What are the different types of agent actions?

• A. Audio agents, visual agents, and sensory agents
• B. Manual actions only
• C. JSON agents, code agents, and function-calling agents.
• D. Random actions

Answer: (C)

The main idea here is how an agent actually interacts with tools and the environment. The best answer names the action interfaces that an agent can use to perform tasks: JSON-based actions, code-based actions, and function-calling actions.

JSON agents rely on structured JSON payloads to specify what the agent should do next or which tool to invoke. This keeps communication clear and machine-readable, making it easy to parse and route actions through a pipeline.

Code agents perform actions by generating and executing code within a runtime. By writing and running code, the agent can carry out complex operations, logic, or interactions that go beyond simple predefined commands.

Function-calling agents use a predefined set of functions or APIs. The agent’s decisions translate into function calls, which can be validated and executed safely, enabling controlled interactions with external services.

Other options describe input modalities or vague behaviors rather than distinct action interfaces. Audio, visual, and sensory are about how information is perceived, not how actions are carried out. Manual actions imply human involvement, which isn’t an inherent action type of autonomous agents. Random actions aren’t a real category of agent actions and don’t describe a meaningful interaction pattern.

So, these three categories—JSON actions, code actions, and function-calling actions—cover the typical, concrete ways an agent can act to accomplish tasks.