Interaction Design and Software Engineering (IDSE) @ ICSE’26

In large-scale agile systems development, the lack of a unified requirements engineering (RE) process is a major challenge. This five-year longitudinal case study with Grundfos AB derived RE principles through qualitative data collection spanning more than 25 sprints, ~320 weekly synchronisation meetings, and seven cross-company workshops (2019–2024). Six key RE principles were identified and retrospectively validated through focus groups with senior leaders and expert evaluation with Bosch, Ericsson, and Volvo Cars.

Explores the nature of folklore in software engineering — the narratives, myths, rituals, and informal knowledge that circulate among practitioners — and provides conceptual foundations for studying this phenomenon.

Examines how software practitioners perceive socio-emotional gaps in human-AI collaboration, based on semi-structured interviews with ten practitioners, and identifies the functional dimensions needed to bridge those gaps.

Investigates using large language models to automatically summarize execution logs from automotive software systems, with the goal of accelerating debugging workflows for engineers.

Presents REST-at, a tool that uses large language models to automatically establish and maintain traceability links between natural-language requirements and test cases.

Explores whether large language models can generate accurate and useful natural-language explanations for boundary test cases, supporting comprehension and review of automatically generated tests.

A multi-organisational case study examining data annotation errors in AI-enabled perception systems for automotive applications, involving six companies and four research institutes across Europe.

Reports on a practitioner study investigating how data leakage manifests in machine learning pipelines for automotive perception systems and how development teams detect and mitigate it.

Addresses a disconnect between high-level AI policy and actual organisational practice. Through a case study of a large Engineering, Procurement, and Construction company plus five supplementary interviews with technology leaders, identifies four governance models: High-Assurance, Cautious, Enterprise-Driven, and Developer-Led. Shows how organisations select models based on their risk profiles and how governance choices shape resulting technical system designs.

Addresses how domain expertise concentrated with few specialists creates organisational constraints. Proposes a software engineering framework that captures human domain knowledge for AI agents in simulation data visualisation by augmenting an LLM with a request classifier, a RAG system for code generation, codified expert rules, and visualisation design principles. Evaluation across five scenarios with 12 evaluators demonstrates a 206% improvement in output quality, with the agent achieving expert-level ratings in all cases versus the baseline.

Challenges common assumptions about refactoring in research software contexts, arguing that purely technical improvements are insufficient for achieving meaningful software quality in academic settings.