Computer-using agents may spend tokens interpreting the screen and choosing every next click. If the same path must run repeatedly, MacroLoop can record that deterministic portion and replay it locally.
Token savings come from removing repeated decisions
An agent can decide which batch or workflow is appropriate once. MacroLoop can then handle the fixed clicks and keys without asking the model to analyze every screen again.
Do not confuse navigation with model usage
Prompts submitted to an external AI service may still consume tokens under that service’s plan. MacroLoop reduces the agent-control overhead; it does not make model inference free.
THE TAKEAWAY
Use AI where the next action is uncertain and MacroLoop where the next action is already known.OFFICIAL REFERENCES