No content calendar. I write it when something works, or when something breaks badly enough to be worth the warning.
In a hurry? Every question this site answers, on one page. 80 of them, straight answers, each one linked back to the post it came from.
An ICM audit that only reads your files can't see the rot. The 4R audit runs my folders in a throwaway copy. Then the 4R audit plants real mistakes to test my guards. Here are the receipts from my own Sunday audit.
Page 1 of Google for "AI operating system" gave me a storage vendor, a research repo and an office building app. 6 results, 6 industries, 6 meanings. Not 1 written for somebody running a business. Here is the version that was missing.
My entire business is 9,916 files in 1 folder and Claude Code finds anything in it without me pasting context. The 3 parts of Jake Van Clief's ICM folder structure, why it doesn't rot in 3 weeks, and how 5 businesses share 1 tree. 41 folders at the top.
Anthropic shipped smart reports and shared skills to Team and Enterprise plans. A Claude Code skill is a folder with 1 text file in it, and you can make 1 in about 2 minutes without writing code. The 14 I run my business on, free, and what each 1 gets you.
Every video I publish becomes a Skool lesson, a landing page and an email, in under 10 minutes, off 1 folder and 1 prompt. The 4 things a lesson needs, the 6 files the package makes, and the 1 zip that does 3 jobs. 571 members in under 60 days.
A Marine sent me 3 screenshots. 4 AIs in 1 vault, 12 side copies stacked in Claude Code, 5 project folders. The map his own AI drew counted 43. The interview, the map, the proposal, the move, written out step by step, and the free starter folder that runs the first 2.
Cowork sees your files. It doesn't see the map, because it never goes looking the way Claude Code does. 1 line in the instructions box and a front door file fix it. I set it up live, muted the wrong mic for 3 minutes, and left it in.
My channel's entire catalog is 27 videos and all of them landed inside the last month. The bill was $26.76. This is the line that made them, the receipts off the console, and the video walking every stage of it.
I emailed 373 zip files in about 2 weeks. Every one of them froze the day it landed. So I pushed a paying client's folder system live instead, on camera. The upload failed 4 times before it worked and I left that in.
Anthropic put Claude inside QuickBooks, PayPal and HubSpot. Every setup guide on page 1 starts at the connector. None of them tell you what to file underneath it. Here are the 6 folders and the 2 text files, set up live for a gutter guy of 9 years.
My old cold email way was 3 tools and 4 days of waiting. This is me replacing it on camera with Lead Gen Jay's Consulti and the ICM folder system underneath, pulling 500 real practice owners while you watch. It crashes. I left that in.
A hook makes something happen at a fixed moment, whether the AI remembered it or not. What they are, the 2 running on my own business, and the narrow case where an agency owner actually needs one.
Every list mixes 2 completely different things together and never tells you which is which. The ones that ship with the tool, the ones you write yourself, and how to know which of the 3 you actually need.
This is the feature the $5,000 course is built on. What a subagent really does, the 1 job worth using one for, and why a roster of 7 helpers falls apart 3 months in.
A skill is a folder holding instructions for 1 job, opened when the job matches. Nothing to install, no code to write. What they are, the best practices, and why the part of Claude Code built for developers works best for people who are not.
People search for the memory switch the day after it forgot what you were doing. There's no switch. The memory is text files read at the start of every session, and that's better news, because a text file is something you can open and fix.
Settings, skills, shortcuts, helpers and memory, each one named in job terms. There are 2 copies and the split matters. A business owner needs exactly 2 of the things inside it, and skipping the rest costs you nothing.
Every best-of list is developer tools, because developers write the lists. This one is the 6 skills that write headlines, sales letters, blog posts, niche research, videos and pricing. All free, all text files you can open and make yours.
A skill is a folder of instructions for 1 job. A plugin is the box the folder can arrive in. The difference in plain language, and the honest order for a business owner: filing first, skills second, plugins last.
Everybody leads with a system nobody has heard of, and the DMs die. The play that works: fix 1 pain the owner already knows about, deliver it as files and folders, and let the client ask you what that was. With my first 30 days of receipts.
A CLAUDE.md file is a plain text note that sits in your folder and gets read before your AI does anything else. It is not code, it is not a setting, and there is nothing to install. It is a note on the front of the drawer.
Every rule below cost me something before it became a rule. My map file has lived between 77 and 137 lines across every version I have ever saved, and the versions that hurt were always the long ones.
Here is the file. Copy it, change the words, done. The plain version first and the explanation after, because you came here for a file and not for an article.
