Writing
- The future of SaaS is tailored software
I built an NFC business card platform with a friend who runs a print shop. What we learned along the way says something about where software is going.
- The ticket is the thinking
How I treat every task, bug, and finding as a spec — and why that turns a kanban board into a compounding knowledge system.
- How we stopped hoping our AI was right and started knowing it
We built a fully automated campaign recommendation pipeline. The AI was fast, impressive, and wrong in ways we didn't notice for weeks. Here's what fixing that actually looked like.
- My sync daemon wiped 28 tasks in two seconds. Here's what I learned.
A war story about building a Notion ↔ Obsidian sync daemon — what broke, why 28 tasks vanished, and the three simple fixes that made it resilient.
- We can each build anything. So why can't we build together?
Notes from AI Bar Camp Berlin 2026 — my talk, what the room taught me, and the shared context problem nobody has solved yet.
- The future of AI agents looks a lot like 2010.
The thing missing from most agent setups isn't model quality or tool use. It's project management. Old patterns, new actors.
- You were never paying for the software.
The AI-kills-SaaS debate is asking the wrong question. SaaS was always a data product. AI doesn't change that — it makes it more true.
- CTR is not the story. It's the first sentence.
Optimizing search ranking for a sponsored ads product taught me something about proxy metrics: CTR tells you one thing. It doesn't tell you everything — and the gap is where the interesting problems live.
- Most AI-generated Jira tickets are useless. Here's the system that fixed mine.
A set of Claude Code skills that encode your team's ticket philosophy once — so every Jira output starts grounded, not generic.
- We could only manually review a fraction of campaigns each week. Now every single one gets analysed every night.
How I automated campaign review for a large book of advertisers using Python, Bedrock, and GitHub Pages.
- My tax script kept breaking. Claude Code read the whole thing and fixed it.
How switching from patch-by-chat to Claude Code's full-context collaboration changed the way I build and maintain automation scripts.
- My daughter named it. I just had to build it.
Building a printable coloring page generator with Gemini, Supabase pgvector, and Puppeteer.
- Education leave rocked my work
Five days at a remote Generative AI course. I went to learn Python and LLMs. I did not expect to ship.
- AI Invoice Automation (Gemini)
A self-learning pipeline that processes German tax invoices from Google Drive using Gemini AI — zero cost, fully automated.
- Smoobu → Google Sheets Automation
Automatically syncs Airbnb/Booking.com reservations from Smoobu into separate sheets for your cleaner, accountant, and yourself.
- Interactive Portfolio — AI Recruiter Copilot
A dynamic portfolio that acts as my AI proxy — processes 10+ years of product leadership data to evaluate job fit and answer recruiter queries.