I decided that it was time to graduate from using regular chats with AI to using Claude Cowork. When I saw ICM, the lightbulb came on for how I was going to implement it. My previous state was to operate in multiple chats, download artifacts to an unorganized Downloads folder, and my work output lived in several places. This helped me organize and gave me a way to structure my work around what I used AI for. Instead of thinking through a workflow, I took some time to reflect on which topics in my day to day life I tended to use AI for help with. Those became my root folders. Inside each of those folders are the types of work or the mental modes that I work with them on. As an example, some of my root folders turned out to be career, health, work, homelab, garden, etc. When I am working on a certification, each cert has a workspace in the career folder. If I generate an artifact to help with exercise, it goes to the health - fitness folder. I built a skill that scaffolds any new directories into the structure. The new folder gets a default AGENTS.md and CLAUDE.md file (just a redirect to AGENTS.md) created that maps the folder. I tune the rules. This has allowed my work with AI to shift from long chats to build context, into an interactive workspace where my prompts are simply the tasks and the category. The context lives in the structure. Claude knows which certifications I have, which ones I am working on, and why I pursue them already. It already knows which plants I grew in the garden last season and which beds they were in. It knows which apps I wrote with AI, what their function is and the URL of the github repo. I'm excited about what this opened up for me and looking forward to developing workflows using this system. I've got some specific ideas revolving around system administration and security tasks that I plan on working on. Hope this helps someone else. Start from where you are and look for one way to improve your current systems. Implement it. Repeat the process.