References: How tool use turns a model into an agent

Source material

Source curriculum (structural mirror, cited as further study):
• Microsoft, "Tool Use Design Pattern" (AI Agents for Beginners, Lesson 04)
  Author: Microsoft Cloud Advocates
  Lesson page: https://github.com/microsoft/ai-agents-for-beginners/tree/main/04-tool-use
  License: MIT
Clawdemy's lessons are original prose that follows the pedagogical arc of this
source. We do not reproduce or transcribe it; we cite it as the recommended
companion. All rights to the original materials remain with their authors.

Direct link to the source lesson: github.com/microsoft/ai-agents-for-beginners/tree/main/04-tool-use.

Read this next

• Tool Use Design Pattern (Microsoft) by Microsoft Cloud Advocates. The practitioner version of this lesson, with runnable code samples and framework examples. MIT-licensed, free to read and fork. Start here if you want to see the four-step exchange in working code rather than pseudocode.

Going deeper on function calling

A short, durable list. Each is a primary source for how a real model provider formalizes the tool-call exchange this lesson describes.

• OpenAI: Function calling guide. The provider docs for defining tools as schemas and parsing the tool calls a model emits. The concrete shape of step 1 (the menu) and step 2 (the request).
• Anthropic: Tool use with Claude. The equivalent specification for Claude models. Useful for seeing that the four-step exchange is provider-agnostic: the same shape, different field names.
• Model Context Protocol. An open standard for how tools are described and called across models and applications. Relevant once you care about reusing tools across different agents; the subject of a later lesson in this track.

Adjacent topics

Where this leads inside this track.

• Choosing an agent framework. The next lesson. If the menu, the loop, and the watch-and-execute machinery are always the same shape, the question becomes whether to build them yourself or use a framework that provides them. We weigh hand-rolling against frameworks and how to choose.
• The tool-use design pattern in depth. A few lessons ahead. This lesson covered the mechanism of a single call; that one covers how to define a tool well, so the model reliably knows when and how to use it.
• Securing agents. Later in the track. A model that can trigger real actions through tools is also a new attack surface. That lesson covers what can go wrong when a tool call is influenced by untrusted input.