In AI agents, how are Tools typically represented to the LLM?

• A. Tools are encoded as binary executables
• B. Tools are hidden in the model's weights
• C. Tools are described textually in the system prompt for the LLM to use
• D. Tools are only used during training

Answer: (C)

Tools are described textually in the system prompt for the LLM to use. This setup gives the model a clear, external interface for capabilities like querying databases, calling APIs, or performing specialized computations without altering the model’s internal weights. Each tool is defined with a name, a brief description of what it does, and the expected input and output formats. When the agent needs to perform a task that requires a tool, the LLM uses that defined interface to decide when to call the tool and what arguments to pass. The actual tool execution happens outside the LLM, typically by a tool runner, and the results are fed back for the LLM to incorporate into its next steps. This separation makes it easy to add or update tools without retraining, and it keeps the model focused on reasoning and planning rather than manipulating binaries or internal parameters. Tools are thus used at inference time, not only during training.