Give the model the right context.
Distinguish state, memory, and context. Design retrieval that supplies attributable evidence rather than a larger prompt.
A BOOK FOR SOFTWARE ENGINEERS
Intelligence
needs engineering.
AGENTIC AI
Architecture, Theory, and Practice
Build systems that reason, plan, and act.
Understand the architecture that makes them reliable.
01 / THE FIRST EDITIONFROM INFERENCE TO ACTION
14 chapters115 technical figuresPython case studiesArchitecture before frameworks
01 / THE APPROACH
Beyond the demo.
Into the system.
A capable model is the starting point. Building an agent that operates reliably takes software engineering.
This book examines what surrounds the model: the tools it can use, the context it receives, the state it preserves, and the rules that govern its actions. It connects these choices to familiar ideas from distributed systems and software architecture.
For software engineers, architects, researchers, and graduate students who want to understand the design decisions behind agent systems.
Concrete mechanisms. Explicit trade-offs.
Examples you can reason about.
02 / WHAT YOU’LL LEARN
Make the important
decisions explicit.
CONTEXT & RAG
Turn proposals into controlled actions.
Design tool contracts, explicit plans, and agent harnesses. Preserve progress across sessions and verify when work is complete.
Know what happened. Measure what worked.
Trace observable execution, evaluate outcomes, and compare quality, latency, and cost across complete systems.
Constrain authority as capability grows.
Choose when to delegate, establish ownership and approvals, and enforce boundaries around tools, memory, and execution.
03 / INSIDE THE BOOK
One architecture.
Four perspectives.
Start with the agent loop. Follow it through context, tools, and coordination into the realities of production.
Four parts · Fourteen chapters
IFoundationsTHE MODEL AND ITS INPUT
- 01Foundations
- 02Prompting
- 03Retrieval and Retrieval-Augmented Generation (RAG)
IIBuilding BlocksCAPABILITIES AND COORDINATION
- 04Tools
- 05Observability and Evaluation
- 06Multi-Agent Systems
IIIRuntimeEXECUTION AND ADAPTATION
- 07Planning
- 08Agent Harnesses and Production Runtime
- 09Memory
- 10Learning
IVSystemsORGANIZATIONS AND OPERATIONS
- 11Agent Organizations
- 12Optimization
- 13Safety
- 14Future Directions
04 / THEORY, PUT TO WORK
Follow the decisions.
Understand the code.
Concise Python examples sit alongside the explanation. Follow an OAuth migration, a repository investigation, a dependency audit, and a governed delivery workflow.
The accompanying bbagents library turns the book’s concepts into typed, executable components for plans, policies, context, memory, and observability.
- 01ObserveCollect relevant evidence.
- 02DecidePropose the next action.
- 03AuthorizeCheck policy and approvals.
- 04ExecuteApply a bounded operation.
- 05VerifyInspect the actual outcome.
THE MODEL PROPOSES. THE RUNTIME ENFORCES.
BUILD WITH A CLEARER MENTAL MODEL
Your next agent
starts here.
Agentic AI
Architecture, Theory, and Practice
First editionPDF edition
Agentic AI
Fourteen chapters on the engineering
behind systems that reason and act.
$9.99 USD
Checkout opening soonPurchases will be processed through Stripe.
A FEW DETAILS
Before you begin.
Who is the book written for?
Experienced software engineers, architects, researchers, and graduate students. Familiarity with Python and software architecture will help you get the most from the examples.
Is it tied to a particular framework?
The book focuses on architectural mechanisms and trade-offs. It uses concrete tools and protocols, including LangGraph, MCP, and A2A, while keeping the core ideas independent of any one vendor.
Does it cover harnesses and agent skills?
Yes. Chapter 8 covers harness composition, reusable skills, work across sessions, and production execution. Context management, evaluation, and security are developed in their respective chapters.
How do I purchase a copy?
Checkout is not open yet. When available, the purchase button will take you to Stripe, where the price, format, and delivery details will be confirmed before payment.