I Built My Own Executive Assistant With Claude.
About 14 months ago, I tried to build this exact thing with ChatGPT.
I wanted a system that would clean up my files, brief me each morning, and act like an executive assistant who actually knew what was going on in my life.
We got close.
But we never finished.
I found myself stuck in a loop of scripts I didn't understand, with no clear sense of what I was even building. The tech wasn't ready to help a non-developer build a tool tailored to me.
My desktop is always a mess. My downloads are about as cluttered as it gets.
Two days ago, I sat down with Claude and completed my adaptive executive assistant.
It's not an app. It's a folder on my Mac, a few hundred lines of Python, and one terminal command. Here's how it works.
Step 1: It Cleaned My Hard Drive
I moved my digital assets to an external hard drive. (I too feel the memory shortage.)
My laptop is just the cockpit.
Then I built a tool that watches my Downloads, Desktop, and inbox folder.
It looks at every file, decides what it actually is using AI, and moves it to the right place. Daily. Via one terminal command: tidy.
Step 2: It Briefs Me Every Morning
I type brief.
A markdown file opens in Obsidian.
It shows me three things: what files are waiting for my attention, what loose ends I'm carrying across my projects, and a curated news feed filtered through my investment thesis.
It's not the news.
It's my news.
The headlines that matter to my work, with one line each on why.
Then I upload that brief into a fresh Claude conversation.
That's where the workday starts.
It knows what's on my plate.
We work from there.
Step 3: It Learns, It Adapts
End of day, I ask Claude for the debrief. What did we do, what stayed loose?
I save it.
Tomorrow's brief retrieves it.
The system has continuity - it remembers what happened yesterday.
What's Next
I'm wiring my Gmail and calendar into the morning brief. So when I sit down with my morning chai, I see what's on my plate before I access the web.
The system isn't done. It'll never be done. And that's the point.
Folks - I'm not a developer. I couldn't build this 14 months ago. And I built it in two days.
The Prompt - Copy This Into Claude
If you want to build something like this for yourself, paste the prompt below into a fresh Claude conversation. It's crafted to accommodate anyone.
I want to build a personal operating system that organizes my files, surfaces my priorities, and learns from how I work. I'll use you as my executive assistant. Help me build this iteratively.
Here's what I want, in order:
1. A FILE-SORTING SYSTEM. I want to designate one external drive (or one main folder) as the home for my real work. Then I want a tool on my computer that watches my Downloads, Desktop, and any inbox folder I designate, and automatically routes files to the right destination on the drive. The categories should be specific to my actual life — clients, projects, finances, reference material — not generic. The tool should use AI to classify each file. When something doesn't fit, it should ask me whether to create a new category. I want to trigger it with one terminal command.
2. A DAILY BRIEFING. Every morning I want to type one terminal command and have a markdown file open that shows me: what's waiting in my file inbox, what loose ends I'm carrying across my active projects (read from a markdown file I maintain), and a thesis-filtered news feed. The thesis is something I'll write — a private document describing what I care about, what I'm tracking, what triggers action, and what to ignore. The news filter reads my thesis and surfaces only what matters.
3. A CHAT-DEBRIEF LOOP. The morning brief is meant to be uploaded into a fresh Claude conversation as the daily seed. I'll work with you throughout the day in that chat. At end of day, I'll ask you for a debrief — what we worked on, what got decided, what stayed loose. I'll save that debrief as a markdown file. Tomorrow's brief should read it and surface yesterday's debrief at the top. Continuous thread, not isolated sessions.
4. EMAIL AND CALENDAR (later). I want to eventually integrate my email accounts and calendar so the brief shows me what's on my plate before I open any browser tab.
Before you write any code, ask me clarifying questions:
- What's my work / what are my projects / what are my priority buckets?
- What does my current file chaos actually look like — what folders, what types of files, what's the storage situation?
- What would I want my "thesis" to filter for — what news matters to me?
- What are my current "loose ends" — the unresolved things I'm carrying right now?
- What apps am I in (Mac/Windows/Linux, terminal comfort, markdown reader preference)?Don't build before you ask. The hard part isn't the code — it's getting the spec right. Once you understand my context, build the smallest useful version of step 1 first. We'll iterate from there.
I want this system to evolve over time. Propose changes; let me approve them. Don't drift silently.
Start by asking me your first round of clarifying questions.
That's it. Paste it into Claude. Answer the questions thoughtfully. Build the smallest piece first. And iterate.
If this resonated, reach out. I'm building this in public and I want to hear from people doing the same. Questions, follow-up topics, "I tried it and X happened" - all of it useful!