AFlow automates the creation of agentic workflows by searching in a code-represented workflow space using Monte Carlo tree search. It replaces manual workflow design with machine-generated solutions, building workflows from configurable nodes (LLM invocations) and reusable operators such as Generate, Format, Review, Revise, and Programmer. The system iteratively selects, expands, evaluates, and updates workflows based on task performance, demonstrating potential to outperform handcrafted agent workflows on various tasks.