This AI entrepreneur is developing agents that can plan ahead for the unexpected
TL;DR
Danijar Hafner’s office in San Francisco’s SoMa district sits mostly empty. His brand-new startup is still in stealth mode and doesn’t even have its name on the door. On the day I visit, there’s only one other person there, and little in the way of furniture. But what it lacks in decor, it makes up….
Nauti's Take
The opportunity is real, because today's agents fail exactly where a task drifts away from what they saw in training, and Hafner's work on world models targets that gap directly. The risk is maturity: a stealth startup with no product, no benchmark, and a research line that has so far shown its strengths mainly in games and simulations.
Teams building agent workflows today should stay with narrow tasks and hard guardrails, and treat this direction as a medium-term bet.
Summary
Danijar Hafner’s office in San Francisco’s SoMa district sits mostly empty. His brand-new startup is still in stealth mode and doesn’t even have its name on the door.
On the day I visit, there’s only one other person there, and little in the way of furniture. But what it lacks in decor, it makes up…
Sources
