Study finds dynamic concurrency in coding agents helps long tasks but hurts short ones
Researchers from Nanjing University and UIUC ran paired experiments on the dynamic concurrency mechanisms of three coding agents — Codex, Claude Code and Kimi Code — across 354 software engineering tasks and 2,124 executions, analysing full main-agent and sub-agent traces. Results varied by task type and system: on LoopsBench, Claude Code's completion rate rose from 7.1% sequentially to 21.4% with concurrency, and Codex rose from 34.6% to 57.7% on NL2Repo; on SWE-bench Verified, Claude Code fell from 83.0% to 59.0% and Kimi Code from 77.0% to 59.0%, while Codex stayed at 71.0% in both modes. Trace analysis produced a taxonomy of 4 categories, 13 subcategories and 28 concurrency failure modes (such as missing owners, pseudo-concurrency, load imbalance, mistimed result merging and shared-environment conflicts), plus three mechanisms where concurrency paid off: complementary work, parallel exploration and independent verification.
The study says support for spawning sub-agents does not by itself mean reliable collaboration — task decomposition, execution governance and result integration determine whether parallelism turns into gains, offering a reference for coding-agent design and evaluation.
新智元 2026-10-19 10:36 北京
新智元报道
Coding Agent正在进入新的发展阶段。过去，它们主要修复Bug、解决GitHub Issue或实现相对独立的功能，SWE-bench等Benchmark已为此建立了成熟评测范式；如今它们开始承担从完整代码仓库生成、跨语言迁移，到需要持续探索与验证的长周期软件工程项目，任务规模和执行时间不断增长，单个Agent的上下文容量、推理能力和执行效率逐渐成为限制。
一种自然的解决思路是让多个Agent同时工作。近年来，Codex、Claude Code和Kimi Code等前沿Coding Agent开始支持动态并发（Dynamic Concurrency）：主Agent在执行过程中自主决定是否创建子智能体、何时创建，以及如何协调它们的工作。
理论上，这种方式能将复杂任务拆分，让不同Agent并行探索、实现和验证。
#Codex#Claude Code#Kimi Code#Nanjing University#University of Illinois Urbana-Champaign#SWE-bench Verified#LoopsBench#NL2Repo#RepoZero#ProgramBench