Agentic AI Engineering by Paul Iusztin

Agentic AI Engineering by Paul Iusztin is a self-paced course for engineers building AI systems beyond initial prototypes. Its 34 lessons move from agent foundations through two complete projects, production operations and an independent capstone.

The course addresses engineering challenges around the model, including context management, evaluation and observability. Its materials comprise 2.29 GB of MP4 video and open-source project code.

Skills you’ll build in Agentic AI Engineering

• Build production-ready AI agents using LangGraph and LangChain
• Implement ReAct loops, tool calling, and context engineering
• Complete two end-to-end projects: Nova research agent and Brown writing pipeline
• Apply LLMOps: evaluation, observability, and CI/CD deployment on GCP
• Design agentic systems using FastMCP and Gemini on Google Cloud Platform
• Build and present an independent capstone agentic AI project

Agentic AI Engineering course breakdown

The curriculum has four parts and a capstone. It begins with the components of an agent, applies them in two projects, then turns to the operational practices needed to evaluate, deploy and scale systems.

Part 1: Agent foundations

The foundations section covers tool calling and ReAct loops, alongside context engineering and structured generation. It also introduces memory systems and retrieval-augmented generation (RAG) as components of agentic applications.

Part 2: Nova research agent

The first complete project is Nova, an autonomous deep research agent built from scratch. The project follows the agent end to end rather than stopping at an isolated demonstration.

Part 3: Brown writing pipeline

The second production project is Brown, an agentic workflow pipeline for writing. Paul built both Brown and Nova while developing the curriculum and uses them as the course’s projects.

Part 4: LLMOps

The LLMOps section covers evaluation and observability, as well as authentication, CI/CD on Google Cloud Platform and scaling. These topics address how to assess and operate an agentic system beyond the initial build.

Independent agent capstone

For the capstone, learners design, build and present an independent agentic AI project using the methodology introduced throughout the course.

Who the course is for

The course is intended for developers moving beyond LLM wrappers, ML engineers working to make prototypes hold up in production, and AI practitioners seeking a structured approach to LLMOps. Technical founders can also use it to understand the agent design decisions involved in leading AI projects.

About Paul Iusztin

Paul Iusztin is a senior AI engineer with more than 10 years of experience and 20 shipped production AI applications. He founded the Decoding AI newsletter, which has more than 40,000 subscribers, and authored the LLM Engineer’s Handbook.

Frequently asked questions

What does Agentic AI Engineering cover?

The course covers agent foundations, two end-to-end projects, LLMOps and an independent capstone. Topics include tool calling, memory, RAG, evaluation and deployment on Google Cloud Platform.

Who is the course designed for?

It is aimed at developers, ML engineers, AI practitioners and technical founders who want to work with agent architecture and production systems. It is not presented as an entry point for learning the basics.

What tools are used in the course?

The curriculum covers LangGraph, LangChain, Gemini, FastMCP and Google Cloud Platform.

What materials are included?

The course contains 2.29 GB of self-paced MP4 video and uses open-source code for its projects.

Who is Paul Iusztin?

Paul Iusztin is a senior AI engineer, the founder of Decoding AI and the author of the LLM Engineer’s Handbook. He has more than 10 years of experience and has shipped 20 production AI applications.