In this episode, I talk with Harald Schäfer, CTO at comma.ai, where he is leading one of the most interesting autonomy efforts in the world:

They work on end to end driving and generative world models is changing how small teams can compete with billion dollar labs.

We talk about his path from electrical engineering in Belgium and Santa Barbara to joining comma as one of the earliest engineers. Harald explains how he helped turn a hacker project into a focused engineering team that ships reliable autonomy to thousands of real users.

He walks me through comma’s move to a single neural network that controls the car from video input, why deleting code is often more powerful than adding more, and how his team uses world models to train on billions of synthetic miles that never existed on real roads.

Harald also shares what it is like to build inside a company with no CEO, why simplicity beats complexity in autonomy systems, and how the new comma 4 and Body 2 signal a move beyond cars into general robotics.

If you work in robotics, autonomy, or AI systems, this conversation is packed with lessons about engineering clarity, avoiding brittle stacks, and shipping real products with small teams.