rqlite is a lightweight, open-source, distributed relational database built on SQLite and Raft.
Antithesis has published a post on how they use AI Agents to inject bugs in rqlite, then check whether their tests catch them. Along the way their tests found three real bugs in rqlite, which I have since fixed.
The idea is one I’ve pushed for years: a test you’ve never seen fail is a test you can’t trust. I used to flip the expected result by hand. They go further, breaking the code itself and letting an agent do the work. Of the 13 properties they tested, their tests caught the bugs the agents planted. For the other two, the tests missed them, so the harness needs more work there.