文章背景与核心概要
智能体工程系统(Agentic engineering systems)在通用软件工程基准测试中展现出了令人瞩目的性能,然而它们在企业资源计划(ERP)领域特定语言(DSL)中的应用效果长期以来鲜有探索。为了填补这一空白,研究人员推出了 BC-Bench,这是一个专为评估智能体在微软 Dynamics 365 Business Central 所使用的 DSL——AL 语言中的真实世界任务表现而设计的全新基准测试。
该论文的核心亮点包括:BC-Bench 包含了从两个微软拥有的生产代码库中人工整理的 101 个任务,真实还原了企业资源规划(ERP)的开发工作流;通过改编 SWE-Bench 方法论,该基准测试克服了 AL 生态系统资源公开较少、环境配置复杂等特有挑战;除了代码功能性之外,BC-Bench 还评估了测试生成能力,并支持包含视觉上下文的多模态问题描述。评估结果表明,在通用基准上观察到的性能提升并不能可靠地转化为 AL 生态系统的优势,这凸显了建立特定领域评估框架的迫切需求。
BC-Bench: Evaluating Agentic Engineering in a Domain-Specific Language for ERP
BC-Bench: Evaluating Agentic Engineering in a Domain-Specific Language for ERP
Authors: Haoran Sun, Klaus Marius Hansen
Submitted: August 21, 2026
Primary Subjects: Software Engineering (cs.SE), Artificial Intelligence (cs.AI)
arXiv: [arXiv:2608.20851 [cs.SE]]
Authors: Haoran Sun, Klaus Marius Hansen
Submitted: August 21, 2026
Primary Subjects: Software Engineering (cs.SE), Artificial Intelligence (cs.AI)
arXiv: [arXiv:2608.20851 [cs.SE]]
📌 Summary
📌 Summary
智能体工程系统在通用软件基准测试中表现出令人印象深刻的性能,但它们在企业资源计划 (ERP) 领域特定语言 (DSL) 中的有效性在很大程度上仍未得到探索。
Agentic engineering systems have demonstrated impressive performance on general-purpose software benchmarks, but their effectiveness in enterprise resource planning (ERP) domain-specific languages (DSLs) has remained largely unexplored.
为了填补这一空白,研究人员引入了 BC-Bench,这是一个旨在评估智能体工程在 AL(微软 Dynamics 365 Business Central 使用的 DSL)真实世界任务中表现的新型基准测试。该论文的主要亮点包括: * 真实世界任务整理: BC-Bench 具有从两个微软拥有的生产代码库中提取的 101 个手工整理的任务,反映了真实的 ERP 开发工作流程。 * 生态系统适配: 适应 SWE-Bench 方法论,该基准测试解决了 AL 生态系统的独特限制,包括有限的公共资源和复杂的环境配置。 * 先进的评估指标: 除了代码功能外,BC-Bench 还评估测试生成能力,并适应包含视觉上下文的多模态问题陈述。 * 主要发现: 对多个前沿模型和智能体框架的评估(通过多运行指标考虑不确定性)表明,错误修复任务中的模型间性能差异超过了智能体框架之间的差异。至关重要的是,在通用基准测试中观察到的性能改进并不能可靠地转化为 AL 生态系统,这强调了对特定领域评估框架的迫切需求。
To address this gap, researchers introduce BC-Bench, a novel benchmark designed to evaluate agentic engineering on real-world tasks in AL—the DSL utilized for Microsoft Dynamics 365 Business Central. Key highlights of the paper include: * Real-World Task Curations: BC-Bench features 101 manually curated tasks extracted from two Microsoft-owned production repositories, mirroring authentic ERP development workflows. * Ecosystem Adaptations: Adapting the SWE-Bench methodology, the benchmark tackles the distinct constraints of the AL ecosystem, including limited public resources and complex environment provisioning. * Advanced Evaluation Metrics: Beyond code functionality, BC-Bench assesses test generation capabilities and accommodates multimodal problem statements containing visual contexts. * Key Findings: Evaluations across multiple frontier models and agent harnesses (accounting for nondeterminism via multi-run metrics) reveal that model-to-model performance variations in bug-fixing tasks outweigh differences between agent harnesses. Crucially, performance improvements observed on general-purpose benchmarks do not reliably translate to the AL ecosystem, emphasizing the critical need for domain-specific evaluation frameworks.
🔗 Links & Resources
🔗 Links & Resources
- 全文访问:
- 查看 PDF
- HTML 版本(实验性)
- TeX 源码
- 许可证: 知识共享署名 4.0

- 外部书目工具:
- NASA ADS
- 谷歌学术
- 语义学者
- Full-Text Access:
- View PDF
- HTML Version (Experimental)
- TeX Source
- License: Creative Commons Attribution 4.0
- External Bibliographic Tools:
- NASA ADS
- Google Scholar
- Semantic Scholar