The New Mythical Man-Month in Vibe Coding Period
Why faster implementation makes judgment, taste, and conceptual integrity even more important.
A little about me
Master’s student in Computer Science
Tsinghua University · Beijing, China
Exploring how intelligent agents perceive, reason,
and act — together, and in the real world.
I’m a master’s student at the Department of Computer Science, Tsinghua University (2025–2028), supervised by Prof. Zhiyuan Liu. I hold dual bachelor’s degrees in Mathematical & Physical Sciences and Software Engineering from Tsinghua.
My current research focuses on embodied agents: building intelligent agents that perceive, reason, and act in complex physical and interactive environments. Previously, I worked on LLM agents and multi-agent systems, exploring how interaction improves individual capabilities, how collaboration emerges, and how to organize agents for individual and collective intelligence.
I also care deeply about open source. I’m a core contributor to ChatDev and led its second-generation version, ChatDev 2.0 — DevAll.
My dream is to build an agent with superhuman capabilities in both reasoning and real-world action — one that learns and grows through its own experience of the world.
“A new generation of agents will acquire superhuman capabilities by learning predominantly from experience.”
We’ve released JustRUA! Project page · Paper.
ChatDev 2.0 - DevAll has been accepted to the EMNLP 2026 Demo Track with reviewer scores of 7, 7, and 8 🤓.
Our work AgentXRay is accepted at ICML 2026.
🚀 We have released ChatDev 2.0 - DevAll! We welcome issues and PRs. And special thanks to all the contributors for their hard work.
Our work Multi-Agent Collaboration via Evolving Orchestration is accepted at NeurIPS 2025. Hope this is a good start.
Our technical report about GUI Agent is released on arXiv.
I graduated from Tsinghua University with dual bachelor’s degrees in Mathematical & Physical Sciences (MPS) and Software Engineering.
Our team’s work about “Multi Agent Cross-Team Collaboration” is accepted by ACL 2025 Findings.
Our team’s work around “Scaling LLM agent collaboration” is accepted by ICLR 2025.
Our team’s work around LLM Agent under information asymmetry is accepted by NeurIPS 2024. Congrats 🥳!
Our two papers have been accepted to ACL 2024, main conference 🥳. Thanks to my best mentor and all co-authors!
† Equal contribution. * Core contributor. Highlighted entries indicate first / co-first authorship or a core-contributor role.
Citation counts: Semantic Scholar · checked .
Notes on agents, learning theory, and building things.
Why faster implementation makes judgment, taste, and conceptual integrity even more important.
Exploring search, world models, and reinforcement learning for language-model planning.
Notes on PAC learning, generalization bounds, Rademacher complexity, and VC dimension.
From tree search and exploration to neural guidance and self-play in AlphaGo Zero.
Teaching assistant for Towards Artificial General Intelligence at Tsinghua University.
I serve as a reviewer for ARR, EMNLP, CVPR, and ICLR.