°2: Digital Organizer - Used Claude Code to clean files and build a writing workspace
Two experiments with Claude Code: one partial success, one partial failure
This is part of my ongoing series about how AI shows up in my daily work. I share (in <1000 words) specific use cases where I used AI as a collaborator. To write each piece, I ask the AI four questions:
How did we collaborate together? → summarizes activity
What was my role versus your role in solving this? → delineates between human vs AI role
What made this interaction work well (or not work well)? → determine if interaction was collaborative, strategic, one-sided, transactional, etc.
What should I do differently next time? → suggestions for the future
I pair with AI to write these pieces, but I review, reflect, and add my own perspective throughout.
Claude Code - the better IDE?
Claude Code is heavily used by developers, but it’s being increasingly adopted by non-devs like myself for managing our “digital brains.” I’ve been using Cursor and VSCode for all my AI agentic IDE work, but have been dancing around using Claude Code for awhile because using the Terminal scared me. But, I discovered that you can use a Claude Code extension IN VSCode - so that’s my way in.
I have two problems that I wanted to tackle that I read about in Lenny’s post about Claude Code:
(1) Streamline my digital storage - in general Not scattered exactly - more like spread across Google Drive, Apple Cloud Storage, local Documents, and Downloads. Multiple copies of the same file living in different places.
(2) Streamline my writing process - specifically I want to write more consistently. I need a better way to store notes in a single location so I can actually pull from them instead of losing ideas in random folders.
The Process
Part 1: Cleaning up my digital storage
I opened Claude Code in my terminal and started with the most annoying problem: figuring out what was actually on my machine versus what was in cloud storage.
Prompt: Help me identify what files I have locally, what’s duplicated across Google Drive and Apple Cloud, and what’s old or unused that I can delete.
Claude scanned my system and found patterns I couldn’t see on my own:
Files with “copy” or “version 2” in the name scattered across locations
Projects from 2+ years ago I’d completely forgotten about
Downloads folder full of PDFs I’d never opened again
Too many files with strings of letters to count 😬
Claude asked questions like: “When did you last access this?” “Is this duplicated in your Google Drive?” “Do you still need this project?”
The result: Cleared out duplicates, archived old projects, and created a cleaner structure where I could actually find things.
Part 2: Building a writing workspace
Lenny’s post mentioned Helen Lee Kupp’s use case about “organizing scattered thoughts into coherent articles.” I found her writing workspace template and forked it.
What I found immediately useful:
Writing examples folder - A place to store my previous work so Claude can match my voice
Raw notes folder - A designated spot for messy, unorganized thoughts
But here’s where it gets tricky: my notes don’t live in one place.
They’re scattered across Apple Notes, Google Drive, voice memos on my phone, random docs. I tried to get Claude Code to read my Google Drive files locally, but it can’t actually access the content. The files show up as .gdoc pointers, but Claude can’t read them directly as is.
Yes, I know - a Google Drive MCP could solve this. But from what I can tell, there isn’t a plug-and-play version I can just install. I’d have to build one myself, which defeats the purpose of “quickly setting up a writing system.” [If you’re reading this and know otherwise - leave a comment!]
What’s actually missing: The processing of information
This workspace is helpful, but it isn’t solving my fundamental problem - how do I actually get information INTO this repo?
I need an explicit processing step. Maybe a daily 30-minute review where I sort through the day’s notes and move them into the raw notes folder. Something like Zettelkasten - the deliberate act of processing and organizing thoughts, not just dumping them.
And honestly? I should have this kind of processing practice anyway. The more I learn about LLMs, the more I realize: they’re only as good as the data you give them. Garbage in, garbage out. This is exactly why MCPs (Model Context Protocols) are so powerful - they give the model access to structured, controlled data.
More on this exploration to come.
Quick reflection on the collaboration
(from Claude, lightly edited for brevity):
What worked: I brought real problems and forced explanations instead of just accepting solutions. Claude provided pattern recognition and asked clarifying questions. Together we adapted a template to match my actual workflow.
What to try next: Experiment with a hybrid approach - draft in GitHub with Claude Code, polish in Google Docs. Stop trying to force one system.
So, what’s the point?
(Back to me now)
Designing systems for how one actually works vs how one “should” work is hard.
Even as I write this piece about building a writing workspace in Claude Code, I’m already fighting the urge to copy it into Google Docs for “real” editing. Old habits die hard. [oh, and I did copy into google docs - I couldn’t “think” in markdown.]
Here’s the reality: I live in the Google ecosystem. Everything I do is in Google Docs and Drive. Claude Code can see those files exist (they show up as .gdoc pointers with document IDs), but can’t actually read or write to them properly. The content comes out wonky.
So I’m stuck in this in-between space:
Draft in markdown using Claude Code?
Immediately copy to Google Docs for “real” editing?
Force myself to stay in GitHub to prove I can?
Build some complicated Google Drive MCP server so Claude Code can work with my docs?
I don’t have the answer yet. And honestly? I’ll probably end up doing exactly what I always do - copy this into Google Docs the moment I’m done. [I did.] Maybe that’s the hybrid workflow I actually need. Maybe the GitHub workspace is just for capturing and Claude collaboration, while Google Docs is where I polish and publish.
Or maybe I’ll surprise myself and actually use the system I built. [I didn’t.]
For now, I have a cleaner machine and a workspace that might work. Sometimes that’s enough to start - even if you’re not sure you’ll stick with it.
If you’re stuck in digital chaos or a consume-only cycle, Claude Code is worth trying. Start with a real problem you care about solving. Adapt what others have built to fit your brain. And be honest about what works - even if it’s different from what you expected.
‘Designing systems for how one actually works vs how one “should” work is hard.’
Exactly what a good PM knows and a new PM needs to learn. Forcing new user habits and behaviors is sooo tough.