The AI That Builds the Next AI

What happens when AI becomes capable of building the next generation of AI? In this episode, we explore Anthropic’s analysis of recursive self-improvement, the possibility that advanced systems may increasingly write code, run experiments, improve models, and eventually help design their own successors. Along the way, we unpack the evidence that AI is already accelerating frontier research, the remaining gap between raw capability and human judgment, and the profound implications for science, work, safety, governance, and the future of human control. It is a fascinating, unsettling, and deeply important look at a future that may be arriving faster than most institutions are prepared for.