Farnaz Fotrousi

Large Language Models (LLMs) are increasingly applied to software engineering (SE), yet their potential for autonomous, role-oriented collaboration remains largely underexplored. Understanding how multiple LLM-based agents coordinate, maintain role alignment, and converge on solutions is critical for SE, as naively allowing agents to interact does not reliably lead to correct or stable outcomes. Recent empirical studies show that unstructured or poorly understood interaction dynamics can result in error propagation, premature consensus on incorrect solutions, or prolonged disagreement that prevents convergence, even when correct partial solutions are present early in the interaction. As an initial step towards addressing this underexplored area, we undertake a systematic analysis of conversations between two agents, a Designer and a Programmer across 12 model combinations from 7 open-source LLMs (Gemma 2, Gemma 3, LLaMA 3.2, LLaMA 3.3, DeepSeek-R1, MiniCPM, and Qwen3). Our systematic approach reveals three key dimensions of multi-agent interaction: efficiency (the speed and stability of convergence), consistency (the degree of role alignment visualized by BLEU and ROUGE), and effectiveness (the extent of compilation success and error resolution). Results show that the DeepSeek-R1:DeepSeek-R1 pair was unique in converging to the correct solution from the very first iteration and sustaining it consistently to the final iteration, while LLaMA 3.2:LLaMA 3.2 and Qwen3:Qwen3 demonstrated strong Designer:Programmer role alignment despite of diverging from the correct solution. The other pairs deviated from the task, never to converge to a result. These findings advance understanding of agentic programming and highlight the need for further research on understanding and calibrating convergence and stop conditions essential for future autonomous SE.

Context: Software evolution ensures that software systems in use stay up to date and provide value for end-users. However, it is challenging for requirements engineers to continuously elicit needs for systems used by heterogeneous end-users who are out of organisational reach. Objective: We aim at supporting continuous requirements elicitation by combining user feedback and usage monitoring. Online feedback mechanisms enable end-users to remotely communicate problems, experiences, and opinions, while monitoring provides valuable information about runtime events. It is argued that bringing both information sources together can help requirements engineers to understand end-user needs better. Method/Tool: We present FAME, a framework for the combined and simultaneous collection of feedback and monitoring data in web and mobile contexts to support continuous requirements elicitation. In addition to a detailed discussion of our technical solution, we present the first evidence that FAME can be successfully introduced in real-world contexts. Therefore, we deployed FAME in a web application of a German small and medium-sized enterprise (SME) to collect user feedback and usage data. Results/Conclusion: Our results suggest that FAME not only can be successfully used in industrial environments but that bringing feedback and monitoring data together helps the SME to improve their understanding of end-user needs, ultimately supporting continuous requirements elicitation.

Crowd-based requirements engineering (CrowdRE) is promising to derive requirements by gathering and analyzing information from the crowd. Setting up CrowdRE in practice seems challenging, although first solutions to support CrowdRE exist. In this paper, we report on a German software company's experience on crowd involvement by using feedback communication channels and a monitoring solution for user-event data. In our case study, we identified several problem areas that a software company is confronted with to setup an environment for gathering requirements from the crowd. We conclude that a CrowdRE process cannot be implemented ad-hoc and that future work is needed to create and analyze a continuous feedback and monitoring data stream.

Collecting and using user feedback as a method to support requirements engineering, might undermine user rights. This becomes apparent when looking at related areas, e.g., research in user experience, where collecting user feedback also plays an important role. In such settings, researchers need to ensure that the stakeholders' rights and integrity are respected. This paper identifies and discusses some of the ethical challenges and issues a researcher can face, using an example case. Focusing on user feedback, this case can serve as an example for CrowdRE, i.e. several of our findings might apply to CrowdRE in general. However, further research is needed as our work mainly reflects the challenges experienced by the authors of this paper.

Feedback communication channels allow end-users to express their needs, which can be considered in software development and evolution. Although feedback gathering and analysis have been identified as an important topic and several researchers have started their investigation, information is scarce on how software companies currently elicit end-user feedback. In this study, we explore the experiences of software companies with respect to feedback gathering. The results of a case study and online survey indicate two sides of the same coin: on the one hand, most software companies are aware of the relevance of end-user feedback for software evolution and provide feedback channels, which allow end-users to communicate their needs and problems. On the other hand, the quantity and quality of the feedback received varies. We conclude that software companies still do not fully exploit the potential of end-user feedback for software development and evolution.