What are examples of observations in AI agents?

• A. User prompts only
• B. Training data only
• C. Error messages, API responses, sensor readings, and system logs.
• D. Environment variables only

Answer: (C)

Observations are the signals the agent perceives about its current situation. They come from the environment and internal systems and are used to update the agent’s understanding of state and to guide decisions. Examples include sensor readings from a robot’s cameras or lidar, responses returned by external services via APIs, and error messages or system logs that reveal what happened during execution. This kind of data tells the agent what is actually happening right now, beyond what it was told to do or what it was trained on.

In contrast, user prompts are instructions from a person that set goals or tasks for the agent to achieve; they’re inputs the agent responds to but aren’t perceptual data about the world. Training data is historical information used to train the model, not live observations the agent uses to act. Environment variables are configuration settings, not dynamic percepts the agent senses.