Shunyu Yao

Shunyu Yao has made foundational contributions to the development of language agents. He introduced the ReAct framework, the first to combine reasoning and acting within an agent paradigm. This approach established a scalable and generalizable foundation for building language-based agents. At its core, ReAct enables large language models to perform interpretable internal reasoning before making decisions and taking actions. This significantly improves controllability and greatly expands their applicability in real-world tasks. Today, ReAct has become the most widely adopted method for building language agents across academia and industry.

To further explore the real-world potential of language agents, Shunyu developed new interactive environments such as WebShop, SWE-bench, and tau-bench. Unlike traditional reinforcement learning platforms like Go or Atari, these benchmarks reflect high-value, practical tasks such as web interaction, software engineering, and customer service automation. These environments have become essential for evaluating agent generality and enabling practical deployment. Correspondingly, systems like SWE-agent have achieved breakthroughs in code generation and interactive debugging.

Building on this foundation, Shunyu became a core contributor to OpenAI’s first generation of agent products in 2025, including Operator and Deep Research. Operator focuses on general computer system interaction, turning abstract digital environments into executable tasks for agents. Deep Research targets knowledge-heavy domains like science, law, and finance, enabling advanced assistant and collaboration capabilities. These innovations represent a major step toward mature language agents and illuminate a promising path to AGI.