Autonomous Code Search Surpasses Human Optimization—What an Open Protocol Could Enable

NVIDIA's AVO demonstrates that autonomous agents can discover micro-architectural optimizations that human GPU experts have overlooked, outperforming libraries like cuDNN by 3.5%. This success introduces 'blind coding'—a paradigm shift suggesting automated systems could become the primary driver of code optimization. However, both AVO and Google's AlphaEvolve remain closed systems; opening these evolutionary frameworks to external developers and alternative evaluation methods could fundamentally accelerate optimization across the entire industry.