Autonomy is the capacity of an agent to decide how to pursue a goal rather than follow a prescribed sequence of actions. It selects tasks, orders them, interacts with the outside world, and evaluates whether its actions support the goal. section 9.1.1
In an LLM-based agent, autonomy depends on:
- A prompt establishes the goal, while callable functions provide ways to retrieve information or affect external systems. section 9.1.2
- The LLM chooses which function to call, when to call it, and what data to pass, using returned information to continue solving the task. section 9.1.2
Function selection instead of a fixed workflow
An agent given a complex task might call a web API, scrape a page, or run web automation, choosing only the functions needed rather than executing every available function in order. section 9.1.1 section 9.1.2
Figure 9.1 An outline of an LLM agent using function calling
Use the function-calling outline to trace how a goal, selected function, external result, and subsequent decision form an autonomous loop. section 9.1.2
