As Physical AI transitions from controlled laboratory demonstrations to real-world deployment, the core challenge shifts from capability ("Can it walk?") to systemic operational viability ("Can it work?"). This keynote looks past viral demos to address the practical friction points of scaling embodied autonomy in unconstrained environments. We will examine the critical data bottleneck stalling the industry, contrasting the scalability of passive video data against the high-fidelity precision of human teleoperation. We will also explore how the rise of predictive world models is enabling robots to generalize and adapt to novel surroundings. Lastly, the session introduces a clear, structured paradigm for assessing true, real-world commercial viability.