Foundation Models and Search-Based Engineering: New Roadmap for AI-Driven Software Development

A quarter-century of work in search-based software engineering (SBSE)—the practice of applying metaheuristic optimization techniques to solve software development challenges—has proven effective across the entire development lifecycle. But the rapid rise of foundation models like large language models now opens fresh questions about how these AI systems can work together with traditional SBSE approaches.

What the Research Covers#

Researchers have published a new roadmap paper on arXiv examining this intersection. The work identifies three main areas of opportunity:

• FMs enhancing SBSE: Using foundation models to strengthen the optimization techniques already in use
• SBSE advancing FMs: Applying search and metaheuristic methods to improve foundation model capabilities
• Deeper integration: Finding synergies between the two approaches to unlock entirely new possibilities

Why This Matters#

The emergence of large language models and other AI foundation models has shifted the landscape of software engineering research. With FMs now capable of understanding and generating code, the opportunity exists to rethink how traditional SBSE methods—which have relied on algorithmic search and evolutionary computation—can adapt to or collaborate with these newer AI capabilities. The roadmap also flags open challenges and identifies emerging domains where these combined approaches might deliver the most value.

Looking Forward#

The authors frame this moment as pivotal: the future trajectory of SBSE in an era dominated by foundation models remains open. Rather than viewing FMs and SBSE as competing approaches, the research suggests exploring how they can amplify one another to tackle harder problems in software development and beyond.