FRONT COVER MATTER

AFTERWORD SKILLS COVERAGE & LEARNING OBJECTIVES This section maps each Part of the book to the specific skills and learning objectives it builds, so you can see at a glance what you will be able to do by the time you finish each stage of the book, and use it afterward as a quick reference for where a particular skill lives. Part One - Foundations of Artificial Intelligence (AI) and Language Models: explain what AI and large language models are and are not, describe how a model turns text into a response without resorting to mathematics, trace the path from a raw model to a working application, and distinguish the three professional disciplines that surround the model. Part Two - Introduction to Harness Engineering as a Discipline: define what a harness is and why it is necessary, explain why harness engineering emerged as its own field, apply the seven-layer ETCLOVG taxonomy to any agent system, and map the components of a production harness onto that taxonomy. Part Three - The Execution Layer (Running the Agent Loop): implement the core agent loop, manage state and control flow across a run, choose and apply control structures for decision routing, and select an appropriate runtime execution environment for a given harness. Part Four - The Tooling Layer (Giving the Agent Capabilities): define what a tool is from the harness's point of view, design and register tools in a tool registry, apply design principles that make tools easy and safe for a model to use, work with the Model Context Protocol (MCP) and other tool standards, and recognize the common categories of tools an agent relies on. Part Five - The Context Layer (Managing the Agent's Working Memory): treat context as the scarce resource it is, recognize and prevent context rot and context bloat, construct context deliberately with a context builder, apply compaction and summarization techniques, and design a memory hierarchy appropriate to a given task. Part Six - The Lifecycle Layer (Birth, Life, and Death of an Agent Run): start a run correctly, use lifecycle hooks to intervene at the right moments, enforce budgets and resource limits, and end a run cleanly and safely. Part Seven - The Observability Layer (Seeing Inside the Agent): explain why observability matters for agent systems, instrument a harness with tracing and spans, build logging, metrics, and dashboards that surface real problems, and use replay to debug a run after the fact. Part Eight - The Verification Layer (Evaluation, Testing, and Quality): build an evaluation harness, design evaluations that measure what actually matters, choose appropriate scoring methods, interpret benchmarks and leaderboards critically, and test the harness itself as rigorously as the agent it runs. Part Nine - The Governance Layer (Safety, Security, and Compliance): design a safety architecture for an agent system, apply permissions and access control correctly, recognize and defend against prompt injection and tool poisoning, secure the Model Context Protocol (MCP) and its supply chain, apply sandboxing and isolation techniques, and account for responsible AI, compliance, and risk. Part Ten - Multi-Agent Systems and Orchestration: decide when more than one agent is actually warranted, choose an orchestration pattern suited to a task, implement the coordination mechanics that make multi-agent systems reliable, and design communication protocols between agents. Part Eleven - Memory, Knowledge, and Retrieval: distinguish the types of agent memory and when to use each, implement retrieval-augmented generation (RAG), and manage knowledge for agents over time. Part Twelve - Frameworks, Standards, and the Tooling Ecosystem: survey the landscape of agent frameworks, apply relevant standards and conventions, evaluate platforms and managed services, and navigate the open-source landscape with informed judgment. Part Thirteen - Performance, Cost, and Reliability Engineering: reason about the economics of agents, apply cost optimization techniques, apply latency optimization techniques, and design for reliability and error recovery. Part Fourteen - Deployment, Operations, and Agent Operations (AgentOps): take a harness from prototype to production, choose a deployment architecture suited to a system's needs, run agent operations (AgentOps) day to day, and respond effectively to incidents involving agents. Part Fifteen - Specialized and Applied Harnesses: build coding agent harnesses, build computer-use and browser agents, build research, writing, and knowledge agents, and adapt harness design to other specialized domains. Part Sixteen - Building a Harness from Scratch (Capstone Projects): build a minimal harness from nothing, extend it with context management, extend it again with tools, the Model Context Protocol (MCP), and safety, extend it further with evaluation and observability, and extend it a final time into a multi-agent, production-hardened system - five escalating, hands-on projects that put every earlier Part into practice. Part Seventeen - The Profession, Ethics, and the Certification Path: describe the role of the AI harness engineer, apply professional practice and collaboration norms, reason through the ethics and societal impact of this work, and prepare systematically for certification. Appendices A through H provide durable reference material for use after the book is finished: a glossary of terms and acronyms, a component reference, a patterns catalog, safety and security checklists, evaluation templates, further reading and resources, a lab setup guide, and a full resume-portfolio-building capstone project suite. PART ONE: FOUNDATIONS OF ARTIFICIAL INTELLIGENCE (AI) AND LANGUAGE MODELS Part One gives you the ground you need to stand on. You will learn what AI is and is not, how a language model works without any mathematics, how a model becomes a working application, and how three related professional skills fit together around the model. By the end of Part One you will be ready to understand why a harness is needed at all.