CoCoEvolve: Evolutionary Optimization for AI Systems
Snowflake, Wednesday, June 3rd, 2026
CoCoEvolve uses AI coding agents instead of LLMs to automatically optimize AI systems through evolutionary techniques.
CoCoEvolve is an evolutionary optimization framework developed by Snowflake that uses CoCo, their AI coding agent, to automatically propose, test, and implement improvements to AI systems. Unlike traditional LLM-only approaches, CoCoEvolve can interact with live systems, self-validate changes before expensive evaluations, and apply verified domain-specific skills.
In benchmarks, CoCoEvolve improved a Cortex Agent from 22% to 89.9% on the DABStep Hard benchmark, boosted dbt task pass rates by 8.5%, and improved AI Function accuracy by 41.3%, demonstrating significant advantages over manual tuning and LLM-only optimization methods.