Context engineering is the discipline of deciding what information to include in a model's prompt or session: user identity, prior interactions, relevant documents, system state, and instructions. An AI model is only as useful as the context it receives.
A customer service agent handling a billing dispute that receives the customer's account history, the specific transaction in question, the relevant billing policy, and the customer's prior communications is working from well-engineered context. Most enterprise AI failures are not model failures — they are context failures. The model had the capability to help; it just did not have the right information to do so.
