Skip to main content

About the Handbook

What is the AI Agents Handbook​

AI Agents Handbook is a self-study guide for developing AI Agents,

  • covering all topics of Applied AI development
  • flexible: you can move quickly, studying only the core material, or delve into additional materials
  • practice-first
  • zero to hero: you don't need any prior knowledge to start
  • the last chapters are dedicated to the cutting edge of science and business.

Our goal is to create the world's first end-to-end course on AI Agents development. We are inspired by the legendary resource d2l.ai.

Quick links:

Approximate total volume of core material: 126 hours.

Values​

  1. Agile and up-to-date - we add materials as quickly as possible, adapting to constantly improving technologies
  2. Practice-First - practical assignments in each block
  3. Community-Driven - students and authors help develop the handbook - how?
  4. Zero to Hero - the handbook is planned to be a high-level abstraction that can be studied without in-depth CS and DS knowledge
  5. No empty value - we create materials ourselves only if we add value to what already exists in the world

Principles​

  1. Text format is used to allow for instant changes, as the AI Agents landscape becomes 50% obsolete every 12 months
  2. Modules consist of materials from different teachers, in order not to engage in the "rewriting of other people's materials" that is classic for university courses
    • it also gives the opportunity to use lectures from the best teachers in the world
    • and you as a student can at any time deviate from the program and delve into topics that interest you
  3. If you find the current material useless or outdated - let us know!

Program​

The main course material is presented in three blocks: basics, junior, senior. You can register on roadmap.sh and interactively mark completed modules: AI Agents 2025 Roadmap.

The entire program is divided into blocks, each block consists of modules, inside the module one topic is covered (For example, "Basics of LLM").

1. Basics (~6 hours)​

In the basics block, you will write your first agent, gain basic knowledge of LLM, how to work with them, and learn how to write prompts - this is all you need to start learning and developing Agents.

In additional modules, you can study lightweight backend development, LLM use cases, and master vibe-coding.

2. Junior (~40 hours)​

  • The junior block starts with completing the huggingface course on agents.
  • Next, you will learn about the Retrieval Augmented Generation (RAG) architecture, which is important for business.
  • You will go through a workshop on agent and multi-agent architectures
  • You will start mastering one of the industrial frameworks for agent development: LangGraph, Pydantic AI or LlamaIndex
  • And you will implement a large project of your own!
  • The block concludes with modules on AgentOps: infrastructure, agent performance evaluation, accelerated development.

After completing this block, it is recommended to start looking for your first job at a middle-level position. You can continue your education in parallel with your job search.

3. [TBD] Senior (~60 hours)​

Mastering agents at a high level is quite easy - therefore, immediately after the junior block comes the senior block. Here we will return to the topics already studied and delve into them: LLM, prompting, backend, RAG, cognitive architectures, benchmarking. New modules for us will be online agent evaluation, feedback-based improvement, meta-agents, Agent2Agent protocols. At the end, you will implement a vertical industrial-grade project with AgentOps infrastructure, memory, and much more.

4. [TBD] TechLead (100 hours +)​

Agent developers cannot be characterized as people who completed certain material a long time ago and have since stopped learning. On the contrary - Agents are in the fastest growing part of the industry. Constant learning, FOMO, the search for new technologies - all this is the romance of an AI Agents developer.

This size is planned for already practicing developers to share experience. It will mainly consist of:

  • materials on developing agents on modified models
  • review of scientific articles
  • insights from industry leaders
  • evaluation of the best changes in the global landscape
  • other frontier topics

5. Not AI Agents (~20 hours)​

A module with all the additional information that is not directly related to LLM:

  • backend development
  • DevOps
  • AI security
  • ethical aspects
  • vibe-coding
  • meta-learning
  • and other topics

Modules from this block are located along the main material.

tip

All these modules from "Not AI Agents" are not required to complete. However, if you are a beginner developer - these modules will help you in developing more production-ready agents and finding a job.

If you are already an experienced developer, then you will probably only appreciate the materials on meta-learning, vibe-coding.

What this handbook does not contain​

  • Training on no-code platforms (but if you want to lead - we will support you)
  • Training in programming, the basics of python, git, working with a computer
  • Training in Data Science, ML, DL, CV, NLP, applied mathematics
  • Methodological materials on writing resumes, job search, passing interviews; support in the learning process, a community of motivated students (but all this is on merkulov.courses)