Studies how models alternate reasoning with actions, then tests whether tool-using agents complete requests consistently across conversations.
Why read Shunyu Yao?
Yao’s coauthored ReAct work gives you a concrete starting point for agents: make a decision, use a tool, read the result and decide what to do next. The project page contrasts this with reasoning alone and actions alone, and includes a failed task that can be inspected rather than hidden behind a successful demo.
His coauthored τ-bench paper asks a different question: does the agent repeatedly reach the required outcome while talking to a user and following domain rules? It compares the final database with the expected state and distinguishes one success from success on every repeated trial. The simulated users and simplified domains make it a controlled research test; the paper also explains why a correct final state can still miss a policy violation.
Start with the original
Selected work
-
Research walkthrough
ReAct: Synergizing Reasoning and Acting in Language Models
How do reasoning and external actions fit together? This coauthored project walkthrough shows alternating thoughts, actions and observations, including a failure example. The linked paper was first submitted on 6 October 2022.
Cite this selection
XFollowList, “ReAct: Synergizing Reasoning and Acting in Language Models” reading notes in the Shunyu Yao profile. https://xfollowlist.com/people/shunyu-yao#work-9c9b04f70c57
The linked work is the original source; this permalink identifies XFollowList’s reason to read it. Original publication date is unknown.
Profile-source checking date and references · Dated reading map
-
Research paper ·
τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains
How reliable is an agent that changes records? Coauthored with Noah Shinn, Pedram Razavi and Karthik Narasimhan, this paper compares expected and actual database outcomes and measures consistency across repeated trials. Read the setup before interpreting the scores.
Cite this selection
XFollowList, “τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains” reading notes in the Shunyu Yao profile. https://xfollowlist.com/people/shunyu-yao#work-1fd5587af573
The linked work is the original source; this permalink identifies XFollowList’s reason to read it. Original work published .
Profile-source checking date and references · Dated reading map
Projects & roles
Community rating
Sources & further reading
Profile sources checked . This describes checking these references, separately from the publication dates of selected works.
- Yao’s own research site links @ShunyuYao12 and lists both coauthored projects. Its February 2025 OpenAI role text is historical.
- The author record and original submission date establish Yao’s coauthorship of τ-bench.
- Current X title confirms Shunyu Yao and @ShunyuYao12. The current bio states Language agents; no employer claim is inferred.
Cite this profile’s source check
XFollowList, Shunyu Yao profile sources. Sources checked 2026-10-11. https://xfollowlist.com/people/shunyu-yao#sources-checked