Back to index
OtherSelf-paced, several units with hands-on notebooks, a few weekends of real work·Free

Hugging Face AI Agents Course

4.4

One of the best free ways to actually understand AI agents rather than just talk about them. It is properly hands-on, framework-broad, and demanding enough to leave you able to build, as long as you are comfortable with Python.

What We Liked

  • Completely free and genuinely code-first, you build agents rather than watch slides
  • Covers several frameworks, smolagents, LangGraph and LlamaIndex, not just one vendor
  • Teaches the underlying mechanics of agents, tools, memory and the reasoning loop
  • A real final project and an active community to work through problems with

What Could Be Better

  • You need solid Python and basic API comfort before you start
  • Agent tooling moves fast, so the odd notebook drifts out of date
  • Self-paced and demanding, easy to stall halfway without a deadline
  • Assumes you already understand LLM basics, this is not your first AI course

Detailed review

Everyone wants to learn AI agents right now, the search demand for it is enormous, and most of what is on offer is either a vague conceptual overview or an expensive bootcamp promising to make you an agent engineer in a weekend, so a free course from Hugging Face that just makes you build the things is exactly the antidote. This is code-first from the start. Instead of explaining what an agent is over three videos, it has you construct one, wiring up tools, giving it memory, and watching the reasoning loop actually run, and that hands-on framing is where the understanding comes from. The part I respect most is that it refuses to marry you to a single framework.

You work through smolagents, which is Hugging Face's own lightweight library, but also LangGraph and LlamaIndex, which means you come away understanding the shared concepts underneath rather than just the syntax of one vendor's product, and that transfers no matter what your team ends up standardising on later. It teaches the mechanics that matter, how an agent decides which tool to call, how it holds context, how the loop of think, act and observe actually works, and once you have built that yourself the endless agent hype online suddenly becomes legible. There is a real final project rather than a toy exercise, and an active community around it, which genuinely helps when a notebook breaks and you need a second pair of eyes. I do need to be clear about who this is not for.

You need working Python and at least basic comfort calling an API before you arrive, because this teaches agents, not programming, and if your coding is shaky you will burn your energy on syntax instead of concepts. It also assumes you already grasp the basics of how large language models behave, so it should not be anyone's first ever AI course, it is the second or third one, the one you take once you can already prompt and code a little. Agent tooling moves quickly, so expect the occasional notebook to have drifted out of date against the latest library version, which is the price of learning something this current, and as with anything free and self-paced the lack of a hard deadline is the real risk, this is demanding enough that it is easy to stall in the middle if you do not commit to a schedule. Push through it though, and you end up in the small group of people who can actually build an agent that does something useful rather than just describe one in a meeting, and you got there for nothing, which is hard to argue with.

[ final ]

The verdict.

The course I would point a developer to first when they want to build agents and not just chat about them. It is harder than the polished paid alternatives and you come out actually able to ship something, which is the whole point.