Japanese AI startup Sakana AI has established a Recursive Self-Improvement (RSI) research lab to develop AI systems capable of rewriting and enhancing their own code. This four-phase program aims to create AI architectures that optimize themselves autonomously, from the initial design phase to advanced AI research. Sakana AI plans to develop "Agent-Native Models" designed for open-ended tasks, which will later evolve into "AI Scientists" capable of independent research. The third phase envisions systems that improve their own foundation models and architectures by writing, testing, and verifying code. In the long term, Sakana AI aims for "Democratized AI," making powerful AI systems accessible even to smaller players and counteracting the hardware arms race among major tech corporations.