During his short tenure, Agarwal contributed significantly to Meta’s AI research, including enhancing “thinking models” through RL scaling, using synthetic data in mid-training, and improving on-policy distillation methods. He expressed gratitude toward his teammates for their collaboration.