Stage 2 of 10 · 7 lessons · ~2 weeks

Tool calling from scratch

Build the agent loop yourself with one provider API and no framework: define tools, read tool calls, return results and errors, handle parallel calls, then read a second provider to see what is universal.

Start: Defining tools

Lessons· one lesson for each topic, in this order

Do this

Get an API key from one provider and set a monthly spending limit.
Define three tools with clear names, descriptions and schemas.
Write the loop with the official SDK and print every call and result.
Return error results for a missing record and a timeout, and watch the model recover.
Trigger a turn with parallel calls and return all results in one message.
Add a step limit and test that it stops a looping run.
Read a second provider's tool-calling guide and write down the field mapping.

Start with· follow one of these

GuideFree
Claude Tool Use Documentation
Anthropic · Intermediate
Read Overview, Define tools and Handle tool calls, then follow the tutorial that grows one call into a full loop
GuideFree
OpenAI Function Calling Guide
OpenAI · Intermediate
The same loop on the Responses API: read it after your first provider to see what is universal

Practise

Anthropic Courses
Free notebooks: work through the tool use course, which builds tools and a loop exercise by exercise
OpenAI Cookbook
Runnable function-calling examples to compare with your own loop

Keep handy

JSON Schema: Getting Started
The Getting Started guide covers every schema keyword you need for tool arguments

Also worth your time

Writing Effective Tools for AI AgentsRead
How to name, describe, scope and evaluate tools so models use them correctly

Before moving on: A framework-free agent in your language with three real tools, error results the model can act on, parallel call handling, a step limit and a printed trace of every call, plus a note mapping it to a second provider's API.

Tauseef's advice

Do this stage even if you plan to use a framework. Fifty lines of your own loop will teach you more about agents than a week of framework tutorials. If you want a compact reference while you work, Sefism's LLM APIs cheat sheet at /cheatsheets/llm-apis/ covers the request shapes across providers; this stage goes further into the loop itself.