The official Salesforce AI Research account announced its paper UserRL: Training Interactive User-Centric Agent via Reinforcement Learning. The work presents a unified RL framework that uses eight gym environments to train agents for multi-turn user interactions. Listed authors are qiancheng1231, LiuZuxin, aksh_555, _Jason_Q, Zhiwei Liu, and Haolin Chen. The preprint is available at arxiv.org/abs/2509.19736. The post comes directly from the company's AI research division and describes the paper's focus on moving beyond static benchmarks.