Building AI Agents from Scratch with Python

▎ AI agents are the next frontier of software development. We're no longer talking about chatbots that answer questions — we're talking about autonomous systems that reason, plan, use tools, and execute complex tasks end to end.

▎ Building AI Agents from Scratch with Python is a hands-on guide that takes you from simple LLM calls to production multi-agent systems. Across 15 chapters, you build Aria — an AI agent that starts as a basic loop calling the Anthropic API

and evolves into a complete system with persistent memory, planning, RAG, multi-agent orchestration, security guardrails, and deployment with Docker and FastAPI.

▎ Each chapter adds a real layer to the project: tool calling with a decorator-based registry, short and long-term memory, planning with ReAct and replanning, state machines with checkpoints, Retrieval Augmented Generation with a custom vector

store, testing with LLM-as-Judge, prompt injection defense, and observability with metrics and alerts.

▎ By the end, you extract everything into a reusable SDK and deploy Aria as a REST API with streaming, session management, and monitoring — ready for production.

▎ This is not a theoretical book. Every concept comes with complete, runnable Python code, practical exercises at three difficulty levels, and a common mistakes section that will save you weeks of debugging.

▎ Who this book is for:

▎ - Intermediate Python developers who want to build real AI agents

▎ - Engineers already using LLMs who want to go beyond simple wrappers

▎ - Teams that need to ship agents to production with security and observability