Hugging Face built an agent course that teaches the competition

What it does The repository hosts the Hugging Face Agents Course, a free MDX-based curriculum that moves from agent fundamentals through framework-specific implementation to a final, benchmarked project. It covers how agents use LLMs, function calling, and special tokens, then dives into building with smolagents, LangGraph, and LlamaIndex. The course wraps up with automated evaluation and a leaderboard for student submissions.

The interesting bit The course gives roughly equal billing to smolagents, LangGraph, and LlamaIndex—an unusual move for a framework vendor—and even includes a bonus unit on fine-tuning LLMs for function calling. The finale treats agent building as an engineering challenge rather than passive reading: students create, test, and certify their work against an automated benchmark.

Key highlights

• Four core units plus three bonus tracks, including observability, function-calling fine-tuning, and a Pokemon-playing agent unit
• Explicit coverage of smolagents, LangGraph, and LlamaIndex
• Final project includes automated evaluation and a student leaderboard
• Source content is open and written in MDX
• Prerequisites are minimal: basic Python and passing familiarity with LLMs

Verdict Worth bookmarking if you want a structured, framework-agnostic on-ramp to agent development. If you are already running production multi-agent systems, you are not the target audience.