Waymo Co-CEO Dmitri Dolgov: "Move Fast And Ship Safely"

Building and scaling physical AI requires transitioning from successful demos to reliable, safety-critical products by climbing an exponential ladder of performance "nines." Unlike digital AI, physical agents must navigate the high costs of error, latency constraints, and the lack of a pre-labeled internet-scale dataset. Success depends on a "structure-augmented end-to-end" architecture that leverages learned representations while respecting physical laws and road rules. A robust ecosystem—integrating the agent, a high-fidelity generative simulator, and a rigorous critic—creates a self-reinforcing flywheel that accelerates development. Ultimately, safety metrics and transparent, evidence-based performance data serve as the primary competitive moat, as seen in Waymo’s record of being 17 times safer than human drivers in serious injury crashes. This approach transforms AI from a technical experiment into a reliable, life-saving service that earns public trust through consistent, real-world operation.