I gave Opus 4.8 all the markdown files that Fable and I had created up to that point, in hopes that Opus could help me plan the model switch. And the reason Fable was so good, other than that he is the best coder on the planet, is Jake's ICM system, which I learned and built in this Skool class. The Foundation Fable built upon. My final prompt to Opus (we had a good conversation before this, and I saved my Fable context window with /export, so if need be and Fable returns, we can restart where we left off): -------------- Fable 5 was writing to my Obsidian Vault, and he has a Website folder with 1. Audit (the rebrand and rebuild was coming from a Replit project I built. So I gave Fable all my code from GitHub, and the Audit file is pertaining to the migration-audit) 2. Brand 3. PRD files 4. Context.md 5. ia-nav-structure 6. launch-action-plan 7. Would you like to read these files before helping me with a handoff.md file? 8. (he said yes) Opus: I've read all ten. This is unusually disciplined work — the layered ICM system, the keep/rework/kill audit, locked IA, five approved PRDs, and a staged runbook. A few things stand out that actually change how the handoff should be built: The vault is the plan; the build repo is the truth. Your real build lives in port-a-insider with its own CLAUDE.md tracking milestones M0–M7. The vault tells Opus what to build; the repo + its git history tell it what's already built. So the handoff's job isn't to re-summarize the plan you've already documented beautifully — it's to bridge Opus into that repo and tell it how to figure out where Fable stopped. Your brand-voice risk is lower than I'd feared. The insider-messaging doc is genuinely thorough — hero three-beat, the generous-club principle, the extended never-list, the Dispatch format, Porter's posture. That's the kind of concrete voice spec that survives a model switch well. The continuity risk here is build-state, not voice.

Can someone point me to a good primer on Pi harness? I can follow the breadcrumb trail after that… :-)

I wanted to mark this one out loud. @Bas Rosario is the first member who isn't an admin to reach Level 7 in here. He earned every bit of it. If you've spent any time in the forums, you've run into Bas. He's the one greeting new people on the welcome board before anyone else gets there. He's the one leaving real feedback on your work, not a thumbs-up and a move on. And he teaches by meeting you exactly where you are. He's teaching his wife, a cosmetologist with no tech background, to think in systems. He explained ICM to her with a birthday cake. A cake is a system, he said. It has steps. You gather the ingredients, you preheat the oven, and you do them in order. That's what ICM does with the AI. It hands it one step at a time. That's the gift. He can take the thing some of us are still wrestling with and make it small enough to hold. And when you try to hand him the credit, he won't take it. On Saturday's call, the room joked that he's the new authority around here. He shot it down on the spot and pointed at the people he learned it from, naming them one by one. He builds things to route you to the right help, not to put himself at the center of it. He does this constantly. Not in bursts when he wants attention. Every day, in the background, for people who didn't ask and couldn't repay him. That's the whole thing. Level 7 isn't a reward for showing up. It's what showing up for other people looks like after you've done it for this long. And no, he's not a bot. I know the running joke. The steadiness almost makes the case for it, in a room full of people building AI. But that's the tell. Bas has said he shows up here every day on purpose, out of genuine gratitude for what this place has given him. A bot can post every day. It can't be grateful. Bas comes back because he is. That's the most human thing in here. Bas, thank you. This community is better because you're in it. If Bas reviewed your work or helped you when you were stuck, say so below. He should hear it.

This will tell you a lot about the future and the way I think, give it a read if you have time. https://jakevanclief.substack.com/p/the-machine-is-smart

There is a method of using Chinese models like: Qwen, MiniMax, Zai, Moonshot, etc. That are hosted in the US on the latest NVIDIA B300 datacenter GPUs, where your data stays private and is never trained on. This is where Ollama, the self hosted go to source for local models comes in. Ollama offers a selection of cloud based models. Simply append your model call with :cloud https://ollama.com/search?c=cloud&o=newest What's your favorite model?