دانلود کتاب مهندسی بافت (Context Engineering)

Artificial intelligence (AI) is driving one of the fastest technology adoption cycles in recent history. The most visible part of this wave is generative AI (GenAI), a class of AI systems designed to produce novel content from natural-language input, including text, code, images, music, and video. At the core of text-based GenAI are large language models (LLMs), deep learning models capable of interpreting and generating human-like language. In practice, the effectiveness of an LLM-based system depends on its context, meaning the information available to the model at inference time. Context engineering is the discipline of selecting, organizing, and updating that information so the model receives the right inputs in the right form, at the right time, and within the limits of its context window. This chapter introduces the foundations of context engineering and how this book is organized around them. It then reviews the evolution of AI systems, from simple interactions with LLMs to workflows, agents, and agentic systems. Next, it examines the limitations of LLMs that make context engineering essential for building modern AI systems. Finally, it provides an overview of the sources that shape context in LLM-based systems: instructions, external knowledge, tools, memory, state, and user prompts. Prompt engineering is the practice of designing and refining the input for an LLM (called prompt) to guide its outputs toward more accurate, useful, or creative results. It gained rapid attention in early 2023 after the public release of ChatGPT and other widely accessible models, when millions of users discovered that response quality depended strongly on wording, structure, or punctuation. This discovery revealed both the power of these models and the unpredictability of a process that often relied on trial and error. However, prompt engineering addresses only part of the problem. Over time, practitioners recognized that LLM behavior is shaped not only by prompt wording but also by the broader context in which the model operates. This broader view gave rise to context engineering, the discipline of equipping AI systems with the information needed to complete a task effectively. In practice, this means deciding what to place in the context window, the limited amount of text the model can process at once, usually measured in tokens, the internal text units used by the LLM. Consider a simple example. In a prompt-engineered approach, the user writes the prompt “Summarize the attached incident report and identify the root cause.” In this case, the LLM depends mainly on the prompt wording and its prior training, which can lead to omissions or misinterpretations. In a context-engineered approach, the application retrieves the report, selects the most relevant sections, and injects them into the model call. The prompt may remain the same, but the model receives a curated context that makes the task more reliably solvable.