5 Things I Learned Running OpenClaw For 30 Days

0:00So you know that whole idea of AI agents running your business? Well, I'm going to break that down how I'm implementing it in my business and how I'm able to pump out so much content by using Open Claw. I actually did it over the last 30 days of kind of getting the feet on the ground and making progress with getting Open Claw going and how best I'm going to automate it. I now have a team of AI agents working non-stop 24/7 on the Mac Mini behind me. YouTube scripts, thumbnails, Twitter posts, blog articles, podcast outreach. I'm going to show you and talk about exactly how I built it, some troubleshooting that I went through over these last 30 days from completely uninstalling Open Claw and restarting over and just focusing on

0:45the small things that I wanted to automate first and actually going back and forth with giving it personality and how it kind of learns how my content style is can be very helpful for people that are, you know, trying to get into content creation. This isn't a clickbait experiment or anything. I'm not going to tell you how to make a million dollars in 10 days with Open Claw. I'm just talking you through the process of how I got it set up, what I'm automating and how I'm saving about 10 hours every week from not doing redundant tasks. So, let me walk you through this. First, we have our first agent, which is the brain controlled by Opus 4.6. I use the OAUTH token through Anthropic. It has been one of the best. I've tried Codex, I've tried GPT-4, I've tried Mini Max, Quinn,

1:31I tried uh Kimmy K2 and all those just aren't as good as the brain, the the one that facilitates everything within your Open Claw. And Opus, I know it eats a lot of tokens. I'm sure you can get away with Sonic 4.6 as well, but Atlas is the one who I communicate with through Telegram and through the dashboard on the Mac Mini as well. So, Atlas is the chief, the CEO. It's an AI agent running in the background right here on the Mac Mini called Open Claw. You heard of it. I know you have. It's an open source 100 I mean 100 over 150,000 stars on GitHub now. Over 80 developers working on this project. So the cool thing is I own all my data. So I have no monthly hosting

2:16bills. I don't have a VPS server or anything like that. The Mac Mini was 500 bucks. Everything stays local, secure. I have a lot more control over it. But here's what makes Atlas the first agent I created different from like a Chat GPT or Claude. Atlas actually has an identity. So there is a markdown file associated with Atlas the agent. And the emoji is a globe because his whole vibe is sharp, resourceful, direct, and gets things done without the fluff. I don't want to be, you know, I want things done and told to me like how it is. I don't want any fluff. I don't want any romanticizing of anything of what we're trying to complete and accomplish with Open Claw. So another important markdown file is the soul markdown file. And I know it sounds weird, but it's exactly

3:02how it sounds. It's you need some soul. And the soul is the mind, will, and emotions of the personality traits of who you're going to be communicating with. So it's basically telling the agent that it's not a chatbot. That it's not just an ordinary LLM. It gives it some character. So it's becoming someone, right? And it defines its personality, it has opinions, it's resourceful before asking help, it earns trust through competence. So when I'm talking to Atlas, it doesn't feel like I'm talking to like a generic AI is what I'm saying. Like we've had the Chat GPT experience years ago where it's just kind of gets redundant and it's the same old same old. So it doesn't feel like that. Feels like I'm talking to an actual business partner that's tailored to my specific business and needs.

3:49Here's the crazy part is it has real memory. There is a memory markdown file that persists across all sessions. So, when Atlas wakes up each morning, he reads his memory file and remembers who he is, what happened yesterday, and what our mission is. I'll just refer to Atlas as he because I don't want to mix up terms. It's a person, right? He keeps notes about me in the user markdown file. So, this is like the time zone, what to call me, the funnel structure, everything about what I'm trying to accomplish, and everything it knows about me. This is very important to do when you're setting up your open claw. You want to deliberately tell it's your missions, your goals, a little bit about yourself, so it can understand how to best serve you. So, every single

4:34session, Atlas reads the soul markdown file to remember who he is, and the user markdown file to remember who I am. The memory markdown file for what happened before, plus daily memory files in something called a feedback log, also a markdown file. So, he doesn't repeat the same mistakes twice. It's like giving an AI agent an actual brain with personality, memory, and self-awareness, which I know sounds scary. Atlas is more than alone. We have several other agents, though I do recommend spending the majority of your time setting up your first agent, which is the one you're going to be communicating with the most and using a strong LLM to be able to facilitate all the demands that you are requiring. So, I built Atlas and a whole team.

