davidkimai/context-engineering

Context Engineering is a learning resource, not a runnable software tool. It is a handbook and in-progress course for people who work with AI language models and want to go deeper than writing better prompts. The central idea, quoted from AI researcher Andrej Karpathy, is that "context engineering is the delicate art and science of filling the context window with just the right information for the next step." The difference between prompt engineering and context engineering, as framed here, is scope. A prompt is what you type to the model. Context is everything the model sees: your instructions, example conversations, retrieved documents, memory from earlier sessions, tool outputs, and the flow between steps in a multi-step task. Context engineering is the practice of designing and managing all of that deliberately, not just writing a better sentence. The repository is organized as a progressive course with four stages. The first stage covers the basic building blocks: single prompts, few-shot examples (showing the model a handful of examples before asking it to do something), memory, and multi-step flows. Later stages get into more technical territory, including how to set up systems that retrieve relevant documents before querying a model, how to coordinate multiple AI agents working in parallel, and what the research literature says about how context affects model performance. The course structure uses a biological metaphor, moving from atoms to cells to neural systems, to show how complexity builds incrementally. The content draws from a review of over 1,400 research papers on the topic, and the README links to several academic papers from groups at IBM, Princeton, and other institutions. A 12-week course syllabus is under construction. This resource is aimed at developers and technically curious people who already use AI tools regularly and want a more principled framework for getting better results, particularly when building multi-step or multi-agent AI workflows.