How Shunyu Yao is Shaping the Second Half of AI with Agents

Shunyu Yao, a 29‑year‑old AI researcher, graduated from Hefei No.45 Middle School, won a silver medal in the National Olympiad in Informatics (NOI), scored 704 in the Chinese college entrance exam, entered Tsinghua’s elite Yao class, and earned a Ph.D. in Computer Science (language and reinforcement learning) from Princeton before joining OpenAI.

01 Yao's Growth Path

During his studies he recommended the classic book Gödel, Escher, Bach , emphasizing interdisciplinary thinking. He later became a co‑founder of Tsinghua’s rap club, a volunteer recruiter, and chair of the Yao class council.

Tree of Thoughts (ToT): enables large language models (LLMs) to iteratively think and improve reasoning.

SWE‑bench: a benchmark dataset for evaluating LLM capabilities on software engineering tasks.

SWE‑agent: an open‑source AI programmer.

ReAct: synergizes reasoning and acting in language models.

His early research shifted from computer vision to natural language processing after a MIT exchange, where he discovered the synergy between psychology and AI.

Research Highlights

At Princeton he joined Karthik Narasimhan’s group, focusing on NLP and reinforcement learning. His first Ph.D. project, CALM (2020), treated language as a medium that converts human experience into actionable candidates for agents.

Subsequent works include:

WebShop: a large‑scale simulated e‑commerce environment where agents follow complex textual instructions.

SWE‑bench and SWE‑agent: evaluating agents on real‑world programming tasks.

ReAct: enabling LLMs to reason and act simultaneously when interacting with external environments.

“If your pre‑training already contains everything, reinforcement learning is just a skill that activates those abilities.” – Shunyu Yao

He argues that the emergence of GPT‑3.5 highlighted the importance of prior knowledge, allowing agents to generalize across tasks.

02 The Second Half of AI

In April 2024 Yao published a blog post titled “The Second Half”, declaring that AI’s development has entered a new phase where the focus shifts from model building to defining meaningful real‑world tasks and evaluating systems accordingly.

He sees his role evolving from a pure researcher to a “product manager” who discovers valuable problems and builds solutions, while also contemplating entrepreneurship—aiming to create a trillion‑dollar company based on agents.

Yao’s journey illustrates a blend of deep theoretical foundations, interdisciplinary curiosity, and a drive to translate AI research into practical, general‑purpose agents.