Developers of concurrent software strive for two contradictory goals. On the one hand, the software should be reliable, which requires to carefully synchronize concurrent accesses to shared data. On the other hand, the software should be efficient, which is difficult to realize in the presence of unnecessary synchronization operations. To help developers balance these two goals, this talk explores program analyses that detect correctness and performance problems. The presented analyses are automatic, in the sense that they run on existing software with minimal effort, and precise, in the sense that all reported problems are true positives. The key idea is to combine techniques for generating concurrent tests with dynamic analyses that detect problems even though there is no explicit specification. We present two implementations of this idea that detect problems in widely used thread-safe classes, and discuss open challenges to be addressed in future research.