Large language models

Bestselling book on building LLMs. A practical, project-driven manual for engineers who want to understand how modern language models are built — and where they fail — by writing every layer themselves. From a scalar autograd engine to RLHF to fused specialists, in 36 hands-on projects with deliberate sabotage experiments. Build it. Break it. Measure it.

Legacy modernization is an evidence problem before it is a code-generation problem. Use Claude Code to recover hidden behavior, build a verification harness, and move critical systems in controlled, reversible slices without handing production decisions to AI. Includes 26 practical chapters, 86 diagrams, and a runnable companion lab.

Software development is changing fast, and Claude Code is at the center of that shift. Learn how to work effectively with AI agents to write code, automate workflows, and build larger projects with confidence. From setup and prompt design to real-world engineering practices, this book provides a practical guide to modern software development in 2026.

Learn how to build, run, and optimize llama.cpp from the ground up. This book covers everything from compiling the code and working with GGUF models to deploying fast, production-ready local LLM inference.

It's never been easier to build an AI agent — and never been harder to make one that actually works. This book takes you from language model foundations to production-ready multi-agent systems with the depth to predict failure before it happens, engineer graceful degradation over catastrophic failure, and take absolute architectural ownership. Get the paperback from amazon.

Your C# skills are worth more today than they were a year ago — if you know how to put a language model in the loop. This book shows you how, with the Microsoft Agent Framework: real tools, RAG, multi-agent orchestration, plus the hosting, observability, and safety that separate a demo from a system you ship. Nineteen chapters. 120 runnable projects. No Python detours. Just C# and .NET 10.

Code Is the Side Effect"Software engineers are not primarily code writers. We are clarity traders — and that hasn't changed."You've seen the demos. The AI builds a whole feature from a sentence. The agent writes tests, fixes the failing ones, opens the PR. It's remarkable.Then you come back three months later. The codebase is a tangle. Nobody knows why anything is the way it is. The agent that built it has no memory of what it decided or why. And every time you ask it to add something new, it breaks two things you didn't know were connected.This is the pattern that nobody talks about. AI coding tools make the easy parts of engineering dramatically easier. They leave the hard parts untouched — and they create new hard parts that didn't exist before.Ways of Working is the book for engineers who want to work with AI agents rather than be gradually replaced by them — who understand that the tools are genuinely powerful and genuinely limited, and want to build practices that get the most from each.What you will actually learnThe world model framework. Before an agent can build anything well, it needs to understand what it's building and why. This book teaches you to give agents what they need: a structured, queryable representation of your architecture, your component contracts, your behavior specifications, and your code patterns. No world model = no sustained agentic development.Intent documentation. The most expensive bug in agentic codebases is not a hallucination — it's a decision made without context. Why is this rule here? Why is this boundary where it is? Agents can't infer rationale from code. You have to write it down.Spec-Kit and formal specifications. GitHub's Spec-Kit brings machine-readable, traceable, CI-verified specifications to engineering teams. This book shows how to use it to turn requirements into agent inputs that are precise enough to generate correct implementations.Graph explainers. Tools like Graphify and Understand-Anything transform codebases and documents into queryable knowledge graphs — giving agents navigable context instead of flat text. This is the memory substrate that makes multi-agent systems reliable at scale.Agent architecture that holds. What makes an agent coherently itself? When do file-based agent systems break down and what replaces them? How does constraint-based coordination (borrowed from holocracy) solve the autonomy-coherence problem that has stumped AI researchers for decades?Claude Code, for real. A complete treatment of Claude Code's CLAUDE.md convention, permission model, hooks, and slash commands. Plus the oh-my-claudecode ecosystem: 15+ specialized agents, workflow orchestration patterns (autopilot, ralph, ultrawork), and the skills framework for team-specific automation.The AI-native organization. What genuine AI-native teams look like beneath the marketing. How to hire, structure, and lead them. What language-oriented programming and constrained natural language mean for the future of the human-code relationship.Who it's forEngineers who are past the "should I use AI?" question and into the "how do I use it without losing my engineering integrity?" question.Senior engineers. Engineering managers. Technical leaders. People who have noticed that the more they delegate to AI, the less certain they feel — and who want to understand why.From the AuthorI've been building production systems with AI agents for years. Not demos — systems that had to work reliably across months, maintain themselves as requirements changed, and produce outputs that engineers could understand and defend.That experience has made me skeptical in both directions.Skeptical of the "AI will do everything" vision — because I've watched too many AI-generated codebases collapse under the weight of accumulated misunderstanding.Equally skeptical of the "nothing fundamentally changed" position — because the engineers who treat AI coding tools as just faster autocomplete are making a category error they'll pay for in months of maintenance debt.Something genuinely new is happening. This book is my attempt to think about it clearly.

Learn how to build intelligent, production-ready AI agents with the OpenAI Agents SDK. Through practical Python examples and step by step guidance, you'll master everything from the fundamentals to advanced multi-agent workflows, tools, and deployment.

Learn Claude Code by building real projects. This hands-on companion turns the Claude Code Masterclass workshop into a practical self-paced guide for planning, coding, testing, reviewing, refactoring, and shipping software with AI.

Build smarter AI systems that go beyond the limits of large language models. Retrieval-Augmented Generation is a practical guide to designing, implementing, and scaling RAG applications with modern retrieval techniques, vector databases, and real-world deployment strategies.

Mastering Qdrant for RAG Applications is your practical guide to building production-ready RAG systems with the leading open-source vector database. Learn how to design, optimize, and scale high-performance vector search using Qdrant through clear explanations, real-world examples, and hands-on code.

The hardest part of building AI is not the model. It is everything around it. The Art of Harness Engineering is a practical guide to turning AI prototypes into reliable products. It covers testing, guardrails, observability and governance, giving you the tools to build AI systems people can trust and organizations can run with confidence.

Most vector search books start with the database. This one starts with the machine. Build a search engine from scratch, then push it from brute force to billion-scale retrieval with SIMD, HNSW, quantization and distributed systems. By the end, vector search won't be a black box. It'll be something you know how to build, tune and scale.

Somewhere there is a server with two GPUs for which no operational process exists. This book turns that into an operable platform — hands-on, in 23 labs: vLLM, KServe, LiteLLM, the NVIDIA GPU Operator, Keycloak, OpenBao, ArgoCD, pgvector. Not a tutorial: a reference work that shows the derivations behind every setting.

AIエージェントの構築が、これほど容易だった時代はない。そして、実際に機能するものを作ることが、これほど難しい時代もない。本書は言語モデルの基礎から本番対応マルチエージェントシステムまで、失敗が起こる前に予測し、壊滅的な障害ではなく優雅な劣化を設計し、完全なアーキテクチャの所有権を確立するための深さをもって、あなたを導く。ペーパーバック版はamazonにて好評発売中。