AI agents use reinforcement learning (RL) by interacting with their environment and receiving feedback in the form of rewards or penalties. This feedback helps the agent to learn optimal strategies over time, enabling it to maximize its cumulative reward in future scenarios.