Using RL to Double an Agent's Effectiveness in Production Debugging