AI agency is decision-making in which an AI model’s outputs guide the selection of the system’s next action. Code agency instead leaves that choice to the executing program. section 1.1.1
Two views differ on how much decision-making an agent must delegate to AI:
- The primacy view requires the AI model to control most decisions, such as planning with an LLM or selecting actions with a vision-language-action model. section 1.1.1
- The indispensability view counts software as an agent when AI is essential to its operation, even if fixed code orchestrates the overall process. This favors tractable control in reliability-critical applications. section 1.1.1
Personal assistant
Code can route a request through speech recognition, an LLM, calendar tools, and a final response. Agency shifts between code and model along this workflow. section 1.1.1
