Your Second Brain Has a New Reader

Yesterday I gave a talk at the DevOps Meetup in Freiburg, hosted at JobRad. The title: Your Second Brain Has a New Reader.

The core thesis is simple: most of us have spent years building personal knowledge bases — Obsidian vaults, Notion workspaces, org-mode files, plain markdown. And most of that knowledge sits there unread. Write-only graveyards.

AI agents change this. Not because the models are smart, but because they’re readers. They consume your accumulated context and act on it. The agent is only as good as your notes — but good notes in, and it becomes scarily capable.

What I Covered

The write-only problem. I asked the audience to raise their hand if they have a personal knowledge base. Most hands went up. Then I asked who searched it in the last 24 hours. Yeah.

Three real examples of what changes when notes have an active consumer:

1. Course prep — I said “Prepare for my course in Barcelona.” The agent found the email with the syllabus link, downloaded the Google Doc, read student feedback from my Obsidian vault, and created a prep document with action items. One sentence, five minutes.

2. Morning planning — Every day at 6 AM, my agent triggers a morning routine. I say “What didn’t I finish yesterday?” — it reads yesterday’s daily note, finds unchecked tasks, carries them over. Two minutes of chat gives me a structured daily plan.

3. This talk — The meta-recursive moment. I said “Let’s work on the meetup talk” and the agent searched my memory, found the event details, read the existing outline in Obsidian, and built these slides. The talk about AI reading your notes was prepared by AI reading my notes.

The architecture: files, not features. My stack is Obsidian (961 files) + an AI agent + qmd for semantic search. But the principle is tool-agnostic. It’s just markdown:

• MEMORY.md — curated long-term knowledge
• memory/YYYY-MM-DD.md — daily raw logs
• HEARTBEAT.md — periodic checks
• USER.md — who you are, preferences
• TOOLS.md — your specific setup

The agent wakes up fresh every session. These files are its memory. No vector database needed at personal scale. Everything is git-tracked, human-readable, portable.

The feedback loop. Conversations become daily notes, daily notes get fact-extracted into durable knowledge, and the knowledge base maintains itself. The uncomfortable question: do you trust an AI to edit your notes? My answer: PR model for your brain — separate spaces, git-tracked, always revertable.

The Meta Moment

The entire presentation — including the slides — was generated by my AI agent reading my Obsidian notes. During the talk, this got the biggest reaction. The slides are a single HTML file, zero dependencies, Terminal Green aesthetic (JetBrains Mono, scan lines, CRT vignette). Built with the frontend-slides approach.

View the slides →

Audience Questions

The audience asked about cost and infrastructure:

• Token costs: ~€100/month currently, budgeting for €200 as usage grows
• Hardware: A small Linux server (nexus) running the agent 24/7 + a Mac mini for development

Takeaways

1. Your notes are an API — treat them like one. Structure matters.
2. Context > model — the same model with your vault context vastly outperforms the same model without it.
3. Files > features — markdown files beat any proprietary system.
4. Start small — one AGENTS.md file describing your project changes everything. GitHub already supports this for Copilot.

The ROI is real: 30 minutes of structured writing gives you weeks of better AI interactions. You don’t need a fancy setup. Start with one markdown file today.