Agentic Loop

When you invoke Claude Code (ask it to do something) it uses an agentic loop to perform the requested task.

Flow: Your Prompt → Model → [Gather Context → Take Action → Verify Results] (loop) → Done
While executing the three phases it will perform calls to tools (tool calls) to aid it accomplishing its tasks.
Tools are just functions of code that call to external programs.
It will keep looping, self-reflecting to determine if it has met the goal.
You can interrupt the loop at any time: interrupt, steer, or add context, to help correct the agent if you think it's misguided.