I created my first three-file folder system for a personal blog and writing workspace. When asking claude, the CLAUDE.md included helping me write clear, plain-language blog post and asking clarifying questions before drafting. The CONTEXT.md explained the blog's subject, my target audience, and that good output means concise posts with a friendly but direct tone. The REFERENCES.md provided a few of my favorite past posts and examples of the writing style I wanted. When I asked Claude to write an article on my topic, the result was different because it was familiar with my audience, tone and idea of good work. It felt less generic and more like a draft from someone who had read my blog. Previously, I had to explain my writing style, audience and post structure each time. Now, Claude already has all that information. Key takeaway: writing the three files once means I don't have to keep explaining myself in every conversation, and the quality of the first response is much higher.

I just watched the Jake Van Cleef breakdown, and the biggest thing that stood out to me was you treat your project folder as the source of truth, with agent.md files acting as a "control plane" that redirects the AI to instructions, skills, or state files stored on your local disk.. It clicked because makes you the developer of your own agent's long-term memory and personality.. My main takeaway was how it demonstrated that instructions can be modularized and hidden in the filesystem, turning the agent into a navigator rather than just a conversational partner. The experiment at the end helped me understand how to adopt a "Router-based Onboarding" workflow for a collaborative project. Going forward, I'm going to test it out on a one page website project.

I'm currently using AI for my research. This month, I started applying ICM to build a one page website. One thing that caught my attention in this lesson was the three layers, because with a folder structure, you can start building something and track where you may have gone off track. I'm excited to learn how to create an animated presentation like Jake's. I'm also looking forward to seeing how structured approaches to AI can help me build better systems and reduce complexity.

Hi, I'm Peter and I'm happy and excited to be here. 🙋 A little about me: I work in software product management. I enjoy the intersection between of business and technology people! 🎯 My current goal: Learn how I use an LLM model with Jake and David's ICM methodology for the use cases Jake has clearly explained in his classroom and videos, especially for sequential workflows. 💪 What I'm currently building/working on: I am currently in the discovery and learning phase. I am reviewing the first competition project. I am also testing Jake's "ICM-Architect" skill on an LLM council workflow. Cheers!

I'm currently using AI for building and running a structured prospecting and outreach system — signal detection, research, and drafting — and right now I'm firmly at the building structured systems stage. Everything runs off documented skills, a registry of triggers, and a doer/verifier agent pattern rather than one-off prompts. One thing that caught my attention from this lesson is the idea that context should be legible to a human, not just consumable by the model — if I can't read back what the system knows and why it decided something, I can't debug it or trust it in front of a client. That matters enormously to my work: I'm a Fractional CTO, so anything I put my name to has to be defensible, and "the AI said so" isn't an answer a board will accept. I'm excited to learn how to structure context so it stays consistent across long-running, multi-step workflows, and see how structured approaches to AI can help me cut the re-explaining, reduce the fragility in what I've already built, and get from working prototype to something I'd happily hand to a paying customer.