I'm a dietitian (clinical nutritionist) and practice owner, but (mostly) not a coder. The capstone project is a telehealth clinic with no out-of-pocket cost to clients, and the clinical operating system underneath it. Eduba super kindly gave the green light on a shared seat: my spot, two of us in it, and I pay for it. So I'm looking for a technical person to join me! You'll be creating a product that solves very real, expensive pain, and deploys on my live practice. The full brief is linked here, and I tried to make it comprehensive, interactive and somewhat skim-readable, but attached is also the straight-up html file to just feed to AI instead. Or just ask me questions in the comments. Also for those who aren't necessarily interested in joining me, I would LOVE your feedback/questions/thoughts too. No need to be diplomatic or hold back either ☺️ https://bffhealth.com.au/brief.html (PS- @Alex Brown I think I mentioned I'd tag you in what I was up to, here it is)

I've been diving deeper into dropshipping and learning something new every day. One thing I've realized is that finding products isn't the hard part—building consistent systems for marketing and customer acquisition is. For those who are already getting results: - What's been the biggest game changer for you? - AI tools? - Better product research? - Paid ads? - Organic content? I'm always looking to learn from people who have real experience. If you're open to sharing your journey or exchanging ideas, feel free to comment or send me a DM.

Some in this group with too much time on your hands and have had some illness or disorder for not just reading but also commenting on my posts, have her me say more than once that that Jake is the Alex Hormozi of AI. Third-row, foam-finger, "just one more video at 1 AM while the whole house is asleep" levels of fandom. Then the ACTUAL Alex Hormozi drops a video titled "Why AI won't make you rich in 2026." https://www.youtube.com/watch?v=WttZ6b-KUP4 I clicked in with my chest out, ready to defend the family honor. Nine minutes later I'm sitting there like a kid who just found out Santa outsources and the tooth fairy are just the people who spawned you. Because the man's right. Dangit. His whole case, no chaser: AI is leverage. Real leverage. But it didn't repeal every OTHER kind of leverage that was already minting money before ChatGPT learned to talk. Capital. Media. A team. A sales process that actually closes. An offer that doesn't stink. AI showing up didn't delete a single one of those. Then he drops the part that made me put the cigar and whiskey down. (La Aroma de Cuba Mi Amor and Noah's Mill in case you're wondering what to send me on my birthday). He asked his richest, oldest friends how much they personally touch AI in a day. Off camera, the honest answer was basically "bro, I don't." Their teams do. They don't. The guys with the yachts, gently breaking it to the AI guy that the robots aren't driving the boat. And this is where it kicks those of us who LIVE in this stuff square in the tokens. (I laugh at my own jokes, even when they're in writing, you gotta admit, this one was funny). He says most owners have AI-maxed their capacity, so now they crank out more, faster. Except the extra stuff they're automating was the low-priority junk. So we've built a gorgeous machine that does the wrong thing at the speed of light, and we're out here refreshing Stripe wondering why it's so quiet. His punchline is the one I want tattooed on the inside of my eyelids: the highest-leverage move in a business is making good decisions. Higher leverage than AI. Deciding what NOT to do. He said it's more efficient to decide something isn't a priority than to automate something that isn't a priority. Go ahead, read that twice. I did, out loud, to nobody in particular. And here it is again in case you didn't actually go back to read it.

I explain the details about it on today's high tea . I built a Claude skill that turns a process, an idea, or a messy folder into an ICM workspace. The folder structure does the orchestration. Numbered folders carry the order, the hierarchy scopes context, and plain markdown files hold state. One agent walks the right files at the right time and does the work a multi-agent setup would. Repo: github.com/RinDig/icm-architect 📦 What it does Two modes. 🔨 Build. You describe your work and it pulls out the structure already sitting in how you talk about it. The stages, the points where you stop and check, what stays the same every run versus what is new. Then it picks one of five proven forms and scaffolds the smallest workspace that carries the job. ♻️ Restructure. Point it at a folder, repo, or vault you already have. It reads every file, sorts each one by role, shows you a migration map, waits for your yes, then moves and checks the result. 🧩 The five forms Pipeline, umbrella, record library, knowledge bundle, context map. They mix and nest, so most real workspaces use more than one. ✅ The walk test Every result gets checked cold. An agent with no memory has to open the root, find its way, act, and report status from the files alone. If it can't, the structure gets fixed until it can. ⚙️ How to use it You can honestly just tell claude to download it from the link, but if you're using codex or something else it will just have to restructure the claude.md to agents.md Or if you want to do more Hands-On install Claude Code: drop the folder in ~/.claude/skills/icm-architect/, then say "ICM this" or "build me a workspace for X." Claude apps: zip the folder and upload it under Settings, then Capabilities. Fork it, break it, tell me what you built. 👇

Back in June I tore my whole setup down and rebuilt it. Everything had been living under one folder called AI project: eight or more dated copies of my second brain from a nervous-backup habit, plus a 96 GB junk drawer holding my real coding work, about twenty Remotion projects, and my business documents all together. Disk was never the issue, I had 752 GB free. I just couldn't find anything. So when I ran ICM Architect against it, I wasn't testing the tool against a blank page. I was testing it against a pattern I'd already rebuilt into once and sanded down since. It beat that design in four places. The harness mattered more than the tool, so here it is: I walled it off in a folder outside my workspace, installed the skill project-scoped instead of global so it couldn't leak into every session, wrote down twelve yes/no predictions before running anything, and then designed the workspace myself from the same interview answers and sealed that file before the skill ever ran. What it caught that I'd missed: my real unit of work is a thread of thought, not a video. My Short had gated my long rather than the other way round, and it put that in frontmatter because a folder name can't express an inverted dependency. And it refused to invent six numbers I'd never actually stated, where my own control cheerfully filled them in. It also failed at the thing I most wanted fixed, and that failure was the most useful part. One correction to the repo, from Jake's afternoon tea session: start from a pain, not a plan. I asked myself "what repeats" first and got my idealized pipeline back. Asked "what's most annoying" instead and got a completely different, much more useful answer. That reorder changed the whole output. A tool that beats your first draft isn't interesting. One that beats your third is telling you something. Be willing to rebuild, then be willing to have the rebuild corrected. Full write-up attached, around 2,700 words: what I already had, where the umbrella form turned out to be something I'd built without knowing its name, the compare-and-contrast, what it couldn't do, and the five things I'm keeping regardless. Feed it to your AI if that's easier.