This German AI Startup Thinks Retraining Is a Waste of Time (And They Just Got $12M to Prove It)

Jan de Vries · 2026-07-23

### A New Approach to AI: Think First, Adapt Later For years, the AI industry has followed a familiar playbook: throw more data at a model, retrain it constantly, and hope the world doesn't change too much in between. But a German DeepTech startup, kausable, is betting that there's a better way. The company just raised $12 million in a Seed round to build what it calls a "reasoning-first" AI—a system that adapts to new situations without needing expensive retraining. The round was led by UVC Partners and Entourage, with follow-on investments from German backers HTGF and Mätch VC. A group of angel investors from Black Forest Labs, OpenAI, Google DeepMind, Noxtua, and the European Laboratory for Learning and Intelligent Systems (ELLIS) also chipped in. ### What Makes kausable Different? Most AI models today are like a student who memorizes the textbook but freezes when you ask a question that isn't in it. kausable's founders—three physicists from Heidelberg University and Black Forest Labs—are building something different. Their AI learns from minimal data and adapts to scenarios it couldn't have anticipated at build time. "The world evolves rapidly, and so should AI," says Johannes Haux, CEO and co-founder. "Even the most capable current systems are static and need constant, costly retraining. We're developing a world model that learns on its own and adapts to change." This matters for industries where the ground shifts constantly. Think robotics, energy markets, or finance—where a model trained on yesterday's data can be obsolete tomorrow. ### The Tech Behind the Hype kausable's core innovation is a "world model" that mimics how humans learn. Instead of memorizing patterns, it builds a set of intuitions—a mental framework that lets it adapt quickly with just a little new information. The company recently developed TipPFN, a zero-shot model that predicts rare but high-impact events (think "black swans") in medicine and energy. It also co-authored a research paper with experts from Columbia University that validates the causal reasoning architecture at the heart of its system. That's a big deal: it suggests their approach isn't just theoretical. ### Why Investors Are Paying Attention Pieterjan Bouten, co-founder of Entourage, put it bluntly: "Most AI models are trained to remember the past. kausable is building AI that can reason about the future. Instead of relying on ever-larger datasets and constant retraining, they're developing systems that adapt, infer causality, and solve problems they've never seen before." That's the kind of pitch that gets investors excited. But it's also a huge technical challenge. The company currently has just nine employees, and they'll use this funding to expand the team and push their model forward. ### What's Next? kausable isn't trying to build another niche model. They're going after a horizontal AI that can reason through unfamiliar problems across different domains. If they succeed, it could mean systems that adapt instantly to new situations instead of breaking down when the data shifts. For now, the team is focused on hiring and development. But if their approach works, it could change how we think about AI—moving from systems that memorize to systems that truly understand.