Day 14: Write the two files that make AI smarter forever.
Twenty minutes today. From tomorrow on, Claude opens your folder and instantly knows who you are and how to behave. The spine.
Welcome to Day 14 of the challenge.
This is the most important day of the challenge so far.
It’s also one of the days I’ve been most nervous about you reading. Not because it’s hard. It isn’t. Twenty minutes of doing, ten minutes of reading. But because everything so far has been pitched at “AI is now your colleague” and today is the day we cross into “AI is now your colleague who knows you.”
Two text files. One paragraph pasted into a settings menu. That’s it.
From tomorrow on, every Claude session you open inside your Switched-On folder reads those two files first and behaves as if it already knows you. It does, because you’ll have told it.
A word before we start
Two reassurances before the technical bits.
One: this is not technical. The files are .md text files. The contents are plain English. The setting is a paragraph in a settings menu. If you can write a paragraph and copy-paste a sentence, you can do today’s task.
Two: it persists. From tomorrow, Claude opens your folder and reads these files automatically. You don’t have to remember to do anything. You don’t have to upload anything. You set this up once and the benefit compounds across every chat you ever have.
It looks like infrastructure because it persists. It’s actually just text files in a folder.
Why .md? Why these specific filenames?
Quick aside before we begin.
The files we’re creating end in .md, which stands for markdown. It’s a plain text format that lets you add light structure (headings, bold, bullets) without locking the file into Word or any specific app.
Three reasons it matters:
Future-proof. Any AI can read markdown. Any text editor can open it. Ten years from now, the brain you build today still works.
Lightweight. No formatting bloat. No fonts. No corruption. Just words and a few asterisks for emphasis.
AI-native. AI models are trained heavily on markdown. They read it better than Word documents or PDFs.
The specific filenames matter too. CLAUDE.md is a convention Claude is trained to look for. Put a file called CLAUDE.md at the top of a folder, and Claude reads it automatically every time it opens that folder. Same for MEMORY.md.
That’s literally the whole magic. Name the file correctly, put it in the folder, Claude does the rest.
What this folder will become
Right now your Switched-On folder has the three sub-folders from yesterday plus a handful of outputs from the past two weeks. By Day 30 it’ll have:
Eight foundational files that capture who you are, how you sound, who you serve, what you sell and what you’re known for
A working agent skill that runs on a schedule
A folder of dozens of outputs from work Claude has done for you
A running log of decisions you’ve made with Claude as your thinking partner
The two files we create today (CLAUDE.md and MEMORY.md) are what makes everything else possible.
Twenty minutes well spent.
Step by step
Step 1. Open Cowork in your desktop app, pointed at your Switched-On folder. Opus for this one.
Step 2. Paste this brief:
Create a file at the top of this folder called CLAUDE.md. The purpose of CLAUDE.md is to be the master brief that you (Claude) read at the start of every session opened from this folder.
The file should include:
1. Who I am, in 3-5 lines (you can ask me). 2. What this folder is for (Switched-On, my personal AI workspace). 3. How to behave in this folder. Specifically: use Australian / British English spelling. No em dashes. No corporate jargon. Be direct. Push back when I’m wrong. Ask clarifying questions before generating anything substantial. Save outputs to outputs/ using YYYY-MM-DD_description.md naming. 4. Where to look for things: outputs/, skills/, reference/. 5. A note that MEMORY.md is the running log of what we’ve learned together, and you should read it too at the start of every session.
Ask me the questions you need to write section 1 (about me). Then draft the whole thing. Show me before saving. I’ll edit.
Step 3. Answer Claude’s questions about you. Don’t overthink. Sentences not paragraphs. Use Wispr Flow if it’s faster to talk than type.
Step 4. Read the draft. Edit anything that doesn’t sound like you. Approve.
Step 5. Now MEMORY.md. Paste this:
Now create a file at the top of this folder called MEMORY.md. The purpose of MEMORY.md is to be a running log of what we’ve learned together across sessions. Pattern preferences. Corrections I’ve made. Things I’ve told you to remember. Decisions I’ve made that I want you to remember next time.
Start with three example entries based on what we’ve worked on across the past two weeks. Use this format: - YYYY-MM-DD: [what was learned, in one sentence]
Going forward, you’ll add to this file whenever I tell you to “remember that” or when you notice a pattern worth recording.
Step 6. Read the three starter entries. Add anything obvious you’d want to remember from the past two weeks. Save.
Step 7. Set the global instruction. Open Claude desktop settings. Find Custom Instructions (or System Prompt or Behaviour, depending on which version you’re on).
