دانلود کتاب ساخت عامل‌های هوش مصنوعی تولیدی برای طراحی وب و هماهنگ‌سازی عامل‌های خودمختار با MCP، الگوهای چندعاملی و مهار

Large language models (LLMs) have made it relatively straightforward to add conversational AI to an application. A model can summarize text, answer questions, generate content, and help users complete familiar tasks. However, a convincing demonstration is not the same as a dependable application. Production systems must give models the right context, connect them to trustworthy data and tools, control the actions they may take, and provide evidence of how they reached an outcome. Agentic applications extend an LLM beyond a single prompt and response. They can pursue goals, retrieve information, use tools, maintain state, plan work, collaborate with other agents, and act within an application environment. Each additional capability also creates engineering responsibilities. Developers must decide what belongs in the model, what must remain indeterministic application code, how autonomy should be bounded, and how the resulting behavior can be evaluated and improved. This book presents a practical path from the first LLM interaction to a production-ready agentic system. You will begin with the foundations of agentic applications and learn how prompts, context, model settings, and structured outputs shape an LLM-powered feature. You will then connect models to external capabilities through tool calling, retrieval-augmentedgeneration (RAG), GraphRAG, and the Model Context Protocol (MCP). From these building blocks, you will develop AI assistants and agents using the ReAct pattern, explicit agent architectures, planning systems, orchestration, and bounded autonomy. The book then expands from individual agents to multi-agent collaboration, communication, and reusable architectural patterns. You will also examine anti-patterns that can make agent systems fragile, opaque, or unnecessarily complex. The later chapters focus on the controls required beyond a working prototype. You will evaluate LLM responses and multi-step agent behavior, build an agent harness that enforces limits and records execution, and prepare agents for production with trusted identity, authorization, approval boundaries, durable state, safe recovery, and controlled side effects. Finally, you will use logs, metrics, traces, regression scenarios, and canary releases to operate and improve agents after deployment.