From Obsidian to Claude Code: Building a Programmable Life

A few weeks ago, I came across a post on X. 700,000 views. 4,000+ bookmarks.

The author, Molly Cantillon, runs 8 Claude Code instances to manage her life: product, email, investments, health, writing. Every morning, AI auto-generates her brief. She hit Inbox Zero for the first time ever. AI even found $2,000 in subscriptions she didn’t know she was paying.

She wrote: “A panopticon still, but the tower belongs to you.”

After reading her post, I thought about it for a long time.

Not “wow, that’s impressive.” More like: why was she able to do this?

Three Principles

I started organizing my own system. The core isn’t “what tools to use”—it’s “how to choose.”

Three principles.

Principle 1: Programmable Infrastructure

What does programmable mean?

It means machines can read, write, and operate on it.

All my documents are in Markdown. Plain text, cross-platform, future-proof. More importantly, AI can read it directly.

When choosing services, the first thing I check is whether they have an API. No API means your data is trapped. You can use it, but you can’t take it with you.

That’s why I use Obsidian for my knowledge base, not Notion. Obsidian notes are local .md files. Notion notes live on their servers.

Self-hosting my website follows the same logic. My blog runs on Astro + Markdown. Claude Code can publish posts directly. No admin panel, no login, no copy-paste.

This isn’t technical perfectionism. It’s about enabling AI to do work for me.

Principle 2: Data-Driven Decisions

This is something Minerva taught me.

When I was at Minerva, every decision required asking: where’s the data? Without data, you’re guessing. With data, you’re deciding.

Sounds simple. But it’s hard in practice. Most people’s data is scattered everywhere. Google Analytics here, social engagement there, product metrics somewhere else.

Looking at dashboards? Too slow. Open five websites every day, stare at numbers, then what?

What I want: AI reviews everything and tells me “what needs attention.”

So I built a system called Content Discovery. It automatically scrapes trending content from Reddit, Hacker News, and X every day. AI analyzes, scores, and categorizes. Every morning, I receive a brief.

I don’t go find information. Information comes to me.

Principle 3: Automated Execution

Repetitive tasks shouldn’t be done manually.

Manual	Automated
Open 10 websites daily for news	Cron job auto-scrapes + AI summarizes
Manually publish blog posts	GitHub Actions auto-deploys
Manually compile data	Scripts run on schedule + generate reports

My rule:

Do it once: do it manually
Do it twice: consider automating
Do it three times or more: definitely automate

Time should be spent on decisions, not execution.

The System I’m Building

I don’t have Molly’s 8 Claude Code instances. But I’ve started building my own version.

Cockpit (Personal Command Center)

~/Cockpit/
├── daily/           <- Daily planning
├── projects/        <- Project tracking
├── content/         <- Content creation
├── inbox/           <- Quick capture
└── ideas/           <- Processed ideas

Why Cockpit? Because it’s my command center.

Every morning, I sync with my AI manager here: what got done yesterday, what to push forward today. Once aligned, I head to individual projects to execute. Before signing off, I come back to update the log and dump any thoughts into inbox.

Everything is Markdown. Claude Code can read and write directly.

Content Discovery (Daily Brief)

Auto-scrapes Reddit, Hacker News, X for trending content. AI analyzes and scores. Every morning, a brief is generated.

GitHub Actions runs daily. I just scan “what’s worth paying attention to.”

What’s Not Done Yet

Honestly, I’m not at Molly’s level yet.

Email automation? Still manual
Finance tracking? Haven’t connected APIs
Health data integration? Haven’t started

But that’s fine. Systems are built gradually, not all at once.

Where You Can Start

If you’ve read this far, you might want to try it yourself.

You don’t need to build 8 systems at once. Start with one pain point.

Ask yourself three questions:

What do I repeat every day?
What information do I spend the most time finding?
What do I most often “forget”?

The answer is your starting point.

Tool Recommendations

Purpose	Beginner	Advanced
Notes	Obsidian	Obsidian + MCP
Automation	Zapier / Make	GitHub Actions / Cron
AI Assistant	ChatGPT	Claude Code
Website	Notion Site	Astro + Markdown

Don’t overthink the beginner choices. What matters is starting.

Final Thoughts

Molly’s system is impressive. But it took her months to build.

My system is still being built. This post isn’t a “I did it” share—it’s a “I’m doing it” record.

Before, we were shaped by tools. Whatever app we used, we were limited by its logic.

Now, we can shape our own tools.

The tower is still a tower. But the keys can be in your hands.

Further reading: Molly Cantillon’s original post

Want more content like this? Subscribe to my newsletter