“Combining Supervised Fine-Tuning (SFT) with Reinforcement Learning (RL) has become a common approach for producing production-ready AI agents.”