What Context Engineering Is, and Why It Replaced Prompt Engineering

You probably know about prompt engineering: the craft of wording a question or instruction so the model gives you a good answer. If you have used a chatbot and learned that "write me a polite, three-sentence reply declining this meeting" works better than "reply to this," you have done prompt engineering.

Context engineering is the next level up, and the shift in language is deliberate. The term was popularized by people like Tobi Lutke (CEO of Shopify) and Andrej Karpathy. Lutke described it as "the art of providing all the context for the task to be plausibly solvable by the LLM." Karpathy described it as "the delicate art and science of filling the context window with just the right information for the next step." Both definitions point at the same shift in attention: away from the single clever sentence, and toward the whole bundle of information the model sees before it responds.

The articles in this folder make a strong claim that is worth stating plainly: when an agent fails, it is usually not because the model is too weak. It is because the model was handed the wrong context. In the words of the first article, "Most agent failures are not model failures anymore, they are context failures."