Identify defects specific to coding-agent-assisted development

Determine which defects observed in the repository catalog and MateriApps retrieval-agent development were specific to coding-agent use and which would also arise in conventional software development.

Background

The study is an experience report produced by a small development group and does not include a controlled comparison between coding-agent-assisted and conventional software development. Consequently, the authors cannot attribute the observed defects specifically to coding agents. Establishing that attribution would require a controlled comparative study.

References

Even if they do, it remains unclear whether such code exhibits the same fault patterns as human-written programs.

— How effective are traditional test criteria at detecting bugs in large language models generated code?  (2609.09315 - Hamidi et al., 8 Sep 2026) in Section 1, Introduction

In particular, we did not perform a controlled comparison between agent-assisted and conventional software development. We therefore cannot determine which of the observed defects were specific to coding-agent use and which would also arise in conventional development.

— Building a research-software catalog with a coding agent: from hackathon prototype to public deployment  (2609.04711 - Yoshimi et al., 4 Sep 2026) in Section 4.4, subsection “Limitations and future work”