Inference

The process of running a trained AI model to get a response. Every agent action runs inference, which costs money and time. Reducing inference calls — through caching, batching, or pre-computed shared state — is one of the highest-leverage optimizations in agent systems.