Autonomy is an agent’s capacity to determine and continue its own actions instead of following a predetermined sequence. The agent assesses the current context, chooses a tool or action, observes the result, and repeats until the task is complete. section 1.3.2
A practical autonomous agent follows a continuing control loop:
- Assess the current state and progress toward the goal. section 1.3.2
- Choose whether a tool helps, then select the tool and its parameters. section 1.3.2
- Execute the action, observe its result, and decide whether to continue or stop. section 1.3.2
- Ask for clarification only when the user’s intent is genuinely ambiguous; otherwise execute immediately. section 2.3.2
- Extend the available capabilities by generating and executing a custom tool when predefined tools cannot handle the requirement. section 1.3.2
Weather request
For “What’s the weather like in Seoul today?”, an autonomous agent searches immediately when a web-search tool is available, rather than asking whether it should search. A more complex task may continue through many tool calls before producing its answer. section 2.3.2 section 1.3.2
