I’ve been watching a lot of content on working with AI and wanted to share what I’ve learned from my own experience. My background includes jet engine maintenance in the military, IT and information security, and hands-on trades like carpentry and welding. I’ve used AI across both digital and physical projects, and after a couple of years of experimenting, I’ve settled into two main approaches: 1) I start with a clear idea and use AI to refine it and move faster than I could on my own. 2) I start with a rough concept and use AI to help develop it into something more complete. Both approaches work, but I’ve found I strongly prefer the first. I enjoy taking time to think through ideas on my own. I’ll bring a notebook into my workshop or outside, sit in the sun, and sketch or loosely wire frame concepts. That process—just thinking and creating internally—feels like play. Sometimes I go a step further and mentally simulate the idea as if it already exists. I “use” it in my head, test how it functions, and push on its weak points. Doing that has helped me catch design flaws and structural issues early. I treat those like small stress tests, iterate on them, and once the idea feels solid, I bring it to AI to accelerate the build-out. For anyone new to working with AI, a few things stand out: 1) Expect friction. You will make mistakes, lose work, break things, and backtrack. That’s part of learning any new tool. 2) This technology isn’t perfect. It will fail or give bad output at times. Step away when needed and rely on your own problem-solving skills. 3) Treat it as a creative process. There’s real enjoyment in building something, not just finishing it. 4) Focus on quality. Some ideas aren’t worth rushing. Focus on what’s meaningful or interesting, even if it takes longer. 5) Keep it light. Experiment, build things with friends, and don’t take it too seriously. It’s a tool meant to support you, not replace you. 6) Work in a collaborative manner. Some times asking for feedback, suggestions, or thoughts from a model can bring back useful information.

I hit a usage cap recently and added funds to continue. When I returned to Claude Code, I was still seeing the limit notification, which was confusing at first. After a few minutes, I realized the issue: I had previously set a monthly usage limit. Even though I added funds, I still needed to increase that monthly limit to actually use them. Just goes to show that sometimes an issue isn't actually an issue. Its just a configuration change that needs to be applied.

Project for this morning 1) Wipe my spare laptop 2) Install a headless linux server 3) Install ollama3 Done! This works without internet, but it is verrrry slow to respond. My main plan is to leverage this as an overnight agent. Something I can give 1 simple task to and let it do its thing.

Last week I was adding things to a project in CC - VS Code. My /context is 80k from the start. I thought it would be a good idea to have all the skills, all the repositories, everything together, and when I wanted to do a project I could just say "Claude, I want to do X, Y, and Z." What's the best way to do this with everything we have? But I think it's better to have everything separate. How do you guys do it? Thank you in advanced.

I have quite a few questions. Everyone is talking about AIOS as their own product and the possibility of a monthly retainer, but since you mainly focus on the power of Claude Code, I’m not really seeing how this is actually deployed to clients. Since Claude Code runs locally, you can do a one-time setup—folders, skills, config, context, etc.—and hand it over, but that’s where it ends. You don't have ongoing access to their local machine. On the other hand, if Claude Code is only meant for developing these systems, and the actual client solution is built from scratch using the Claude API, then it kind of loses me. The raw API doesn't have "skills," plugins, or commands out of the box—it honestly doesn't even come close to a polished tool like Claude Code. So, my main question is: What is your actual approach? How do you realistically build and deploy these AI systems in the way everyone describes? Thanks a lot! 🙏