Where prompt engineering shapes one instruction and context engineering shapes what the model can see, loop engineering shapes how many times and under what conditions the agent acts, checks its own work, and decides it's done — a discipline that grew alongside long-running autonomous coding agents like Claude Code.