Define an agent based on the transcript's summary.	An agent is a language model that has access to a set of tools and is executed repeatedly until a goal is achieved.
How does an agent gain knowledge about its operating environment?	It relies entirely on using tools to inspect the environment, rather than relying on RAG or detailed prompts.
What is the primary function of most tool calls in agents like Claude Code or Computer Use?	They are intended to gather information about the environment (e.g., the codebase or screen).
Describe the operational pattern of an agent.	Agents run tools in a loop until a goal is achieved or an unrecoverable error occurs.
What design principle should guide the selection of tools for an agent?	The agent should be given a small, focused set of tools, where each tool has a very clear purpose.
Why should agents only be used for high-value problems?	To ensure that if the agent makes a mistake, the failure does not have a large economic or safety impact.
What is the agent's relationship with outside knowledge?	It has no knowledge of the outside world and relies entirely upon being given a set of tools to inspect the environment.
What is the best way to ensure that an agent is effective?	By creating evaluation criteria and evaluating the agent against those criteria (evaluations).