What is context engineering?

Context engineering is choosing what information an AI model is given with each request. A model only works from what it is sent at that moment. So an AI product puts together a bundle every time: its instructions, relevant documents, search results, saved notes about the user, the conversation so far and the new question. Context engineering decides what goes in, in what order, and what to leave out when there is too much.

Why it matters for a business

Two products on the same model differ mostly here. Which documents the search returns, how much history stays in, what a summary of an old conversation keeps: those choices decide whether the model has the right material in front of it, and a better model does not rescue a window filled with the wrong material. It is also where cost is controlled, because every token in the window is billed on every request.

Example

Let’s say a support assistant summarises the first 20 messages of a long thread to make room. The summary keeps that a credit was issued and drops that the customer refused the credit and asked for the money back. On the next turn, the assistant offers a credit again.

Go deeper

AI wrappers: what gets built around the model

Related terms

• Context window
• Context rot
• Prompt engineering
• System prompt
• RAG (retrieval-augmented generation)

All terms, A to Z