How Claude Code Actually Works: The Agentic Loop, Not Autocomplete

Autocomplete predicts the next few tokens from the file you're looking at. Claude Code runs a loop: decide what to do, do it with a real tool (a file read, a shell command, an edit), look at the actual result, decide what to do next — until the task's done or it needs you. This piece traces that loop step by step: what happens tool-call by tool-call when it fixes a bug, how permission modes govern autonomy, and where CLAUDE.md/memory/settings actually live. Claude Code Interview