Igor Babuschkin is cofounder and CEO of River AI, which is developing personal AI that users can adapt and control. He previously cofounded xAI and contributed to DeepMind’s AlphaStar research.[1][3][5]
Start with AlphaStar: Grandmaster level in StarCraft II →
X Career
Babuschkin describes starting in physics on CERN’s LHCb experiment before moving into AI. At DeepMind, he worked on WaveNet and AlphaStar; at OpenAI, he worked on reasoning and pretraining. He later cofounded xAI and describes leading work across infrastructure, training, multimodal systems and product. His current personal site presents investment in AI founders through Babuschkin Ventures.[1][2]
Contributions
Training a population of StarCraft agents
Babuschkin is credited on DeepMind’s AlphaStar account. The system first learned from human replays, then improved through reinforcement learning in a league of agents. Some league members exploited the main agents’ weaknesses, producing opponents that discouraged a narrow strategy from dominating training. The final system reached Grandmaster level in the full StarCraft II game. The league mechanism matters because repeatedly beating one opponent can leave a player vulnerable to a different strategy.[3]
Personal AI and model adaptation
Babuschkin introduced River AI in June 2026. Its first public product, the River API, lets developers adapt models at the training layer. He describes personalization and continual learning for agents as the next layer: systems that learn a person’s preferences rather than requiring repeated supervision. This is the company’s development direction, with the API as the initial released component.[5]
Ideas
Babuschkin’s investment statement emphasizes founders who build AI with a long-term view and regard it as a tool for expanding what people can do. His personal account connects that position to his progression from scientific research to building and financing AI systems. The statement describes the intention behind his investment work rather than a demonstrated safety result from any particular company.[1][2]
Tools, writing & talks
Writing, research & talks
Start here
AlphaStar: Grandmaster level in StarCraft IIdeepmind.googleExplains imitation learning, reinforcement learning and league training for a competitive real-time strategy agent.
Read the league-training explanation to understand why playing against varied opponents can improve a learning agent. The game supplies an intuitive example; no reinforcement-learning implementation experience is required.
Credited research contributor[3]- Igor Babuschkin: Aboutbabuschk.in
A first-person account of his research and company-building career.
Author[1] - Introducing River AIriver.ai
Babuschkin explains River’s personal-AI direction and its first model-adaptation API.
Author and company cofounder[5]