AI Agent

A system where an LLM is given access to tools and the autonomy to decide when and how to use them to accomplish a goal. Unlike a chatbot (which responds to prompts like a particularly eloquent Magic 8-Ball), an agent can break a task into steps, call external tools, evaluate results, adjust its approach, and iterate until the task is complete. Or until it gets confused and apologizes. Both outcomes are common.

Example: You ask an agent to "find the llms.txt file for example.com and check if it's spec-compliant." The agent might: (1) use an HTTP tool to fetch example.com/llms.txt, (2) use LlmsTxtKit to parse and validate the result, (3) report findings and suggest corrections. You gave it a goal; it figured out the steps.

Why it matters for writers: Agents are the primary consumers of MCP servers and tool definitions. If you're writing documentation for an MCP tool, your audience includes both human developers configuring the tool and AI agents that will read the tool's schema to decide how to use it. Writing clear, unambiguous tool descriptions is a form of technical writing that directly affects agent performance. Your prose is, quite literally, an API now.

Related terms: Model Context Protocol · Tool Use · Orchestration