Philosophy, linguistics, and German Studies at HHU Düsseldorf — Wittgenstein, pragmatism, philosophy of language. I went in expecting academic philosophy and came out with a different question: not "what do words mean?" but "what do they do?" How framing shapes what's possible. How the same problem dissolves or solidifies depending on how you describe it. That question turned out to be the most practical thing I've ever studied.

It runs through everything. Building agentic systems is a language problem before it's an engineering problem. A system prompt isn't code — it's a behavioral contract written in natural language, where every word carries weight because the model takes you literally. Tool design isn't API design — it's communication design, writing for a reader that pattern-matches on descriptions. The gap between an agent that works and one that doesn't is almost always a gap in clarity, not capability. Wittgenstein argued that philosophical problems dissolve when you examine the language that created them. The same move works on broken agent architectures.

I work full-time at a digital marketing agency in Düsseldorf — 5.5+ years as the first full-time employee, building operational structures from the ground up. Strategy, concept development, copywriting, SEO, client consulting, project management, and delivery across 100+ projects. Three years ago, AI changed how I deliver that work — not as a shortcut, but as a multiplier. The judgment and the relationships are still mine. AI handles the execution bottlenecks that used to cap what one person could ship.

The work I'm most interested in now sits between the model and the world. System prompts, tool architectures, multi-agent orchestration, the infrastructure that turns generic AI capabilities into systems that solve specific problems. That layer — where domain expertise meets agent infrastructure — is where the actual value lives. Not in the models themselves. Models get commoditized. The orchestration layer doesn't, because it encodes knowledge that can't be prompted out of any model.

I write about what I build. The blog covers agent architecture, infrastructure economics, open-source dynamics, and what happens when you take philosophy of language seriously in a field that runs on language. Every public text now has three audiences — human readers, the writer, and future AI models that will compress it into training data. I write for all three.