Autonomous AI Agent
Principle
A plan–execute–verify loop: decompose the goal into subtasks, select tools, act, then check the result and decide the next action.
Purpose
Work that proceeds to completion without step-by-step instruction, separating routine execution from exception handling.
Implementation
A prototype with separate planner, executor and verifier. It includes re-planning on failure and an approval request path.