Building Safe AI Agents with Guardrails

Hosted by Alexey Grigorev

215 students

In this video

What you'll learn

Working agent from scratch

Build a minimal agent that accepts input, calls tools, and returns results.

What guardrails actually are

Understand what guardrails are, where they sit, and why prompts alone fail.

Production-ready guardrail patterns

Implement practical input, tool, and output guardrails used in real systems.

Reusable mental model

Leave with a guardrail model that applies across agents and frameworks.

Why this topic matters

As teams move from chatbots to agents, failures become actions. Without guardrails, agents can misuse tools, ignore constraints, and fail silently. This lesson gives engineers a clear mental model and concrete patterns for building safer agents that hold up beyond demos.

You'll learn from

Alexey Grigorev

Principal Data Scientist | Book Author | Instructor to 100k+ Students World Wide

Founder of DataTalks.Club · Creator of the Zoomcamp Series · ML Engineer & Author


Alexey Grigorev is the founder of DataTalks.Club and the creator of the popular Zoomcamp series. With 15 years of experience in software engineering and over 12 years in machine learning, he has built and deployed large-scale ML systems at companies like OLX Group and Simplaex.


An advocate for practical, hands-on education, Alexey has taught over 100,000 students, focusing on a code-first approach to help learners build real-world skills.


In the past, he was an active participant in data science competitions. A Kaggle Master, Alexey has achieved top rankings in several challenges, including 1st place in the NIPS'17 Criteo Challenge and 2nd place in the WSDM Cup 2017: Vandalism Detection.


He is also the author of several technical books, including the widely read Machine Learning Bookcamp.

Go deeper with a course

AI Bootcamp: From RAG to Agents
Alexey Grigorev
View syllabus