xAI Co-Founder: Igor Babuschkin Unpacks the Future of Model Development

The rapid evolution of AI agents, particularly in coding, signals a shift toward personal AI systems that prioritize individual control and local hardware deployment. Igor Babuschkin, formerly of DeepMind, OpenAI, and xAI, founded River AI to address the concentration of power within large, closed-source model providers. By moving inference compute locally and enabling models to learn directly from user interactions, these personal agents better align with human preferences and protect sensitive data. While frontier models excel in verifiable domains like mathematics and coding, the next frontier involves training agents for non-verifiable, long-horizon tasks. Achieving this requires moving beyond centralized data centers toward distributed, specialized post-training, ensuring that AI benefits are broadly accessible rather than controlled by a few entities. This approach balances the need for advanced capabilities with the necessity of maintaining human agency in an increasingly automated world.