Google DeepMind’s new AI agent, AlphaEvolve, is revolutionizing algorithm development by combining large language models with automated evaluators. This innovative system, powered by Gemini models, evolves codebases to address complex mathematical and computing challenges. AlphaEvolve has already enhanced Google’s data center efficiency, hardware design, and AI training processes, with improvements of up to 32.5% in certain applications. It has also made breakthroughs in open mathematical problems, including advancing the kissing number problem and discovering new matrix multiplication algorithms. AlphaEvolve’s success in diverse areas, from data center scheduling to chip design and AI model training, points to its potential for broader application across fields like material science and drug discovery.