Code as the engine for reasoning and acting in AI agents

Xuying Ning, Katherine Tieu, Dongqi Fu, Tianxin Wei, Zihao Li, Yuanchen Bei, Jiaru Zou, Mengting Ai, Zhining Liu, Ting-Wei Li, Lingjie Chen, Yanjun Zhao, Ke Yang, Bingxuan Li, Cheng Qian, Gaotang Li, Xiao Lin, Zhichen Zeng, Ruizhong Qiu, Sirui Chen, Yifan Sun, Xiyuan Yang, Ruida Wang, Rui Pan, Chenyuan Yang, Dylan Zhang, Liri Fang, Zikun Cui, Yang Cao, Pan Chen, Dorothy Sun, Ren Chen, Mahesh Srinivasan, Nipun Mathur, Yinglong Xia, Hong Li, Hong Yan, Pan Lu, Lingming Zhang, Tong Zhang, Hanghang Tong, Jingrui He

This survey reframes the role of code in agentic AI systems. Rather than treating code as only a target output, the authors propose code as agent harness: a unified framework where code serves as the operational substrate for agent reasoning, action planning, environment modeling, and execution verification. The framework spans three layers—harness interface (connecting agents to reasoning and tools), harness mechanisms (planning, memory, feedback-driven control), and scaling to multi-agent systems (coordination and verification). The work covers practical applications across coding assistants, automation, embodied agents, scientific discovery, and enterprise workflows, while identifying open challenges including evaluation metrics beyond task success, verification with incomplete feedback, and safety oversight for safety-critical actions.