A group of researchers has published a roadmap arguing that a 25-year-old branch of software engineering research needs to make peace with large language models.
The paper focuses on search-based software engineering, or SBSE, which uses metaheuristic search algorithms to automatically tackle problems like test generation, bug repair, and code optimization. The authors survey how SBSE and foundation models currently intersect, then map out three directions: using foundation models to make SBSE techniques smarter, using SBSE's search methods to improve foundation models themselves, and building hybrid systems that combine both. There's no new tool or benchmark here - it's a literature review and opinion piece meant to set an agenda for other researchers. It's also already the fourth revision of the paper, a sign the authors keep updating their pitch as the field moves fast underneath them.
SBSE has quietly powered real engineering work for two decades, from automated test-suite generation to program repair, without ever getting the attention now lavished on LLMs. This roadmap reads as much like an argument for that community's continued relevance as it does a technical proposal - a reasonable move when funding and attention chase whatever has "foundation model" in the name.
Roadmap papers are cheap to write and easy to cite; the real test is whether anyone actually builds and benchmarks the hybrid tools this one only sketches out.