Because the real world hasn’t fully come to the table yet.

At NeurIPS Urban AI, I spoke with Fabian Raisch (TU Rosenheim/TU Munich), whose research targets scalable AI-based building control with significant potential: improving efficiency by 10–50%, depending on building conditions and systems.

Fabian shared the most important transition happening right now: his team has been working in the ideal world of simulated data, and now they’re stepping toward real deployment, working directly with heat pump operators and practical constraints.

That’s the pivot point for the entire industry. The AI is moving fast. The research is impressive. The agents are ready.

What’s missing is living, accessible, trustworthy building data, because most buildings are still “quiet. Not every building AI approach needs huge training datasets; some (like what Facil describes) are closer to real-time control and continuous learning through intervention. But either way, we still need buildings to become accessible, observable, and safely controllable.