5:20Because I'm a nerd at heart and I wanted a cool theme. So, I love the Wu-Tang Clan. So, we have the GZA, the RZA, Inspectah Deck. We got Ol' Dirty. We got several other team members to build out. We have Prodigy in there, too. But, that's who I named them after. I just wanted to keep it simple. That's what I love. Curate it however you want. I don't care. So, every morning at 6:00 a.m., I get the agent RZA wakes up via cron job, scrapes five Reddit AI uh subreddits, so trending topics. It also scrapes Hacker News. We have the Brave Search API integrated, so it could pick up trending AI topics. Then, this is where it gets kind of crazy. He generates a full video script in my voice, not generic content, but using a

6:05detailed voice profile that knows I say, "You know," naturally, and I transition with, "So," a lot, that I sign off with, "We'll see in the next one. Be blessed." And what I did here is I have 600 YouTube videos on another channel. I had it go through like my top 100 video transcripts and create and humanize my voice. So, even deliberately creating its writing style to make intentional mistakes because most of the time when you're reading content or reading a script, it sounds like AI, and it just can be a little bit unauthentic. Just I want it to be like me, how I sound. So, I gave it as much content as I could, all my tweets, all my newsletters, all

6:50my emails that I have personally written over the last five years. And this really helps me speak naturally and not have to like be someone who I'm not, which is really important if you're doing any type of copywriting or building any landing pages or anything like that, sending messages, DMs, creating cold emails, things like that. Those are very important. So, the AI knows this. The AI doesn't just write the scripts though. It generates three thumbnails a day using my actual face. We go through the kai.ai API, which has a marketplace of multiple image-to-video models, text-to-video models, and text-to-image models. So, that's what we use, and it basically uses my face as the reference image, and it does like a close-up shot, a dramatic

7:35pointing shot, me at a desk shot, and I'm able to test those thumbnails as we release videos. Each thumbnail has a topic-specific icon like the OpenAI hexagon or Anthropic A. Then it writes the YouTube description with the chapters and a full SEO blog post that auto publishes to our Shipping School community. Um we have our shippingschool.com and our community as well, but I like to post the blogs on there to help with the Google Search Console for SEO. So, all of this happens while I'm sleeping. All of this used to take me at least 2 hours to do by myself. So, at the same time, Inspected Deck, that's the other agent, generates the five Twitter posts for the day, morning update, building public, AI insight,

8:22value post, and the evening reflection. So, all lowercase because that's my authentic style and each under about 6 to 700 characters. Then there's the Jizza, don't forget, who handles the newsletter content on Mondays and Thursdays. Podcast Scout runs at 7:00 a.m. and uses the Brave Search API to find AI founders on Twitter with about 1 to 50,000 followers who are actively building cool projects. Then it generates personalized DM pitches referencing their specific work and includes my Calendly link. That's the one that blew my mind, but this one's even crazier. We spent a day, you know, doing this one, is the Clip Factory. So, this thing runs every 2 hours via the heartbeat. So, every time it runs a heartbeat check, it checks my YouTube

9:08channel via the YouTube API and looks for long-form content. And basically, it takes the recent YouTube videos, transcribes them with Whisper, sends the transcript to Claude to identify the most viral 30-second uh second segments. Then it auto crops them to 9 by 16 with like a Mr. Beast style animation caption. And the whole tech stack is pretty straightforward, but that was the first thing I wanted to build out just to kind of get the thing going as you're going to see more and more podcasts on this channel. That took about a whole day to get going. So, Open Claw is running locally on the Mac mini. You don't need even a Mac mini to get it going. Dust off that old laptop, whatever you have. Mission Control Dashboard, I would say is like the most crucial thing to get going because it helps you organize a lot of

9:53your thoughts, what you're trying to accomplish, build out your dashboard however you want. If you want more of a visual dashboard, if you want more of a minimalist dashboard, whatever, build it to tailor your needs. So, it's all hosted locally with all the tabs like trends, videos, shorts, calendar, Twitter posts, clips, podcast outreach, the blog. And I migrated from SQLite to Neon Postgres SQL because that's just what I've been using for all of my web applications. It's just what I'm used to. So, the data persists in the cloud and I don't lose content when I restart. Another hack that I want to mention is that it's not for everyone. Not everyone needs this, but getting It was what it's called Supermemory.ai.