2 different questions get asked with the same words. One is where Claude keeps its own files on your computer. The other is how you lay out your folders so it finds your work. Here are both.
A rule your AI ignores is worse than no rule, because you stop checking. Most people fix that by making a rules folder. I did too. It did not work, and the reason it did not work is the useful part.
Interpretable Context Methodology is 3 serious words for a filing cabinet. You put plain text files in ordinary folders, arranged so an AI can open one, read where it is, and know what to do next. That is the idea. Everything else is detail.
The folders that are on my machine right now, what goes in each one, and the mistake I made in every single one of them before it settled. Not a diagram of an idea.
I did not invent this and I do not sell it. Jake did, I pay for his group every month, and I run it across several businesses. That is neither the author nor a bystander, which is the seat almost nobody writing about this is sitting in.
Interpretable Context Methodology: Folder Structure as Agentic Architecture, put up on 17 March 2026. I am not a researcher and I run the thing it describes every day. Here is what it says without the language that makes people close the tab.
Somebody is going to tell you this is common sense with a fancy name stapled to it. I want to make that case as strongly as I can before I answer it, because a page that only argues one side is an advertisement.
It asked something you answered an hour ago. It rewrote a thing you already told it was finished. Nothing is broken, you probably do not need to upgrade anything, and the cause is boring in a way that is good news for you.
You can build the empty version yourself tonight and I would rather you did that than paid anybody $5,000 for it. What I send instead is that shape with a real project already in it, because an empty template teaches you nothing.
I cannot write a line of code. I could not in 2006 and I cannot now. The reason telling an AI what I want works is not talent, it is that the files are arranged so it knows where it is. Every piece of technical language taken out.
These 3 get argued about as if you have to choose. You do not. They fix 3 different faults, and almost everybody who is frustrated right now is paying to fix the one they do not have.
Every page that ranks for this wants to rent you an AI receptionist. None of them tell you how to build them for other people. I run 2 of them on the identical 6 folders, and going from the first to the second was 3 text files. That gap is the business.
Somebody called him a charlatan in my comments. I pay for his group every month and nobody made me, so I am probably the right person to answer it. Here are the receipts. Then here is the folder system running on my own screen with no cuts.
Most folder structure advice assumes you run one thing. Agencies never run one thing. Here is how to split brands so they stop bleeding into each other. Here is the 5 minute audit that catches it when they start. And here is the drift that audit found in my own setup the day I wrote this.
Context engineering is deciding what information your AI has in front of it at the moment it does a job. That is the entire concept. Everything written about it after that sentence is either implementation detail for developers, or somebody selling you a framework for a filing problem.
The ICM Folder System is Interpretable Context Methodology set up as plain text files in ordinary folders. One file that is always read first. One file that says what you are working on. Folders holding the work. No platform, no subscription, nothing to install.
Here is the folder structure I run several businesses on. One file that is always read first. One file that says what we are doing now. 4 folders holding the material. The map is 11 lines and it does no work at all, which is the part people get wrong.
Prompt engineering is what you say to the AI. Context engineering is what the AI can see while you say it. That is the distinction. Once you have it, most of the advice you were given sorts itself into useful and useless.
Anthropic, the company that makes Claude, has published its own guidance on context engineering for agents. It is good and it is written for people building software. I am not one of those. Here is what it actually says, and what changes if your job is running a business that now has an AI in it.
Agentic context engineering is deciding what an AI can see when it is taking actions rather than just answering a question. The stakes go up, because a wrong answer is annoying and a wrong action costs you a client. Almost every guide solves this with more agents. That is the part I disagree with.
AI agent memory is the ability to carry information across sessions so the agent does not start from zero every time. It is a real capability and there are real tools for it. It is also the thing people buy when the actual problem is that nothing was ever written down anywhere the agent could look.
When the context window fills, the oldest material drops out of view. The model does not announce it. It cannot tell you what it lost. It keeps answering in the same confident voice it used when it had everything. That gap between confidence and completeness is the problem.
Put something new down and something already there slides off the edge. Nobody announces it. That is why your agent "forgot", why it starts making things up late in a session, and why a bigger window is not the fix.
Every course under that name teaches a developer to build a system. If you got handed the AI and you cannot write code, you need a filing system, not a certificate. Here is the free one.
2006, a secondhand tower, and a typo in the headline every single time. 20 years later I run several businesses on the same 2 text files. Includes the video I made in my first 30 days, which is rough, and I am leaving it up anyway.
If you landed on the blog first, the thing you actually want is the free folder system. Not an empty folder, a website I already started that you finish. The plan written, the homepage half written, and the next job in the desk file.
Get the folder system free