Shunyu Yao
Confidence 0.75 · 2 sources · last confirmed 2026-06-12
AI researcher and first author of ReAct — the 2022 paradigm interleaving reasoning and acting in language models that became the structural core of the modern agent loop. At the time of the work he was a PhD student in the Princeton NLP Group (advised by Karthik Narasimhan), conducting the research during a Google internship with the Google Research Brain team.
Wiki contributions
- ReAct (paper, ICLR 2023) — first author. Introduced the reason–act–observe loop, the dense/sparse thought regimes, and the human-in-the-loop trace-editing result.
- ReAct (Google Research blog) — co-author of the popular announcement (with Yuan Cao).
His work is the genealogical root of the wiki’s harness and agent cluster: the “LLM that uses tools in a loop” definition the wiki adopts from 2026 practitioners is ReAct named four years earlier.
Why he is an entity (promotion note)
Promoted on 2026-06-12: named author on two wiki sources (the ReAct paper and its Google Research blog), crossing the second-source threshold. A strong anchor for any future agent-foundations ingests (Tree of Thoughts, SWE-bench, and related agent-reasoning lineages are natural follow-ons).