10:39And from what I know is it just has a larger memory reach, larger memory context. So, it's able to really remember a lot of the things that we work on, that we do because that's the one common downfall of Open Claw is that it just has a hard time remembering the things that you do and you feel so proud, you shipped some feature, and then the next day it's like, "Hi, how are you? Who are you?" And you're like, "Oh my god, what is going on?" Can be very annoying. So, Supermemory solves that problem. I'd highly recommend that if you're going through that issue. So, for AI, mainly just using Claude for the scripts and the analysis. Gemini Flash seems to be the best for like cron jobs because it's cheaper and Nano Banana 2 for thumbnails with my face reference. It's like 8 cents a

11:24generation. The thumbnails deserve their own explanation because this one was a little tricky. I gave the system three styles using my actual face photo. There's like the desk shot where I'm at my setup, a close-up where my face fills the frame, and the dramatic one where I'm pointing at some logo, you know, the standard YouTube face, whatever. Um and we're testing that. So, click-through rate and uh impressions. So, the voice profile, that's in a another markdown file called boy vo- bow voice, uh has my speech patterns, you know, phrases I use, the energy level, how I do the call to action and things like that. So, when the AI writes the scripts, they actually sound like me. For blog generation, I'm using the Dicky Bush and Nicolas Cole writing frameworks, the 1 3 1 rhythm, curiosity

12:12gap headlines, proper rate of revelation, and the blogs aren't just AI word slop. You can clearly tell nowadays when something is written by AI, and most the time you don't want to read it. It's just so generic. I intentionally want mistakes in my copy, and I'm still overseeing everything that gets published to add in a little bit more of me or maybe my thoughts on what was written. I think it gets you about 90% on the way there. I don't think anyone should really just hit publish. You should have go over it, maybe fix a few things, or add in a few takes of whatever you're going to be publishing. So, all of that like follows a specific proven copper copywriting structure. Another one for the integration, uh I

12:58really love what Jack Fricks is building, and I I think Post Bridge is a great option to um post anywhere you want on social media. So, we got my Facebook page hooked up, Instagram, TikTok, Twitter, um and YouTube for all of this. So, you could connect right into the APIs there. I think it's like $30 a month, not bad. So, it all works, all connects, and we fire it off all the content, but I I do it manually. I don't have it automatically post stuff. So, the thing that gets really interesting is Atlas doesn't just respond when I talk to him. He's proactive. So, every 30 minutes he gets a heartbeat ping and checks is the mission control running? Are the cron jobs healthy? Did today's content

13:43generate any stale jobs that got stuck because they will get stuck? If someone's If something's down, uh Atlas restarts it, automatically alerts me on Telegram, but it also does some of the proactive work, too, like updating the trend signals page, compacting old memory files, syncing everything to super memory. I'll touch on a little bit more is I integrated it and it just really has been helpful. So, Atlas syncs with the daily memory and the long-term memory markdown file to the super memory cloud. That means his context persists even if local files get wiped. He has three layers of memory now, daily logs, which are the raw data, the memory markdown file, which is the curated highlights, and super memory as the cloud backup. So, [snorts] he

14:28periodically reviews the daily files, promotes the important stuff to the long-term memory, and just like human journaling. So, security was a big one for me. That's why I got my own local machine. I wanted to be as secure as possible. Everything runs locally. So, my data never leaves the machine unless I explicitly set up cloud APIs. It's all on a blank slate with no logins or anything like that, um just for safety sake. So, I'm often times using this laptop behind me and, you know, this computer to kind of pull information to um OpenClaw. I'm sure there's a better way to do it, but that's just how I've been doing it. OpenClaw also has a a security health check skill that audits the firewall settings, the SSH config, and system updates. Every Saturday

