Is Search-Based Unit Test Generation Research Stuck in a Local Optimum?

Research in search-based unit test generation has seen steady development in recent years. New techniques and tools have been developed, and empirical evidence has been collected on the wide-ranging capabilities of search-based algorithms for unit test generation, and for many related software engineering practices. But why are developers not generating all their tests automatically yet in practice? In this paper, we argue that there may be deceiving local optima in the research landscape on search-based unit test generation, and we outline specific challenges and opportunities to escape these local optima.