Stop treating AI agents like magic. Learn how they actually work - by building one yourself, from scratch. Every week brings a new "AI agent" framework promising autonomous reasoning and tool use. But if you've never built one without the frameworks, you're trusting a black box you don't actually understand - and when it breaks in production, you won't know why.
AGENTIC AI takes a different approach. Across 14 hands-on chapters, you'll build Scout - a real, working autonomous agent - using nothing but raw Python and direct API calls. No LangChain. No hidden abstractions. Just you, the code, and a genuine understanding of what's actually happening when an AI "reasons," "decides," and "acts."
By the time you reach Book Two in this series (where the frameworks finally show up), you won't be trusting them blindly - you'll know exactly what they're automating, because you built it by hand first.
What you'll actually build:
- A working conversation loop - and the exact reason a model has no memory unless you give it one
- Step-by-step reasoning (chain-of-thought, self-consistency, tree-of-thought) built and compared side by side
- The ReAct pattern, then real tool use - a calculator, a clock, a unit converter - wired in by hand
- Multi-step planning that breaks a goal into an executable sequence
- Persistent memory: session history, automatic summarization, and a durable user profile
- Honest uncertainty handling - an agent that says "I don't know" instead of guessing
- Multi-agent collaboration, with a genuine debate between two independent perspectives
- Real safety guardrails - risk tiers, human confirmation, and a full action audit log
- A rigorous evaluation harness, the same discipline professional AI teams actually use
Two complete capstone projects close out the book: a fully guided build (StudyBuddy, a personal learning companion) and a lighter-scaffolded challenge (TaskFlow, a project-planning assistant) where you apply everything on your own before checking the reference solution.
Who this book is for:
Developers who know Python and want to genuinely understand agentic AI - not just call an API and hope. No machine learning background required. No prior agent-building experience assumed. Just solid fundamentals, explained clearly, built one working line of code at a time.
Written by Dave Malhotra, whose approach favors real, running code over slideware - every chapter ends with a working program, a multiple-choice check for understanding, and a clear path to the next concept.
A free companion package - full source code for every project in the book - is available for download, no registration required.
This is Book One in the Agentic AI Engineering series. Book Two takes everything you learn here and rebuilds it with the professional toolchain (LangChain, MCP, LangGraph); Book Three goes further still, into multi-agent systems, retrieval-augmented generation, and long-term memory.
Stop wondering how AI agents work. Build one, and know for certain.
