Probably. But agent capability has two tests: can it act, and can it recover when the world changes mid-task? Robotics has been forced to care about the second forever. Closed-loop feedback and post-training will matter more than the next slick demo.