DAY 2 – SE.next: Research in the Agentic Era
If AI teammates can run our experiments and replicate our results, and the thing we study is fundamentally changing, what exactly is the researcher’s job?
Here is the uncomfortable truth: much of what SE researchers do today, from building tools to running empirical studies to writing papers, is increasingly within reach of AI agents. But the reckoning runs deeper than our methods, because AI teammates are also rewriting how software gets built, which means the phenomena we study are shifting under our feet. Research problems we have spent decades refining may be quietly losing relevance, even as entirely new classes of problems emerge that we have barely begun to frame. Questions for the room:
- What does “reproducibility” mean when models change weekly and agents behave non-deterministically?
- If AI agents can generate and evaluate solutions at scale, does the researcher’s role shift entirely to problem formulation and judgment?
- Which classic SE research problems are being quietly solved by agents, and what does that free us up to study instead?
- What new phenomena, such as human-agent delegation, specification quality, or trust calibration, deserve the same rigor we once gave to version control and testing?
- When an AI teammate produces the code, should research focus shift from the code itself to the intent, the specification, and the judgment behind it?