Hi everyone! I want to share a recent experience with AI. Although at this point, “odyssey” might be a more accurate word. A while ago, I attended a webinar by Czech AI enthusiast Petr Ludwig about building your own AI team on a virtual server: connecting different models, talking to your agents through Telegram, assigning tasks from your phone, and creating a Human-Agent Wiki where both humans and agents can store knowledge and retrieve it later. 📖 You may know Petr Ludwig as the author of The End of Procrastination. He’s Czech, but the book has been translated into more than 20 languages. He’s also a somewhat controversial figure in the Czech online space. I knew that. What I didn’t realize was how relevant that little detail would become when I decided to share my progress on LinkedIn. So I built myself an AI team Following the webinar, I set up a system with four agents. ⚙️ One acts as the CEO/orchestrator of the team, while the other three specialize in different types of tasks. Under the hood, I’m experimenting with different combinations: - one specialist uses Hermes + ChatGPT - another uses OpenClaw + a lower-tier ChatGPT model - the third uses Hermes + Gemini and is my expensive specialist — I only send it in when I need deeper research rather than everyday tasks And it works. That was basically the point of my LinkedIn post: Hey, I actually built this thing and got it running. I also shared a couple of things I had discovered along the way. ☝️ First, the cheapest ChatGPT Go subscription Ludwig suggested in the webinar wasn’t enough for my setup. I burned through its limits in less than four hours while I was still configuring everything. ✌️ Second, the OpenClaw command-line installer had changed since the webinar was recorded. Somehow, after a fair amount of wrestling with it, I managed to get the original setup running anyway. And let me tell you: troubleshooting becomes significantly more entertaining once you can already complain about the troubleshooting to the CEO of your AI team via Telegram.