How I Run My Entire Business With AI Agents (Full OpenClaw Setup) transcript

I run a marketing agency, a trading bot, a content pipeline, and a YouTube channel. I have one employee helping me do all of it. His name is Max. He's an AI, and he runs 24/7 on a $600 Mac Mini sitting on my desk. [music] Every morning when I wake up, I have a brief waiting for me. My tweet ideas are drafted. My trading bot has been trading for me all while I sleep. My YouTube ideas are researched and my competitor analysis is complete by the time I wake up.

I used to do all of this manually. It took me eight hours to get this done every day. Now takes me 2 hours. I'm going to show you how I built this, how much it costs, and how to set this up in this video. No code required. [music] Let's get into it. So, first up, what is OpenClaw? OpenClaw is an open-source framework that lets you run AI agents on your own machine. Think of it like hiring employees, except they live on your computer.

They work 24/7 and they cost you about a hundred bucks a month. So, as you can see here, these are my four agents that are running. Max is my chief of staff who I talk to every day on Telegram. He manages all the other three agents. Think of him like the manager. There's Sage, my content specialist. He drafts tweets, analyzes performance, studies competitors, finds trending topics, generates about five to seven tweet idea drafts every single day.

Then there's Knox, my trading operations agent. He monitors my Polyarket trading bot 24/7. He checks for errors, investigates losses, reports health every two hours that passes. And then finally, we have Nova, my YouTube strategist. She tracks metrics, researches trending topics, pitches video ideas to me. I mean, this video that I'm filming exactly right now, Nova helped research it for me and even script it based on the information she has on me.

The key thing though, I only talk to Max. Max delegates everything he has to all the other agents. He collects results. He reports back to me on Telegram. It's like having a COO who manages all the team members below him. I mean, let me walk you through a typical morning. Every single day at 7:30 a.m. I wake up to this exactly this message that you see right here. Sage has already scraped what my competitors posted overnight.

He found trending tweets I could react to. He drafted five tweet ideas based on what's actually performing. and he analyzed my last 107 rejected drafts to avoid repeating any mistakes. And the crazy thing is I don't even have to ask for this anymore. This is just an automated job. I get this every single morning. All I have to do every day is just open Max HQ, which is this platform that I built with the help of my AI agent, and see all of my drafts.

We can go to tweets. We can see okay what drafts did it pitch over here. Do I approve? Do I reject? If I press reject, it gives me why. So it asks me why am I rejecting this? So I can keep giving it more and more feedback. So it can keep on improving every single time. And this is what blows my mind. Every single time that I reject one of its submissions, one of its ideas, I get to give it a reason to AI uh inaccurate information over here.

No value and every single time uh it learns from each and every single rejection so that it can keep improving. Already posted something like this. So there you go. Meanwhile, Nox, my second AI agent, has been watching my trading bot all night long. This bot trades crypto prediction markets on Poly Market. It has over 550 trades that it made with a 92% win rate. When something goes wrong, Nox investigates the trading bot, finds the root cause, and either fixes it right away or tells me what happened if it's something crucial, critical that it absolutely must need my approval for.

And finally, Nova takes care of this page in the dashboard, the videos, and she helps me come up with ideas. You can see this one. I gave my AI to build its own AI team. I I let my AI build its own AI team. If I like this idea, I can approve. It can generate a script for me using my voice. I can go into the script. I can edit it. I can see what's to film. I can mark it as filmed. I can track it if it's posted or not. if it's posted, I can track my metrics so that it can study how well my videos are doing and which videos are not doing so well.

So, it can learn every single video I post, what works, what doesn't, what do I need more of, what do I need less of. Okay, let's talk money because everyone asks about that. My total AI spend for the month is about $140 every single month. We have the clawed API which powers all four agents that's about $100 a month. We have the cost of the electricity for the Mac Mini which is about roughly $20 a month and it runs 24/7.

We have the open claw which is completely free and open source so zero cost there. And then we have other APIs, things like Grock for X search or research or trends on X, for example, that I approximately spend about $20 every month on as well. All in all, $140 every month for what would cost $3 to even $5,000 to hire a human assistant, and they'd only work 8 hours a day. The Mac Mini was a onetime $600 purchase and basically paid for itself in month one.

And I want to be honest, this didn't start out at $140. First month, I was burning $300 because agents were doing dumb stuff, for lack of a better word, running expensive models on simple tasks. I optimize my agents over time, put cheap models for simple tasks and expensive models for complex reasoning. So, how do you actually build something like this? I'm not going to walk you through the installation of Open Claw that takes 10 minutes and there's plenty of docs and tutorials out there for it.

