AI agents and tool use: how agents plan and call tools
12 steps · about 8 h 25 min · Intermediate
- How tool use works - Claude Platform Docs (doc, Anthropic, 15 min)
It starts with the core idea: the model emits a structured request, your code runs it, and the result goes back into the conversation.
- Function Calling 101: How AI Agents Execute Real-World Actions (article, DEV Community, 15 min)
A short Python example shows the tool schema, the model's JSON call and the dispatch code in one place.
- LLM Powered Autonomous Agents (blog, Lilian Weng, Lil'Log, 45 min)
It gives the standard map of an agent as an LLM plus planning, memory and tool use, so you can see where tool calls fit.
- ReAct: Synergizing Reasoning and Acting in Language Models (article, Google Research, 15 min)
It explains how interleaving thoughts, actions and observations lets a model plan, which is the idea behind most agent loops.
- ReAct: Synergizing Reasoning and Acting in Language Models (reference, arXiv (Yao et al.), 45 min)
Read the original paper to see the experiments showing that reasoning helps acting, and the method's limits.
- Building effective AI agents (article, Anthropic, 30 min)
It separates workflows from true agents and tells you when the extra complexity is worth it.
- Tool use with Claude (doc, Anthropic, 30 min)
It covers real API details such as client and server tools, parallel calls and tool_result handling, so you can write the loop yourself.
- Writing effective tools for AI agents (article, Anthropic Engineering, 30 min)
Tool descriptions decide whether an agent picks the right tool, and this shows how to design and evaluate them.
- Hugging Face Agents Course (tutorial, Hugging Face, 2 h)
Unit 1 teaches thoughts, actions, observations and tools in simple Python, with quizzes. The certificate needs a free account, but the lessons are open.
- Introduction to smolagents (tutorial, Hugging Face, 2 h)
You build code agents and tool-calling agents in a lightweight library, which turns the concepts into working projects.
- What is the Model Context Protocol (MCP)? (doc, modelcontextprotocol.io, 15 min)
MCP is the open standard for plugging tools into agents, so this is the natural next step once you know how a single tool call works.
- Architecture overview - Model Context Protocol (doc, modelcontextprotocol.io, 25 min)
It shows how clients list tools, pass them to a model and execute the calls, which ties MCP back to the loop you built.