This was such a challenge. I had a idea for a hero landing page design but it was not coming together as I hoped. It got to the point where i was using GEMINI, CLAUDE and CODEX to "just get something close it". Alot of back and forth prompting, folder structure fell apart the first time. I just had to keep engineering it. Tried it forward and reverse engineering it. I was patient but very persistent. I COULD HAVE BEEEN BETTER ORGANIZING AND STRUCTURING MY FOLDERS. It seemed like there was no hope. It seemed no matter how good i got them, the design was just odd, not as direct and very confusing. WHAT DID THIS DO....? ACTUALLY, it inspired me to create my own tools and workflow systems so my designs can manifest exactly how i please and i can have more control of the details. Something ill be speaking and following @Ari Evergreen and @David Vogel about. And of course more videos from @Jake Van Clief. I went into gemini and created parts of the design and had it add controls for it so i can get it looking how i wanted, then took the code and added them to my files or my prompt. This seemed to really advance the design forward. i need to find a better way to update the files. This design was not easy! I want to thank @Shirsho Guha @Marcos Accioly and all others for the feedback on the my other post for my first website. You can see the old version and 1st post about it here: https://www.skool.com/quantum-quill-lyceum-1116/project-2-personal-website-sheesh?p=8319ce31 Definitely learned alot and look forward to adjusting for better workflow and designs. V5 with HERO DESIGN https://www.koachkev.io/

Decided to put some faces and names behind my agents after about 4 weeks in my folder structure. I'm about 3000 files deep and 300mb of markdown and text files. I've still only onboarded about half my clients into the system. I needed a way to visualize what had been built already and where my orchestrator was sending tasks. About half the agents have soul.md built into their instructions as well. This has been a fun project. https://collideascope.co/team/ai-team-roster.html Curious if anyone else has done anything similar.

Twenty-four days ago I posted about Jake's folder system video. This is what happened next. Same foundation — markdown files, orchestration prompts, clear roles. I just kept building. Fifteen named specialists. Each one with a soul file, guardrails, and a playbook. Duke orchestrates. Cash writes. Trace pulls the data. Hank runs the financials. Clint handles the MCP integrations. Behind each one is either a human counterpart doing the real work alongside them — or a role I can't afford to hire yet. Katie who's been with me for 18 years, now has her own orchestrator running the same system. Twenty-seven client folders. Twelve live MCP integrations. One shared repo. The folder system isn't replacing my agency. It becoming my agency. Jake gave me the unlock. This is how it's going.

This took a few weeks. Not building. Training. Tweaking. Breaking. Locking. Running the same flows over and over until the architecture stopped bending. Everyone here knows ICM. What this is… is what happens when you actually live inside it long enough. Not theory. Not clean diagrams. Real load. A few things only showed up under pressure: - The moment where orchestrator wants to execute… and you don’t let it - The cost of letting workers “figure things out” vs forcing briefs to be exact - How fast token bloat creeps in when you don’t treat load surface as a constraint - The difference between a rule you wrote… and a rule you had to write three times At some point, things flipped. The system stopped feeling like something I was managing… and started feeling like something that was holding shape on its own. That’s when the real work began. What’s in here is not “a good setup”. It’s what survived: - multiple passes of weekly audits - repeated cold starts - real production friction - and a lot of “this felt right but didn’t hold” A few things I’d pay attention to if you explore it: - Where boundaries are enforced (not suggested) - What got locked into rules vs left flexible - How briefs are treated as contracts, not prompts - How little the orchestrator actually does Also interesting: The extraction itself. That process alone shows you what was structural… and what was just personal preference. → https://github.com/NFTYoginis/creator-orchestrator-template If you’re deep into ICM already, you’ll see where this goes. Curious what breaks for you — or what holds better than expected.

*** Disclaimer -- This Hermes + Cognee stacking talk is advanced, highly experimental mad-scientist $hit. My setups & ideas are garage tinkering — “duct-tape two beasts and see what explodes” energy. Not production gospel. Can break gloriously. Pearl of wisdom: Master Jake’s ICM method first. Nail prompt and context engineering before touching any of this. If your agent goes rogue, don’t come looking for me. *** ...now with that out of the way ;-) Hermes already ships one of the strongest native memory and learning loops you’ll find in any open-source agent. It doesn’t just “remember” — it actively extracts skills from every task, curates facts into clean memory files, builds procedural superpowers, and gets sharper at your workflow with every run. That’s not marketing fluff. It’s baked in and it works. So the question lands hard: if Hermes already learns over time, why the hell would anyone stack Cognee on top? Because they solve two completely different problems — and together they turn a sharp apprentice into something that feels superhuman. Hermes’ native memory is agent-first and ruthless about efficiency. It keeps context tight, avoids token bloat, runs self-improvement cycles, and turns one-off tasks into repeatable skills. It’s designed to make the agent better at acting for you, session after session, without needing a PhD in memory engineering. Perfect for most day-to-day work. Cognee is a full knowledge engine. It’s built for the stuff Hermes intentionally stays light on: ingesting messy data at scale, structuring it into graphs and ontologies, spotting hidden connections across projects, resolving contradictions, and creating a stable, semantic long-term substrate that multiple agents or tools can actually share. One is the brain that learns by doing. The other is the encyclopedic library that never forgets context and connects dots you didn’t even know existed. They don’t fight. They layer. Hermes recently added clean modular memory providers. That means you can plug Cognee in as the heavyweight backend for deep graph recall and cross-session intelligence while keeping Hermes’ native layers (prompt memory, skill curation, fast SQLite search) for speed and autonomy. No messy overlap. No retrieval wars. Just smarter routing.