Many real-world applications of artificial intelligence need agents that can compete and coordinate with other agents in complex environments. Game of StarCraft, one of the hardest professionally played esports, has emerged by consensus as one of the key challenges on this path. Over the last decade, the best solutions simplified crucial aspects of the game, had superhuman capabilities, or hand-crafted subsystems. Despite all these advantages, none came close to challenging top professionals. In this talk, I will introduce AlphaStar, a multi-agent deep reinforcement learning system, which is the first learning system to ever achieve top tier of human performance in a professionally played esport without any game restrictions (GrandMaster league in the game of StarCraft II). We will focus not only on how this achievement has been reached, but rather on understanding how these results, solutions and modules can be used in other real-life challenges of artificial intelligence.