# Hugging Face's free AI Agents course: the 30% pass mark is the lesson worth learning

> Four units, three frameworks, 3–4 hours a week. The course's real value is that it makes you put your agent on a benchmark and accept a pass mark far lower than the quiz's.

Bản gốc: https://fdetimes.net/en/books-courses/hugging-face-ai-agents-course-review/

To earn the Unit 1 certificate in its AI Agents course, Hugging Face requires you to score at least 80% on the final quiz. To earn the certificate for completing the whole course, the agent you build only needs to score above 30% on the final assignment's benchmark. Those two numbers say almost everything about the course, and about the FDE job.

Answering theory questions correctly is easy. The much lower bar on the benchmark suggests that getting an agent to actually solve a problem is far harder. For a Forward Deployed Engineer, the question that matters is not what you know about tool calling, but whether your agent works on real data.

If you are a developer with 2–8 years of experience looking to move into an FDE role, the course is worth a few weeks. But you do not have to take it in the order it is laid out.

## Four units, two certificate levels

The course is officially titled "Welcome to the 🤗 AI Agents Course". It launched in 2025 at huggingface.co/learn/agents-course, is maintained by Ben Burtenshaw and Sergio Paniego, and was written and developed by Joffrey Thomas and Thomas Simonini. The whole repository is open under the Apache-2.0 licence.

Hugging Face describes the goal as learning AI agents "in theory, design, and practice" using three frameworks: smolagents, LlamaIndex and LangGraph. The prerequisites are basic Python and a foundational understanding of LLMs. Each chapter is designed to take a week, at roughly 3–4 hours.

After onboarding come four units: Unit 1 on foundations, Unit 2 on frameworks, Unit 3 on use cases, and Unit 4, a final assignment scored on a benchmark. Bonus units cover fine-tuning for function calling, observability and evaluation, and an agent that plays Pokemon. Learners can also share their agents on the Hub and compete against each other.

The certificates are entirely free and have no deadline. The foundational level is awarded after Unit 1; the completion level also requires a use-case assignment and the final challenge.

| | Foundational certificate | Completion certificate |
|---|---|---|
| Requirement | Finish Unit 1, score at least 80% on the final quiz | Unit 1 + use-case assignment + final challenge, benchmark above 30% |
| What it proves to an employer | You understand agent concepts | You have shipped an agent measured by a number |
| How to put it on your CV | One short line | With your benchmark score and how you pushed it up |

## What order should you study in if your goal is FDE?

The advice is not to give every unit equal weight. If you have already called an LLM API, you can move through Unit 1 quickly: the aim at this stage is to get the foundational certificate and get on to the practical work.

Unit 2 teaches three frameworks. Pick one and go deep; skim the other two to understand how they differ. What is worth taking away is the ability to recognise the same agent concept written three different ways, not memorising all three APIs.

Unit 3, on use cases, is where you should start writing code without looking at the examples. Treat it as a customer discovery exercise: read the brief as you would a client's requirements, then ask what the input data looks like, which tools the agent is allowed to call, and what counts as "done".

Then comes the most important part, and this is where to change the order: take the bonus unit on observability and evaluation **before** Unit 4, not after. You cannot raise a benchmark score if you cannot see at which step the agent goes wrong, which tool calls are unnecessary, and where it stops too early. That loop of run, measure, fix, run again is a habit worth building now.

**Điểm mấu chốt:** The quiz asks what you know; the benchmark asks what your agent can do. FDEs are paid for the second.

The bonus unit on fine-tuning for function calling is worth taking if you have already seen your agent call tools in the wrong format too many times. Save the Pokemon unit for a free weekend.

## What does the 30% figure teach?

That is also why it is worth noting that the course puts smolagents, LangGraph and LlamaIndex together in a single unit. The arrangement implies that the concepts are the lasting asset, while libraries can be swapped out at any time.

Seen in that light, the 30% threshold should not be read as leniency. It can be understood as a sign that agents on real problems are still far from perfect, and that your value lies in pushing the number up, step by deliberate step.

That is also how to write your CV. Do not stop at "completed Hugging Face's AI Agents course". State what percentage your agent scored on the Unit 4 benchmark, what you changed to take it from one level to the next, and include a link to the Hub so the interviewer can run it themselves.

When reading FDE job descriptions, look for words such as evaluation, observability and tool calling. If they appear, you already have a project to talk about.

The certificate with your name on it is free. The number on the benchmark, and the story of how you pushed it up, is what FDE hiring managers want to hear.

**Thử ngay tuần này:**

- Sign up at huggingface.co/learn/agents-course, finish Unit 1 and take the final quiz to earn the foundational certificate
- Pick one of the three frameworks (smolagents, LangGraph or LlamaIndex) and rewrite one Unit 2 example in it, without copying the sample code
- Open the observability/evaluation bonus unit and write down three questions you will need to answer when your agent fails on real data

## Nguồn

- [Agents Course documentation (Hugging Face)](https://huggingface.co/learn/agents-course/unit0/introduction)

- [huggingface/agents-course (GitHub)](https://github.com/huggingface/agents-course)

- [Agents Course documentation – Get your certificate (Unit 1)](https://huggingface.co/learn/agents-course/unit1/get-your-certificate)

- [Agents Course documentation – Get your certificate (Unit 4)](https://huggingface.co/learn/agents-course/en/unit4/get-your-certificate)

- [HuggingFace AI Agents Course (I Programmer)](https://www.i-programmer.info/news/105-artificial-intelligence/17884-huggingface-ai-agents-course.html)