I'm assuming I'm not the only who's done this and please tell me if it's a bad idea, but I'm assuming we've all turned ICM structure into a Claude skill? If I've got a new project I run /init-max that I've built, and Claude generates folders and MD files based off questions it asks me. It's not allowed to assume answers unless I blatantly say I'm not sure. To build it I've just pointed Claude at module 3 in the foundations classroom and said "make this a skill". Has anyone done the same or is this a terrible idea?

EDIT: the use of copilot-instructions.md is very unreliable. Despite clear instructions to load the ICM workspace and read the setup.md at launch, 9 times out of 10 it fails with no meaningful error. Copilot itself tells me that these instructions are more like suggestions.🤣🤣 Instead I have built a skill in Copilot co-work and I have built an agent for the standard Copilot chat. Both of these will reliably launch into my ICM workspace. -------------------------------------------------------------------------------------------------------------------------------------------------------------- We've gone back and forth, here, on whether ICM can work inside Microsoft Copilot. I can now say it definitely does, and that Copilot Cowork is where it works best. I spent today rebuilding my own ICM workspace, currently in Claude, for Copilot — routing file, room context files, a voice file, branding, shared references — and it maps over almost one-to-one. The folders-and-markdown approach we use is almost exactly how Cowork is built to work. Your files live in OneDrive, Cowork reads them on demand, and you get the same context-on-demand behaviour we're used to. Two things make Cowork good for ICM, particularly in a business environment: - It reads your ICM folder tree straight from OneDrive, so you stop re-explaining yourself each session. - It has native access to Outlook, Teams, calendar and SharePoint, so rooms that interface with those apps or their data get easier - The adjustments were small: Copilot itself rewrote my Claude.md to a Setup.md, native M365 instead of connectors, and tightening the original markdown, which had grown, to keep token use down. NB: Copilot, to me, seems a smidgen slower than Claude (Yes, I have Opus 4.8 enabled in Copilot) Guide and blank template at RockfieldIT/icm-copilot-cowork: Implementing ICM in Microsoft Copilot Cowork

A few months back I picked up a book called Alchemy by Rory Sutherland. He is the vice chairman of Ogilvy, the advertising firm, and he has spent thirty years watching how people actually make decisions versus how engineers and economists think they make decisions. Spoiler: those two things have almost nothing in common. The whole book is built around one idea. People are not rational. They don't pick the best product. They don't respond to the most logical argument. They buy what feels right. What signals something. What fits the story they are already telling themselves. He calls it psycho-logic. The human mind does not run on logic any more than a horse runs on petrol. Here is why that matters right now. And why almost nobody building AI agents seems to have gotten the memo. The AI Agent World Is Built Entirely on Logic Go look at how AI agents are marketed and compared. Every benchmark is about capability. Token limits. Context windows. Reasoning scores. Speed. Cost per API call. The builder community follows the same track. Did it complete the task? Did it hallucinate? How many steps did it take? These are all logical questions. And if Sutherland is right, and I think he is, the entire industry is solving the wrong problem. Because here is what actually determines whether someone uses an AI agent or not. Whether they TRUST it. Whether it makes them feel capable rather than replaced. Whether the experience of using it feels right, not just whether the output technically is right. None of those things show up on a benchmark. Sutherland uses Red Bull as one of his favorite examples. By every logical measure it should have failed. Expensive. Small ugly can. Tastes bad. A logical analysis of the beverage market would have killed it before it launched. But Red Bull tapped into something psychological. The small can signals concentration and potency. The bad taste signals that it is medicine, that it is working. None of this is logical. ALL of it is real. The AI agent equivalent would be an agent that is not necessarily the top benchmark performer but makes the user feel like a superhero when they use it. One that builds confidence instead of anxiety. One where the interaction feels powerful, not just the output.

I'm looking for real-world experience here. I'm considering building a machine to serve a local LLM. I have a custom built machine with a Ryzen 5 and 32Gb of DDR5 sitting here unused for six months after a non-AI project was cancelled. Based on your own experience, what would I need to add to get decent local performance and what would be the best way to access remotely? What model should I use. I have a max budget of €3K additional

Found this GitHub repo that might be the official Fable 5 system prompt, reading through it is quite interesting. They made sure that Fable was as humane as possible. Here's the page if you're interested: https://github.com/elder-plinius/CL4R1T4S/blob/main/ANTHROPIC/CLAUDE-FABLE-5.md