15:14morning a security audit cron sends me a report on Telegram, which is called Prodigy. Atlas has explicit boundaries in soul markdown file. Private things stay private, period. When in doubt, ask before acting externally. He won't send emails, tweets, or public posts without asking first. Internal actions like reading, organizing, learning, those are free. External actions require permission. So, I got over 60 skills installed now. Apple notes, Apple reminders, uh one password integration, GitHub for PR reviews, CL monitoring, coding agents that could spawn Claude code, dev for development, uh weather, Spotify, Sonos control, camera snap, a lot of things that are permanent. Text

15:59to speech, 11 labs, voice calls, Discord, Slack, iMessage, WhatsApp, a bunch of stuff, right? PDF analysis. So, it is like a Swiss army knife that keeps getting new tools. Skills are the key to, you know, having a productive agent. And you could always go to the Claude Hub skills repo or webpage and and have it look at skills that are relevant to what you're trying to build with, you know, maybe have a a limit of, you know, certain skills you don't download if they have a high, you know, vulnerability to malware or something like that. Um I only want skills that are vetted and are secure to avoid any kind of prompt injection. So, the setup took about 30 days to refine this. I've been resetting it, restarting it, fixing errors. I've spent over 100 hours

16:44getting this thing working, but every day there's still something that breaks. And I think that's just an ongoing process. I really don't think Open Claude is quite there yet to far as what we're going to see come in the future. It's definitely something to fun to play around with and actually build something, but I think it's still a work in progress. And there's updates coming out every week. It seems like Peter's posting a new update every week. And it's so much better than when it first started 30 days ago, it was scary. Now there's a lot more security protocols, a lot more skills that are integrated to get it up and running. So, 30 days, 1 through 3, I got Open Claw going. I don't think I slept more than 12 hours in those 3 days, but I did set up Atlas, the identity, and I connected to

17:29Telegram. Days 4 through 7, I really worked on like building out the Mission Control dashboard and setting up the trend scraping from Reddit and Hacker News. Week 2, I added script generation with like my voice profile and thumbnail generation with Nano Banana 2. Week 3 was more of like getting all the social setup through post generation. We got Inspected Deck running on that week. We had the blogs and the auto publishing come out and podcast outreach, too. If you want to jump on the podcast, hey, send me a DM, leave a comment down below. We will be releasing a podcast pretty much every Monday. And that final week, week 4, I added the clip factory cuz now we're doing more long-form. And I just wanted at least three to five clips that I could post every day, like highlights. So, Super Memory Sync was on that final

18:15week, as well, and the security audits got tightened down on that final week. But, 30 days from zero to a fully automated content machine, I do see myself saving more and more time now that I'm not having to do all of the work. Definitely worth it. A $200 a month is what I pay for the Claw plan, and it works. I'm sure if I was shipping more and doing more with it, I would run into problems because I am hitting getting close to hitting the weekly limits, but I haven't hit it yet. So, Atlas knows the mission, and that's the main thing for what we're trying to build. You know, it's in the memory markdown file, 50,000 MRR by the end of this year through Shipping School, YouTube ads, sponsorships, and the newsletter, 100,000 YouTube subscribers

19:00by the end of the year, if I could talk, which is a huge goal, which is why we're going to be cranking out content on here, trying to provide as much value to you as possible. Like, kind of my journey, what I'm doing, and then also updates. So, everything is the a funnel, you know, YouTube plus Twitter plus shorts plus newsletter leads to the Shipping School community. The Shipping School community is just like-minded people where we can hang out and share cool projects and get your app built because it's hard to do this alone. It really is. You can watch all the YouTube videos, read all the Twitter posts, but if you have no one to hold you accountable to what you're trying to build, building is just much harder. It really is. So I have that link down below if you are looking for a like-minded community and you want to actually build what you're thinking about. You want to turn your idea and

19:45monetize it. So everything kind of all leads to that. I'm trying to build the community and we're trying to help people grow. Every piece of content that Atlas generates is oriented to driving people to the Shipping School community and to the newsletter and to this video, basically. So I'm doing the important stuff. I record every video myself. This is like the talking head style. Not a ton of animations, but I just want to be able to give you the raw meat and potatoes, no fluff. So I review the script, I go over it, I cut some out, I use it as is at times, and then sometimes I just have it rewritten. So I approve each thumbnail, too, cuz sometimes they turn out terrible with random text. Every Twitter post usually turns out okay, but I do need to alter

