Hey Clief Notes fam! 👋 I'm Andre Izarra and I'm stoked to be here. I've been learning the past month about ICM created some simple workspace and projects and find this like an Eureka moment. 🙋 A little about me: I've been working in software development the last 5+ years, most of the time as freelance and this last year and a half I've been dedicating a lot of time to IA, learning the tools, ways to use it better and to integrate well LLMs in products. 🎯 My current goal: Learn a way to actually evaluate the accuracy of the ICM 💪 What I'm currently building/working on: The biggest use I have for ICM is a project for my whole work pipeline. I have a workspace for every mayor part of my process, manage conversations to clients until a deal, actually build the software, create the study cases with metrics, create content (starting with this thanks to this community) and repeat. 🤔 My biggest struggle or question right now: I notice the improvements working with agents following ICM but I can't tell exactly how to improve the structure, sometimes I notice problems with the results and I kinda can spot the problems try something different and usually improves, there were few times that got worse but I want as I said a way to actually evaluate the results to find a clear path to improve the structure. Let's get it! 🚀

Thank you Jake Van Clief and David McDermott for having me, and David Vogel for the Hermes assist. This one is ICM in practice. I open my own Hermes and walk every layer of its memory, live: the built-in memory files, session history, skills and the curator, the memory provider I host myself, and the ICM workspace that holds the job itself. The part I think this room will care about: in my workspace, every folder carries a small AGENTS.md router that points to a CONTEXT.md contract. Hermes loads AGENTS.md on its own. It doesn't load CONTEXT.md, so the router is what makes sure the contract gets read. Small router, real instructions 1 step away. Watch it here: https://www.youtube.com/watch?v=GtROa4PfvY8 ICM is why this system holds together, and Hermes is what walks it. Jordan The read through https://agentsasfolders.ai/blog/ai-agent-memory-hermes/

Who I am: Jeff van Leenen, Calgary, Alberta. What I actually do: I'm an Executive at a residential home builder. My job is how the work gets done as the company grows. I was a high school science teacher before this, and it still shows. What I'm building with AI right now: Four things... 1. Our company file system. I'm restructuring it into ICM so it's ready for our staff and AI to actually work from. 2. Teaching our staff about AI and how to use it safely on their own jobs. 3. Folder systems for trade businesses. The first one is for a painter who works for builders. The purchase orders he gets paid on are often wrong and nobody checks them, so I'm building a tool that checks each PO against his price list and what he measured on site. 4. The Idea Engine. It plugs into Claude, saves the ideas I float while I'm working on something else, and brings an old one back when it fits what I'm doing. It's in testing on my own machine. ...plus more... What I'm looking for connection-wise: - Someone who's solved discovery with trades owners. I want to get better at finding the real pain quickly, before anyone talks about tools. - Trading workflows and systems, especially anything built for builders or trades. Best way to reach me: DM here, or comment below.

I keep seeing the same technologies in AI engineering roles: LangChain, LangGraph, RAG, vector databases and agent orchestration. They solve real problems. After building extensively with ICM, I've started wondering how often we introduce that infrastructure before deciding whether the problem actually requires it. ICM approaches the problem from the context side. A workflow is divided into stages. Each stage has one job, known inputs, expected outputs and defined context. The agent reads what that stage requires and produces an artifact that the next stage can work from. Context selection becomes part of the design. That distinction has changed how I think about RAG. RAG is extremely useful when the system cannot know in advance which information will be relevant. If someone can ask arbitrary questions across thousands of changing documents, retrieval has to determine what information should reach the model. That creates a second problem alongside the original AI task: retrieval itself has to work well. Documents need to be chunked. Search and ranking need to return the right material. Retrieval quality has to be evaluated. If an answer is wrong, you may need to determine whether the model reasoned poorly or whether it never received the right information in the first place. There are many applications where that complexity is justified. But consider a structured workflow where the current stage already tells you which references are relevant. If I'm performing a security review and the workflow already defines the security requirements, architecture references and files that should be inspected, I'm not convinced semantic retrieval should automatically sit between the agent and that information. The architecture already knows what context is required. In that situation, retrieving the "most relevant" context at runtime may be less useful than explicitly providing the correct context by design. The failure modes change too. With retrieval, a poor result can come from chunking, indexing, query formulation, ranking or reasoning.

I’m building out an ICM architecture for a PE firm catered towards making due diligence easier and more efficient, I wanted to reach out and see if anyone here is in that space or worked close to it and even any tips you’d have for the correct folder structure/processes. I know a bit but knowing more on your guys’s process would be unreal thanks!