Welcome To The Agents Course! Introduction to the Course and Q&A

Overview

The livestream introduces Hugging Face’s free, open-source Agents Course, explains how its modules and assessments will work, and answers questions from prospective students. The course aims to teach both the theory and practical skills needed to build AI agents, starting with fundamentals and progressing to frameworks and real-world-style projects.

Main Ideas and Course Structure

• The focus is on agents, not just models or a particular framework. An LLM is the model at the core of an agent; an agent is a larger system that uses the model to make decisions and take actions, often using tools. Students will build agents using existing models rather than train or modify the underlying models.
• The course combines theory and practice. Its stated goal is to give learners enough conceptual grounding to work with different agent libraries, alongside hands-on experience building agents and exploring work shared by the community.

Planned Curriculum

1. Introduction and fundamentals: Theory plus a practical introduction to building an agent with smolagents. The first module was already available at the time of the livestream.
2. Bonus material: A planned module on fine-tuning.
3. Frameworks: Work with tools including LangChain, LangGraph, LlamaIndex, and smolagents. The presenters said the course would cover a limited number of frameworks to keep it manageable and keep the emphasis on agent concepts.
4. Use cases: Apply agents to practical scenarios.
5. Final assessment: Build an agent and compare its performance with other students’ agents on a leaderboard, using a GAIA-style evaluation.

Pace, Workload, and Certificates

Modules were expected to be released about every two weeks. The presenters suggested budgeting roughly three to four hours per week, while noting that time needs vary and learners can work at their own pace. The May 1, 2025 completion date was described as a cohort-organizing target rather than an inflexible cutoff.

The course offers two certificates: one for foundational knowledge and another for completing the broader course requirements, including the final assignment. The fundamentals test can be retaken; the presenters said an 80% score was required. The final assessment also allows repeated attempts and has a passing threshold. Certificates are linked to a Hugging Face profile, and learners can choose the name displayed during certification.

Recommended Approach for Learners

• Check your foundations. Basic Python knowledge is recommended for the coding sections, along with some familiarity with LLMs. The first module’s fundamentals assessment does not require coding, so non-coders—including people interested in the topic from a decision-making or project-management perspective—can still engage with that material.
• Use the course materials and self-checks. Review the readings and take the non-assessment quizzes to check your understanding. If you do not pass a test, revisit the material and try again rather than simply guessing through it.
• Keep a regular schedule. Set aside time each week to stay aligned with the course releases, while adapting the pace to your circumstances.
• Learn with the community. Join the course Discord to discuss concepts, find classmates, and get updates. The presenters highlighted the community’s role in helping learners support one another.
• Explore beyond the required exercises. Once you understand the material, try extending the examples, investigating what different model sizes can handle, and thinking about agent use cases you might want to build.
• Use the right help channel. The presenters recommended Discord for course discussion and conceptual questions, and Hugging Face Hub discussions for technical problems involving tests, certificates, code, or tool errors.
• Follow course updates. Register for the course and follow its Hugging Face organization and GitHub repository. The course content was described as open source and intended to evolve with community contributions and changes in the field.

Tools, Prerequisites, and Practical Details

• A computer and a free Hugging Face account are sufficient; a paid account or GPU is not required.
• Many exercises use Hugging Face Spaces, notebooks, and serverless inference APIs, so learners can do much of the work through a browser rather than setting up a complex local environment.
• The course is currently Python-based; JavaScript or TypeScript is not supported in its current format.
• The course is in English at the time of the livestream. Community translations were suggested as a possible contribution.
• The presenters said that small and large language models can suit different tasks; model choice depends on the complexity of the task and available hardware.
• The course is not a continuation of the classic deep reinforcement learning course. The speakers described LLM agents and reinforcement-learning agents as distinct topics.
• Production deployment is not a dedicated course module at that point, though practical use cases and some agent-security considerations were expected. The presenters invited feedback about adding production-focused bonus content.
• Vision-language models may appear as tools in some agent scenarios, such as web browsing.

Other Announcements and Links

• The course was available through the Hugging Face learning pages and course organization, including hf.co/learn/agentcourse and hf.co/agentcourse.
• A hackathon connected with Gradio and smolagents was announced as a chance to apply the skills taught in the course. Further details were to be shared through Discord and email.
• The team also encouraged learners to like the livestream, star the course repository, and follow the course organization.

Speakers and Sources Featured

• Thomas Simonini — Hugging Face developer advocate; co-author of the course and contributor to the Deep Reinforcement Learning course.
• Joffrey — Hugging Face machine-learning engineer; co-author of the course.
• Ben — Hugging Face machine-learning engineer; co-author who focused on much of the LLM material, tests, and certificates.
• Audience members — Anonymous viewers who submitted questions through the livestream chat.
• Gradio and smolagents colleagues — Mentioned as the source of the hackathon announcement; no individual speakers from those teams are identified in the subtitles.