Context Engineering
Context engineering is the systematic management of all the information made available to an AI model or AI agent: system instructions, retrieval results (RAG), tool outputs, short-term and long-term memory, and interaction history. Unlike classic prompt engineering, which aims at the wording of individual requests, it concerns the architecture of the entire information environment across several steps and sessions. Prompt engineering is therefore a subset of context engineering. The term came into use alongside agentic AI and RAG systems, because unstructured or overloaded context can impair the reliability of AI agents in production; the shorthand "context rot" has become established for this in professional discussion.
In practice
For companies running AI chatbots, search assistants or agents, for customer service or internal knowledge bases for example, good context engineering decides answer quality and the risk of hallucination. In practice that means structuring knowledge sources cleanly and dividing them into retrievable units (chunks), so that retrieval systems return precisely the relevant passages; filtering tool output down to what matters instead of tipping raw data into the context; separating the application's short-term and long-term memory sensibly, such as session history against customer history; and actively steering the context window budget to balance cost against answer quality. For websites intended to serve as a knowledge source for RAG systems or third-party AI agents, one thing more applies: clear structure, structured data and well-defined content sections make machine retrieval easier, which is a point of contact with AEO and GEO work. Agencies should point out to clients that a single well-worded prompt guarantees no reliable agent workflows on its own – the system architecture behind it has to be planned as well.