The Folder System You've Been Putting Off? Cowork Will Build It For You.

🎯 What You'll (Hopefully) Get From This: A practical first step for building the three-layer folder architecture using nothing but Claude Cowork and plain English. No terminal, no code, no anxiety. Ultimately, speed to market and the ability to get out of 'analysis paralysis'.

Hope this helps!

---

Yesterday, I came across the latest episode of 'How I AI' with Claire Vo, with guest JJ Englert from Tenex. I'd encourage you to watch this after reading below. Link here: https://youtu.be/jwGQ9CrqVdA?si=k0jRTW1NxslxtqKz

Runtime: ~50 min (fine on 1.5x speed)

---

I've seen a few folks here note nervousness about the terminal window, using code, or claiming to not be developers or engineers but desperate to harness the power of this system and the orchestration capabilities of frontier models.

If that's been you, hopefully the video above and my notes below help.

There's a version of this system you can build today, through conversation, without opening a terminal once. It's called Claude Cowork (if not familiar), and I want to show you how to use it the right way, starting with your actual work, not the system itself.

As Jake mentioned in Foundations, it's one of 3 Anthropic UI's that you can use, and you'll need the desktop app to execute.

---

**Start with the job, not the structure.**

Here's the trap I see people fall into: they try to design the folder system before they understand what it needs to do. They grab a template, rename a few folders, and three weeks later it's collecting digital dust because it never actually matched how they work. Jake clearly warns about this, and you see this in various comments throughout the community.

So what? How can you mitigate this rabbit hole?

The smarter move is to start with the problem (personally, my favorite thing to talk about as a design thinking practitioner). What are the specific things you sit down and do on a Tuesday afternoon?

[Writing proposals. Reviewing deliverables. Drafting posts. Prepping for a client call. You can even go one step further and think about those 'anti-todo' items like draft emails, outline based on dictation, or schedule your calendar properly.

>TIP: If you still don't know where to start: go into the 'Cowork' tab, and there is an 'Ideas' tab in the right pane. Click it, and begin ideation.]

Those are your workspaces. The folder system is just how you make sure Claude has the right context for each one, without dragging everything else in with it.

Jake's Folder Organization Guide makes this explicit: each workspace should represent a different mental mode. If you think differently about two types of work, those are two different workspaces. Your writing brain and your execution brain aren't the same. Claude shouldn't treat them like they are.

>TIP: If you're a Premium Member here, you can also upload the 'Folder Organization Guide.md' file to Cowork to further give it context on how you'd like to approach your build. Explore 'The Vault'. Invaluable resource.

---

**What Cowork actually is.**

Cowork is a tab inside the Claude Desktop app. It gives Claude direct read and write access to a folder on your computer. It can execute multi-step tasks on your machine, create and organize files, and coordinate parallel work without you typing a single command in a terminal.

AKA - Point the 'Task' or 'Project' at the folder, go wild. Cowork can create this stuff for you.

The clearest way I've heard it described: regular Claude chat is prompt and response. Cowork is task delegation. You describe the outcome, Claude makes a plan, and you come back to finished files in your folder. You can step away while it works.

What I've found to be even cooler is that I can go on a walk, dictate my plan, come back to Claude and bang out a PRD (product requirement document) in chat, then hand it over to Claude Cowork and it just runs and creates, outlining the tasks in the UI and staying visually familiar to the chat function.

All around useful.

---

**How you use it to build your folder architecture.**

You point Cowork at a new, empty folder on your computer. Then you describe your work to it in plain language -- not the system you want to build, but the actual things you do.

Something like: "I run a consulting practice. I do three types of work: client deliverables, internal content creation, and business development. For client work I need a folder per client with intake notes, deliverables, and communications. For content, I write LinkedIn posts and long-form articles. For business development, I track outreach and proposals. Build me a folder structure for this, add a CLAUDE.md at the root that tells you how to navigate it, and add a CONTEXT.md inside each main folder that describes what that space is for."

Cowork builds the folder tree, writes your CLAUDE.md with a routing table, and creates starter context files in each workspace. It asks clarifying questions if it needs to. You review what it built, give it feedback in plain language, and it revises. The whole thing takes less time than manually creating the folders yourself.

What's cooler? You can even test and iterate in the same Project or Task, watching it work live.

---

**Where this maps to the three-layer system.**

This is repeat from the Foundations Course (go take this now if you have not... seriously), but in more Cowork terms.

The CLAUDE.md Cowork writes for you is Layer 1, the map. It tells Claude what the project is, where things live, and a routing table that says "for this type of task, go here and read this."

One thing I'd add to your project instructions that most people skip: tell Claude what you're NOT working on. Guardrails are vitally important than context in most cases to keep the LLM on track. It stops Claude from dragging in context you didn't ask for. Also a big session token saver if you're on the Pro plan and limited. If you're in your content workspace, Claude doesn't need to know anything about your client folders. Say so explicitly.

The CONTEXT.md files inside each workspace are Layer 2, the rooms. Each one describes what happens in that space, what good output looks like, and what the process is. Claude only loads a workspace context file when it's working in that workspace. That's how you keep the context window (the amount of information Claude can hold at one time) clean and relevant.

The skills you build over time are Layer 3. A skill is a markdown file (a plain text document) that teaches Claude how to do one specific thing well. DON'T OVERCOMPLICATE THIS. A writing style skill trained on your actual sent emails. A morning briefing skill that reads your Slack and calendar before you log in. Any process you find yourself re-explaining to Claude in every conversation is a skill waiting to be written. You can ask Cowork to build skill files for you as you identify those patterns.

>TIP: JJ speaks about Skills development explicitly in the shared video. Specifically to email and content creation. You already have rich data at your fingertips with the work you currently do. USE IT...

---

**The connector piece is worth knowing about.**

Cowork has one-click integrations with Gmail, Slack, Google Calendar, Notion, and others. You can ask it to analyze 30 days of your sent emails and produce a writing style guide based on your actual voice.

You can have it pull your calendar and build a prep document for every meeting this week. You can set a scheduled task that runs every weekday morning, reads your Slack and inbox, and drops a priority list into your workspace folder before you sit down.

That last one especially. Your system doing real work before you log in is not a small thing. Scheduled tasks adds a 'heartbeat' to your Cowork space (you know... the thing that everyone raves about with OpenClaw?) The powerful thing is that you've made it much more powerful by doing the pre-work above and what Jake teaches here.

---

**The one watch-out.**

Cowork has no memory between sessions, and every session starts clean. That's actually a feature, not a bug. It means your context files are doing the memory work, not a fragile conversation history. But it also means if you don't build the brain file (a document that tells Claude who you are, how you work, and what you sound like), every session starts from zero and nothing compounds.

The folder structure is the memory. Build it first.

---

**TL;DR: Your actual first step.**

Download Claude Desktop, if you haven't already. Open the Cowork tab. Create a new folder somewhere on your computer. Point Cowork at it. Describe your work, not the system you want, and ask it to build the architecture for you.

The three-layer system Jake teaches is the right system that is transferrable for decades to come, regardless of the new fad of model or orchestrator. The terminal window was never the point. If you feel it's been in your way, now it's not.

---

I hope this helps someone!!!

14 comments