A deep reinforcement learning algorithm can solve the Rubik's Cube puzzle in a fraction of a second. The work is a step toward making AI systems that can think, reason, plan and make decisions.