Agentic loop
An agentic loop is the cycle in which a model calls a tool, reads the result, and decides what to do next, repeating until it produces an answer.
This loop is what separates an agent from a chat. It is also what makes agent sessions expensive in a way chats are not: every tool result joins the context permanently, and the whole accumulated context is re-sent on the next iteration.
A loop with six iterations therefore does not cost six times one iteration. It costs closer to the sum of a growing series, because iteration six carries everything iterations one through five produced.
That growth is why exploration is the dominant cost in agent work. Finding a file takes iterations; being told which file takes none.