More about HKUST
Testing, Understanding, and Optimizing Coding Agents: A Survey on Multi-Agent Systems for Code Generation
PhD Qualifying Examination
Title: "Testing, Understanding, and Optimizing Coding Agents: A Survey on
Multi-Agent Systems for Code Generation"
by
Mr. Zongyi LYU
Abstract:
With the rapid advancement of large language models (LLMs), coding agents have
emerged as a transformative paradigm for automating software development
tasks. Among the various instantiations of this paradigm, multi-agent code
generation systems (MACGS) have demonstrated remarkable potential by
decomposing complex programming tasks into specialized subtasks delegated to
dedicated agents. Despite remarkable benchmark performance, the engineering
foundations of these systems-their reliability, internal mechanisms, and
optimization pathways-remain poorly understood, posing critical barriers to
real-world deployment. This survey provides a systematic review of coding
agents, especially MACGS, through the lens of testing, understanding, and
optimization, a perspective that reflects the maturation of the field toward
production-readiness: testing establishes a principled, reproducible
characterization of system behavior; understanding then transforms empirical
observations into actionable mechanistic insight; and optimization, grounded
in that insight, applies targeted interventions that yield principled and
transferable improvements. Specifically, we first categorize the architectural
landscape of coding agents, spanning role-based pipelines and adaptive
frameworks. We then survey different testing methodologies, analysis
frameworks, and optimization strategies. We conclude by identifying open
challenges in these perspectives, positioning this survey as both a map of
current progress and a guide for the research directions that will define the
next generation of reliable coding agents.
Date: Wednesday, 15 July 2026
Time: 10:30am - 11:30am
Venue: Room 3494
Lift 25/26
Committee Members: Prof. Shing-Chi Cheung (Supervisor)
Dr. Shuai Wang (Co-supervisor)
Prof. Raymond Wong (Chairperson)
Dr. Binhang Yuan