Paste this paragraph:
When opening any folder containing a CLAUDE.md file, read CLAUDE.md and MEMORY.md first before doing anything else. Treat their contents as instructions for how to behave in that folder. Confirm at the start of the session that you’ve read them.
Step 8. Test it. Close Cowork. Reopen. Open your Switched-On folder. Type:
What do you know about me and how I work? Summarise what you’ve read in CLAUDE.md and MEMORY.md.
Claude should come back with a clean summary. If it does, the spine is in. If it doesn’t, the section below covers what to check.
If Step 8 didn’t work
Four things to try, in order:
The filename. Make sure it’s
CLAUDE.mdand notCLAUDE.md.txt. Mac TextEdit sometimes adds the .txt extension silently. In Finder, the .md extension should be visible at the end of the filename.Close and reopen Cowork. The global instruction only takes effect on new sessions started after you saved the setting.
The settings menu might be called something else. Different Cowork versions call it Custom Instructions, Personal Settings, or Behaviour. Try all three.
Curly quotes vs straight quotes. If the pasted instruction in your settings menu shows curly quotes rather than straight ones, retype the quote characters directly.
If none of the four work, hit reply and tell me what’s happening. The fix is almost always one of these.
Two weeks in. How’s it landing?
Quick favour before the weekend.
This is the end of Week 2. Two weeks ago, AI was a browser tab and a chat window. Now you’ve installed Cowork, connected two services, built a dashboard, killed a few Sunday Scaries, read a document in five minutes, decided something you’d been weighing, and given Claude its first proper brief on who you are.
That’s not nothing.
I want to hear how it’s going. Three questions, if you’ve got a minute over the weekend:
What’s the moment from these two weeks you’d quote to a friend? The bit you’ve already told someone about.
What’s the thing you haven’t quite cracked? Where the friction is.
What’s one tool, habit or file you’d like to see covered in Week 3 or 4? I’m shaping the back half this weekend.
Hit reply. DM me on Instagram. Comment on the post. They all come to me. I read every one.
One last thing. The women getting most from this challenge are the ones who’ve told a friend. The first time you explain it to someone else is usually when it lands properly. If there’s a woman in your life who’d want this: nishma.co/30daychallenge.
A safety note
CLAUDE.md and MEMORY.md sit at the top of your folder and Claude reads them every session. Treat them like the brain you actually want Claude to have. Don’t write anything in them you wouldn’t be comfortable Claude using as context in any future chat.
Same backup discipline as yesterday. iCloud, Dropbox, whatever you use. The spine is now part of your professional setup.
What you’ll notice
The next chat in this folder feels different.
You don’t have to re-explain who you are. You don’t have to specify “use British English” or “save to outputs/” or “no em dashes.” Claude already knows. The first session after the spine is in is a small wow moment that compounds across every session after.
The other thing you’ll notice (probably tomorrow): you start using Claude for more things, because the cost of starting a new task has dropped. You’re not warming the model up. The model already knows.
Why this works
Most AI tools fail you because every session starts from zero. You re-explain who you are. You re-state your standards. You re-paste your style. Half of the prompt is set-up.
CLAUDE.md plus MEMORY.md plus a global instruction means the set-up is permanent. You write it once. It applies forever. Every prompt from here on is shorter and sharper, because the context is already loaded.
This is the gap between people who get a lot out of AI and people who don’t. It isn’t intelligence. It isn’t tool choice. It’s whether the AI knows you when you arrive. From today, yours does.
Key takeaway
Two text files at the top of your Switched-On folder. One paragraph in settings. From tomorrow, Claude opens this folder and instantly knows who you are, how to behave, and what you’ve learned together.
Twenty minutes today. Compounds across every session for the rest of your working life.
Where this is heading
Week 2 closes here. Week 3 begins tomorrow.
The first two weeks were about giving Claude work to do. Week 3 is about telling Claude who YOU are, properly. Six brain files across the week, each one a different facet of you and your business. Each one saved to the spine. Each one read automatically forever after.
It starts tomorrow with the most important one.
Tomorrow on Switched On
Day 15: Tell AI who you are, properly.
Saturday edition. Ten minutes of Claude interviewing you. Walk out with a one-pager that briefs any AI like a new hire. The first brain file. The one everything else hangs on.
Nishma x
P.S. How did today land? Three buttons, thirty seconds → Tap one
I read every response. The data shapes the rest of the challenge.
Catching up? Days 0 to 13 are on the Switched On Substack archive HERE.
Got a friend or a colleague who this might be useful for? Send her to nishma.co/30daychallenge