20:31them at times. I make all the creative and strategic decisions. Nothing just gets posted automatically. So Atlas is kind of like the prep cook in the kitchen and I'm the chef. I'm dishing the food, right? Uh oh, that that reminds me, we need the chef up in here as one of the agents. We need a Raekwon the chef. So that's that's I got to remember that. Also, what it's like actually like working with these AI agents is kind of mind-blowing. Um for me personally, it's like having a team, like a an actual board of directors or a team that's helping you grow personally or in business. it's pretty cool. Um, still kind of not I haven't gotten used to it. So, it's pretty weird, honestly. Uh, when Atlas sends me a message in the morning in Telegram, it

21:17doesn't just say content generated, say like, "Hey, morning. Brief is ready. Uh, Riza crushed it today, found some spicy drama in the GPT-4 subreddit. The script's solid. The thumbnails are pretty fire, especially with Nano Banana Tube coming out. Uh, the podcast scout has been We use the Grok API and it's, you know, looks for people that would be relevant to jumping on the podcast and sharing what they're building, what they're shipping, what they're learning. So, all that's saving me a lot of time. So, that's literally how he talks because of the sole markdown file. Sometimes he'll push back on some of my ideas. Last week I wanted to pivot the content strategy and Atlas basically said, "That breaks the funnel math we've been building. Here's why." And he was right. So, you want the AI to challenge

22:02you and not just go with everything that you say. The weirdest part is when something breaks and he fixes it before I even notice. Then casually mentions it later. "Oh, by the way, Cliff Factory was having issues with the DALL-E API this morning, so I switched it to the backup method and logged the error for debugging." Interesting. So, that's the cool part. A lot more of a glimpse of into what agentic LLMs are capable of. We're seeing a glimpse of that now. So, it's like having a team of business partners that never sleep. The results in 30 days, we've published 22 YouTube videos, 150 Twitter posts, 45 blog articles, and 78 personalized podcast outreach DMs, and created over 200 short-form clips. My engagement on Twitter is up 340%, my YouTube watch

22:47time is up 120%, and we've had 12 podcast bookings, and as well as over 150 subscribers in the last 2 days. That's crazy. So, Shipping School sign-ups are yet to come in, but I just built the community that has the startup starter pack that has the courses in there and you know that's going to be like the the playground where people can grow come and grow and learn about AI and we'll meet you where you're at. Definitely tailored towards more beginners and people that want to get into building and taking their idea and actually seeing it come to life. So, I'm not working harder by any means. I'm working smarter is what they say, but I'm working actually just more creatively on how I could use these agents and how they kind of would fit my

23:32lifestyle and me personally and in my business. So, Atlas handles a lot of the execution. Like I said, we use Opus 4.6 for all of the execution. And here's what I learned the most is you can't just throw AI at your business and expect magic. The magic is in the constraints. Atlas works because I gave him a clear identity, specific boundaries, detailed memory systems, and a mission he can optimize towards. The tech is just the plumbing, if you know what I mean. The real breakthrough is threatening The real breakthrough is treating AI agents like an actual team. A team of members with personalities, memories, and accountability. So, if you want to build something like this, like your own agent team that's going to fit your business, fit what you're trying to build, maybe just you want to become a

24:18better more optimized person. That's literally what we teach and talk about inside of Shipping School. We do three live calls a week. You get the starter kit that allows you to have database authentication payments plug and play you're trying to build. And I walk you through building that exact setup step-by-step. shippingschool.com is the link. I'll put it in the description below. So, I'm not saying like replace yourself with AI, but work with AI to become more optimized, to buy back your time. I'm saying give AI the grunt work so you could focus on the creative stuff only the humans can do. So, you're kind of like that overseer. So, anyway, that's my 30-day experiment with Claude code and open claw and building cool side projects. And when you stop thinking of AI as a tool and start

25:05thinking it as like a team member, you will be shocked. If you haven't subscribed to this channel already, what are you waiting for? Hit the subscribe button. We got updates coming to you every day. We'll see you in the next one. Be blessed.