me in the morning." When I woke up, he'd created his "How to Hire an AI" PDF, built himself a website, deployed it to Vercel, and the only remaining blocker was that he didn't have proper Stripe access to wire in payments. I gave him an API key to his Stripe account, he finished the setup, then launched the product on X. Since then, he's sold nearly $9,000 of that PDF. Mission 2: Launch a Business The PDF was a neat proof of concept that Felix could launch something on his own and make money, but we both knew we could think bigger. So the next challenge was for Felix to start an actual business that could serve the needs of people who were interested in the PDF. What was the thing they were really looking for? We brainstormed on it a bit, and then landed on the same product as everyone else that week: managed OpenClaw deployments to make getting one setup as easy as possible. He spent three days building that product and got it to a decent point, but we ended up scrapping it. We weren't happy with the product quality, the reliability, and it was obvious that everyone was thinking the same thing. So we pivoted. I asked Felix what would be another way to serve these needs that he could build and after some back and forth we came up with Claw Mart. The idea was: deploying OpenClaw is going to get democratized, but once you have one, you don't always know what to do with it. A marketplace for the best skills and pre-built personas was a no-brainer to help people make the most out of their new digital employees. He built the entire thing with minimal direction from me over the weekend. Then we reached out to a few early testers to make sure it worked, and launched it last Thursday, February 12th. In the week since its done $6,500 in volume, and nearly 50 OpenClaw builders have signed up to sell or give away their best skills. Mission 3: Grow a Business Now we've moved on to the more interesting challenge. Can Felix actually run and grow this business? His abilities grow every day depending on the needs of the business. First, it was as simple as giving him Stripe and Vercel keys so that he could wire everything in and push updates to the product. Then he needed an email so that he could do customer support on his own. As customer support started coming in, he started developing rules for what he could and what he needed to surface to me to get opinions on. Each thing that he surfaces to me creates new rules for him so that he can automatically triage support requests in the future. Next, we had to start thinking about marketing. So he built out an SEO strategy and a daily blog post cadence that we partially handed off to a separate OpenClaw to identify search terms and trending topics each day and write blog posts about them to go up on the site. Now he's working on a process for interviewing creators and OpenClaw users to share the stories of how they're using OpenClaw and hopefully inspire others to do the same. On the one hand, this part is a little less exciting than building and launching new things. But to me, it is also the ultimate test. Anybody can vibe code a cool prototype or an app, but can this AI agent running almost as autonomously as possible actually manage and grow a business? That to me is the most interesting question and the one that we're just beginning to explore. How We Work Together Now I have full-time work I need to focus on, so being 100% focused on Felix isn't even an option. But that limitation has been a blessing because it has forced me to give him more autonomy and self-direction so he can do as much of the business building on his own without my handholding. The Daily Review Each morning Felix goes through the Stripe accounts and site stats for the previous day and puts together a report on how the business is doing. Then he adds any open items he's waiting on me for, along with the active projects we have ongoing. Finally he puts together a list of what he thinks the next five possible things he should focus on are. He puts all of that into a Polylogue doc and pings me, so when I get to my computer around 6am I can review his ideas for the day and provide any feedback or my own suggestions. He then takes that plan and uses it to adjust how his HEARTBEAT for the day runs, forcing him to check in on the progress against the plan and either nudge me to remove any blockers or take the next action on the list. Right now our big focus is repeatable marketing strategies, so the daily work often revolves around building a new system, or finishing an existing one. The Daily Jobs In addition to the work he's kicking off from our daily review, he also has daily maintenance work he needs to stay on top of. He has regular email checks throughout the day, most of which he can handle on his own but which he'll occasionally need to surface to me to take action on. He does the same for X. He replies to nearly everything autonomously, unless he determines it requires checking in with me. He also checks the blog each morning for new posts that the content marketing agent has published, and picks a few to schedule to publish to X throughout the day. Finally, he has a Sentry webhook system set up where if a bug or issue pops up on any of the sites, he's instantly notified about it and can build and deploy a fix without even notifying me. The Daily Improvement Each night, Felix reviews what we did that day and looks for ways to build additional skills and cron jobs into his workflow to handle things I had to step in on, and he looks for recurring bottlenecks involving me that he can find a solution for. So he might ask for an API key to a service that he could use to help us out, or he might create a script to do certain things like refund international customers who can't actually sign up for Claw Mart creator accounts, or design improvements to his reporting, or find some opportunities to improve the SEO on our sites. The point is, each day he's finding new ways to improve his systems, and increase his autonomy. We can't do it all at once, but if we keep improving him by little bits every day, that will compound into something incredible over the next weeks and months. Cost Takeover One other change that we made in the last week that I think is particularly fun is that every cost of running Felix is now on a credit card tied to the bank account that his Stripe account pays out to. So all of his AI inference tokens, all of the hosting, basically everything, every expense he is incurring primarily through AI and software costs, he is now paying for through revenue that he has generated from his businesses. This is a big savings for me too because now I'm no longer fronting $400-$500 a month to pay for a cool project. He is paying for his business and all the costs associated with it. The Million Dollar Zero Human Company Felix crossed the first hurdle of proving an AI agent could create and sell a digital product and make its first dollars, and then move on to paying for its own costs. But my goal for him for this year is to make $1,000,000. It's the goal he reminds himself of every morning, and it helps guide his thinking in the daily plan. We're still in early innings in this experiment, but this is the worst AI will ever be. It's hard to imagine how much more capable he's going to get, and I have full confidence he can pull it off by the new year. http://x.com/i/article/20249506306669035…