This comes straight from the ICM method (I asked Claude to explain to me), so none of it is my invention. But seeing the five forms written out as plain shapes made the whole system land for me, and I'm actively building with it right now, so I wanted to share it the way it clicked. Every ICM workspace is one of five shapes, or a mix of them: 1. Pipeline. Same steps, run over and over, new deliverable each time. Think of a recurring 30-page client document: data in, checks run, sections drafted, reviewed, published. One folder per step, numbered in order. 2. Umbrella. Several different workflows that share one brand, voice, and reference layer. One shared folder holds the facts (brand guide, company snapshot, templates) and every workflow points at it instead of keeping its own copy. One home per fact. 3. Record library. The unit is not a run, it is a thing that accumulates: a client, a lead, a property, a patient. One folder per record, growing over time. The trap here is tracking status by moving files between folders. A status log inside the record is cleaner. 4. Knowledge bundle. The product IS the organized knowledge. A wiki, a second brain, course notes. The value is that a person or an agent can walk it and find things. 5. Context map. The subject is a whole organization. Teams, processes, where everything lives, and a small routing file at the top that answers "where does X go" in one read. The part that surprised me: they compose. My real setup is a context map at the top, holding an umbrella of shared facts, containing pipelines, publishing into record libraries. That's not breaking the rules. That's the normal case. One warning from experience: don't build folders for work you might do someday. I once built a beautifully organized workspace for future tasks, and it sat empty forever because nothing real ever flowed through it. Start with what "done" looks like (a phrase I picked up from @Ari Evergreen. If you can name the deliverable that comes out the other end, and you're already producing it the painful way, build the structure around that. That workspace gets used on day one, because the work is already there waiting for it.

@Jake Van Clief I build Confidential Information Memorandums (28-page investor documents) for my real estate development company. I had a genuinely good Claude Desktop skill for it — detailed instructions, voice rules, financial sanity checks, the works. And it still kept failing in the same place: lawyer edits. Every time counsel sent back changes and we asked Claude to fold them in, it would regenerate the whole document and quietly break things in unrelated sections. More instructions never fixed it. This week I rebuilt it as an ICM workspace instead, and here's the holistic shift: The skill asked the AI to be careful. The structure makes the mistakes impossible. - The document stopped being one generated blob and became ~40 component files (each table body and its footnotes are separate files). A lawyer edit now touches exactly one named file — the other 39 physically cannot drift, because the AI never reads them. - The financial math moved out of the AI's judgment into two small frozen scripts that recompute everything and fail loud. The AI runs them and reads the verdict; it's not allowed to edit them. - Legal text carries a VERBATIM marker that a checker enforces — counsel's words can't get silently rewritten. - Publishing requires an approval file only a human can create, and it gets spent on use — no stale approval can ever push out the next version. Same model, same knowledge. The 2,000 words of instructions got cut up and filed into folders with teeth. A rule in prose is a request; a rule in a file or script is a constraint. That's the whole lesson.

I built a system. It teaches AI how to build visual identities. It doesn't drift. Here's the story: I used to hand-build every visual identity myself. Client work. My own channel. Same routine every time. I explained the colors again. I explained the mood again. Nothing carried forward. Then I tried AI. Same brief, ten times. Ten different looks. One shot came out "noir". The next looked like a postcard. Same words in. Different result out. The model forgets. So I changed the approach. I stopped writing better prompts. I started building in stages. 3 stages. Each one builds on the last. Stage one locks the idenity. Color. Material. Type. The rules for how they combine. Stage two runs that identity through real scenes. A person. An object. A place. An idea. Same rules. New subjects. Stage three puts it in motion. It reveals. It connects. It transforms. Same materials. Same colors. Now it moves. Nothing gets re-explained. Each stage hands its work straight to the next. Drop them in Higgsfield. Prompt each scene. Full animation - no style drift. I tested it on a style I'd never planned to use. "Archival Investigative Montage." Ran it cold. It came out the same quality as everything else. That's the test that mattered. Not the polished demo. The one I didn't prepare for. Are you building multi-step AI systems yet? Or still prompting one shot at a time?

Hey y'all, not a sympathy post. This isn't me saying I'm taking time off for anything. In fact it's the opposite. My grandma passed away last week and I was cleaning out her apartment with my mother and talking to my aunt (both of them it was their mother). I can tell they're struggling. It's why I work so hard honestly on the software community on everything else. Almost none of the money goes to me. I give it to my family, my wife my friends. It's what makes me happy. I'm building a company specifically so my family doesn't have to struggle. Been helping out where I can but obviously been focusing a lot on the business. My uncle also has cancer and so my aunt has been taking care of him and also at the same time dealing with a death of her mother . Best way I can contribute has been through cash and some love here are there. With that though means I don't always have the time to do what I need to do. So honestly, I'm asking for your help if you're willing to donate to my aunt's GoFundMe on top of the help that I'm giving. I've just donated a bit and would love other people to help out. To me, if I work hard I can make sure that my whole family is taken care of. But in the small ways I can help where I can I will do my best too And that goes with asking for help from others. So if you all could help me out help them out. That would be amazing https://gofund.me/45b6655ba

Most "AI workflows" are one phase: type a prompt, hope. Mine has four. The build doesn't start until phase four. By then the AI is barely making decisions. It's executing a contract. Phase one: Brainstorm Open conversation. No structure, no acceptance criteria yet, no scope. I yap. Claude compresses what I yap into a `.md` file in real time. We argue. We rule things out. We name the actual outcome I'm chasing, not the thing I thought I wanted when I started talking. The output of phase one is one sentence: **what done looks like.** CRUSH started as "I want a video plugin that feels analogue." Two hours of brainstorm later the doc said: "14 Metal fragment shaders, a 6-slot effect chain, three global controls (DECIMATE, SPAZ, CHILL), real-time on Apple Silicon, drag-and-drop in DaVinci Resolve." That's the outcome that got dispatched. Phase two: Implementation plan Now I take the outcome and ask Claude to architect it. Files, dependencies, the order things get built in, the places it's likely to fall over. This is still in the main session. Still advisor seat. Zero code written. The plan for CRUSH was a 6-stage build pipeline. Format design, build-time generator, plugin core, 2D shaders, splat engine, AE port. Fourteen Metal fragment shaders authored as `.crush` JSON files. A Python codegen that writes the entire C++ OFX boilerplate at compile time so adding a new effect later is JSON plus a shader function and nothing else. Every stage had acceptance criteria, file scope, and risks named upfront. By the time it was done, the plan was a single markdown artefact a fresh worker could execute against without ever talking to me. That's the goal. The plan is the spec. The spec is the thing. The plan also names risks before any worker meets them. It says: "Metal kernel buffer names must match this exact contract or DaVinci silently rejects the bundle." It says: "Real-time on Apple Silicon at 4K is the gate. If a shader drops frames during scrub, it doesn't ship."