In a keynote presentation at the 2026 Mining Software Repositories Emerson Murphy-Hill, a star researcher at Microsoft, presented his view on the role of an empirical software engineering researcher in the age of generative AI. His talk focused on three themes: the durability, differentiation, and dissemination of research.
Durability
With evolving models and methods, research can become outdated. So consider research that auto-updates in the face of new tools. Another strategy involves focusing on people rather than tools. Disaggregating tools (for example agent vs chat vs completion) can also make the research more resilient to evolving models.
Differentiation
It is true that industry researchers have access to proprietary data, large token budgets, and professional developers. On the other hand, academia affords depth, as e.g. Amy Ko did when she read 900 papers before forming a research question. Similarly, many academic research is often funded over several years, whereas in industry six months is an eternity. Academia can also perform the more valuable and actionable qualitative research, which answers “why” questions, rather than the number-based quantitative (”what” questions) research favoured by industry. Finally, academic research affords credibility, because in contrast to industry, academic researchers don’t have to protect or promote the reputation of specific products and service tags.
Dissemination
Generative AI can help us publish our research, e.g. by reviewing or rephrasing our papers, but that’s not a time-saving silver bullet: substantial work is required to review the AI suggestions, choose which ones to adopt, and apply taste. AI can also help in reviewing papers as a discussant for out thoughts, by suggesting potential flaws, explaining parts of the paper, and reviewing our comments. However, also in this case, it is not a substitute for human review. AI-assisted reviews tend to take longer but are more positive.
Comments Post Toot! Tweet
Vibe coding toward the incident horizon (2026-03-02)
An initial analysis of the discovered Unix V4 tape (2025-12-23)
Why I choose email over messaging (2025-09-26)
Is it legal to use copyrighted works to train LLMs? (2025-06-26)
I’m removing the BSD advertising clause (2025-05-20)
The perils of GenAI student submissions (2025-04-11)
Unix make vs Apache Airflow (2024-10-15)
How (and how not) to present related work (2024-08-05)
An exception handling revelation (2024-02-05)
Last modified: Monday, April 13, 2026 5:06 pm
Unless otherwise expressly stated, all original material on this page created by Diomidis Spinellis is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.