From prompt engineering to context engineering

If you've learned to write good prompts, you already have a head start on the skill that's come to define serious AI work in 2025–2026: context engineering. This lesson defines it and shows how it extends what you already know.

Prompt vs. context — the clean distinction. A prompt is the message you write. Context is the entire payload the model actually receives when it generates a response — which includes your prompt, but also much more: the system instructions, any examples, retrieved documents, the definitions of tools the model can call and the results those tools return, the conversation history so far, and the user's actual request. Prompt engineering is about writing and organizing your instructions; context engineering is about curating and managing the whole set of information the model sees. One influential description (from Andrej Karpathy) called it "the delicate art and science of filling the context window with just the right information for the next step."

Where the term came from. The practice — assembling instructions, examples, retrieved data, and history for a model — predates the label. But the name crystallized in mid-2025: Shopify's CEO Tobi Lütke popularized "context engineering" over "prompt engineering" (June 2025), Karpathy amplified it, and developer Simon Willison documented and endorsed it, predicting the term would stick. Later in 2025, Anthropic published a rigorous engineering treatment of it. So this isn't hype-driven jargon — it's a name practitioners adopted for something they were already doing, now taken seriously.

Why it's the "natural progression" of prompting. Anthropic frames context engineering as the natural next step after prompt engineering — and importantly, as the superset that includes it, not a replacement. Writing a good system prompt is still part of the job; context engineering just widens the lens to everything else that lands in the model's view. So ignore the "prompt engineering is dead" takes — the honest framing is that prompting is one ingredient inside the larger discipline of context engineering.

Why the shift matters. As you build anything beyond a single question — working over documents, using tools, holding a conversation, running an agent — what you put in front of the model becomes the dominant factor in quality, more than the exact wording of any one instruction. The field's attention has moved accordingly: less "find the perfect phrase," more "what configuration of context is most likely to produce the behavior I want?" That question is what this course teaches you to answer.

The mindset: context engineering is the discipline of deliberately deciding everything a model sees — system prompt, examples, retrieved documents, tool definitions and results, conversation history, and the user query — in what order, how much, and what to leave out. A prompt is the message you write; context is the whole payload the model receives. The term was popularized in mid-2025 (Lütke, Karpathy, Willison) and treated rigorously by Anthropic, which frames it as the natural progression of prompt engineering — a superset that includes prompting, not a replacement. As soon as you build beyond a single question, what you put in front of the model matters more than any single phrase, and engineering that context well is the skill this course builds.