Context engineering is the practice of deciding what information an AI should see at each point in a task. It includes the system prompt, the user's request, selected files, retrieved records, examples, tool descriptions, conversation history, and the results of earlier actions. The goal is not to fill the context window; it is to supply the smallest useful set of high-signal information.
Prompt engineering focuses mainly on how instructions are written. Context engineering is broader and continuous. An agent may retrieve a policy only when a question touches compliance, load an agent skill only when its workflow applies, summarize an old conversation before it grows unwieldy, or discard noisy tool output after extracting the important result.
More context is not automatically better. Irrelevant material competes for the model's attention, raises cost, and can bury the instruction that matters. A large context window increases capacity, but someone still has to decide what belongs inside it and what should remain available for retrieval later.
Why it matters at your desk. A researcher using Claude Projects should separate core source documents from old drafts and duplicate notes. A team using Amazon Quick needs permissions and retrieval rules that surface the right workplace material without exposing unrelated files; its move toward persistent work context makes those choices more consequential. Engineers working in Cursor get better results when repository rules, the relevant code, and test feedback arrive at the right moment.
A useful test is simple: if the AI misses something, ask whether it lacked the information, could not find it among the noise, or received it too late. Each failure calls for a different context fix.