Context Engineering
Context Engineering is a methodology in artificial intelligence that involves systematically designing, constructing, and managing the contextual information supplied to AI models in order to guide them toward more accurate and relevant outputs. It moves beyond single-prompt approaches by integrating background knowledge, task instructions, constraints, examples, and other information types to optimize a model's understanding and execution of tasks. It is applicable to dialogue systems, autonomous agents, and complex task processing.
Core Facts
As the capabilities of underlying models tend toward homogenization, differentiated competition at the AI application layer will occur at the level of workflow orchestration and domain knowledge accumulation.
Zhai Lujia emphasized that the solutions provided by AI are not necessarily optimal, that AI is inherently designed to go along with what users say, and that the value of experienced developers lies in judging the reasonableness of AI suggestions.
The core techniques of the Context Engineering phase include RAG (Retrieval-Augmented Generation), memory management, and dynamic document injection
Timeline (last 90 days)
There are generally two approaches to implementing persistent memory: writing information as structured text to an external file and reading it back each round, or using a vector database for semantic retrieval.
Context Engineering focuses on the flow of information across multiple rounds of conversation, unlike prompt engineering, which focuses on the wording of a single input.
A production-grade context system must simultaneously satisfy four conditions: navigable, fresh, fast, and compoundable.
Context engineering is the construction of the entire runtime environment surrounding the model, determining what state the agent can see at each step.
The three common context failure modes for agents are Dead End, Stale State, and Latency.
Context is divided into six types: Instructions, Knowledge, Memory, Examples, Tools, and Guardrails, which can be categorized into Static Context and Dynamic Context.
RAG, long-context filling, and few-shot example construction are all essentially specific techniques of context engineering.
The core insight of context engineering is that model output quality is strongly correlated with the content of the input context window, including code snippets, file structure, conversation history, tool call results, and externally retrieved documents.
An agent's self-correction capability within a runtime environment is constrained by the reasoning quality of the underlying large model.
Context Engineering focuses on how to precisely manage the context window fed into the model, Harness Engineering leans toward the infrastructure layer, and both appeared around the same time as Loop Engineering.
12 more timeline events
All Facts (20)
As the capabilities of underlying models tend toward homogenization, differentiated competition at the AI application layer will occur at the level of workflow orchestration and domain knowledge accumulation.
85%VerifiedZhai Lujia emphasized that the solutions provided by AI are not necessarily optimal, that AI is inherently designed to go along with what users say, and that the value of experienced developers lies in judging the reasonableness of AI suggestions.
80%VerifiedThe core techniques of the Context Engineering phase include RAG (Retrieval-Augmented Generation), memory management, and dynamic document injection
80%VerifiedContext Engineering is a concept that rapidly gained popularity in the AI development community in the first half of 2025, promoted by Shopify CEO Tobi Lütke and others in public discussions.
80%Verified上下文工程（Context Engineering）是2024年后AI开发领域兴起的重要实践，由Andrej Karpathy等人推广
80%VerifiedContext Engineering is a new paradigm that provides AI with a complete, systematic context, including project documentation, code examples, coding rules, development patterns, and validation mechanisms.
80%VerifiedThe concept of Context Engineering was popularized by former OpenAI researcher Andrej Karpathy and others.
70%VerifiedContext Engineering and Prompt Engineering are fundamentally different: Prompt Engineering focuses on optimizing single prompts, while Context Engineering focuses on the dynamic management of context throughout the entire conversation lifecycle.
70%VerifiedAn agent's self-correction capability within a runtime environment is constrained by the reasoning quality of the underlying large model.
65%UnverifiedgeneratePRP 命令会自动执行六个步骤：读取需求、研究阶段、文档收集、创建蓝图、质量检查和生成 PRP
90%UnverifiedContext Engineering围绕三大核心支柱：上下文设计（Context Design）、上下文编排（Context Orchestration）和上下文优化（Context Optimization）
85%UnverifiedRAG（Retrieval-Augmented Generation，检索增强生成）技术的成熟推动了Context Engineering的兴起
85%Unverified上下文工程（Context Engineering）的核心思想是做特定任务时只提供必要的信息
85%UnverifiedAndrej Karpathy等人将Context Engineering类比为传统软件工程中的内存管理
85%UnverifiedexecutePRP 命令执行开发时，案例项目被拆分为 10 个子任务
85%Unverified上下文工程的核心认知是LLM的输出质量完全取决于输入上下文的质量
80%UnverifiedPrompt Engineering解决'怎么说'，Context Engineering解决'给它看什么'，MCP解决'怎么连接'，Skill解决'怎么按需加载技能'，Harness Engineering解决'怎么把这些组成闭环'
75%Unverified上下文工程（Context Engineering）正在成为AI时代最重要的开发技能之一
70%UnverifiedContext engineering is replacing the narrowly defined prompt engineering of the past and is becoming the concept emphasized by the AI Agent development community.
60%UnverifiedThe three engineering paradigms have an inclusion relationship: prompt engineering ⊂ context engineering ⊂ Harness engineering.
60%