What I want to talk about is the part nobody explains. Here's the framework I used. Step one, track your time for one week. Before I built anything, I wrote down every single task that I did for 7 days. Every email, every tweet, every spreadsheet, and I categorized them with creative work, repetitive work, monitoring work, research work. It turns out 70% of my day was repetitive and monitoring tasks, checking analytics, scanning competitors, drafting first versions of content, watching my trading bot, that's agent work.

The other 30% strategy, client calls, final creative decisions, that stays with me. Agents don't replace your judgment, they replace the grunt work that eats away at your day. Step two, start with one agent, not four. I didn't build four agents on day one. I started with Max, just a general assistant on Telegram that I would ask, "Hey, Max, summarize this article. Hey, Max, what's on my calendar?" After two weeks, I noticed I kept asking Max the same types of questions over and over about Twitter, what's trending, what can I talk about?

That's when I knew I needed Sage, a dedicated content agent. Same thing happened with trading. I was checking my bot lock six times a day. So, I built Nox to do it for me. The pattern is when you catch yourself doing the same task repeatedly. That's your next agent. Step three, give your agent a personality file. In Open Claw, that's typically called soul.md. It's a text file that tells your agent who it is. Mine says, "Have opinions.

Be concise. Call me out if I'm about to do something dumb." And this matters more than you think. Without a personality file, your agent sounds like every other chat PT rapper out there. Just generic, hedging, useless, [snorts] >> for a lack of a better word. >> With a personality, it actually sounds like someone you'd want to work with. Not you have to. And then there's even user.md. That's where you tell it about yourself, your business, your goals, your preferences.

The more context you can give it about yourself, the less you have to repeat yourself in the future. I remember when I was setting up my open claw, at first I literally just recorded a 15minute voice note just telling it about my life and about me and about the things I do and my business and the things I like and the things I dislike. And I remember I was like, "Ah, I have even more things to say." and I sent it another 10-minute voice note just telling you about my personal life, my hobbies, the things I like to do outside of work, so it can know as much about me as possible and eventually be able to help me in areas I don't even know I need help in.

Step four, build the feedback loop. This is the whole game. Most people set up an AI agent, get mediocre output, and quit saying, "Oh, AI isn't ready yet." No, you just did not train it, my friend. Every time Sage sends me a tweet idea, I either approve it or reject it with a reason. Too generic. We posted about this last week. You just saw it. No one cares about this topic. Over 110 rejections later, Sage reads all of the rejections before writing anything new.

It knows my voice. It knows what I hate. It knows what performs. I even had it build a system, an automated system that categorizes every rejection reason and tracks which topics I'm tired of. My agents literally study their own failures. And it's important that you put that in the process of your agent, that you tell your AI agent, hey, I want us to build a feedback loop front and center of our workflow. This is the part that separates a useful AI setup from a toy.

You have to treat it like onboarding a real employee. The first month is going to be rough, but by month three, it's writing drafts I actually love and actually want to post. And finally, step five, automate the schedule, not just the tasks. Setting up one agent that responds when you ask is level one. Level two is agents that work while you sleep, my friend. I have scheduled tasks running throughout the day. Whatever I'm doing, whether I'm sitting uh drinking coffee, whether I'm going to the gym, whether I'm showering, my agents are working for me 24/7.

At 7:30 a.m., I get my morning brief with my tweet drafts, competitor analysis, and my calendar for the day. Every 2 hours, I get a trading bot health check analysis. At 10:00 a.m. and 3 p.m. I get fresh content ideas based on what's trending right now. And at 8:00 p.m. I finish my day by getting a performance review of everything we posted and worked on together with my agents throughout the day. I spend time setting these up once and they've been running every day for months now.

When I wake up, work is already done. >> [sighs] >> Look, I'm not going to pretend AI agents are magic. They make mistakes, too. Sage still writes tweet ideas I reject. Nox can't always figure out why a trade went wrong. Nova pitches ideas I've turned down five times already. But here's what changed. I went from doing everything myself to reviewing and approving work that's already done for me. My job shifted from doing the work to directing the work, operating the work, orchestrating the work.

I 10xed my output while doing a quarter of the work. That's not me hyping this technology up. That's my actual life right now. Oh, and if you want to see what happened when I let my AI trade real money for 14 days, you can find this video on my channel. Subscribe so you don't miss it. See you in the next video.