Context engineering

Context includes instructions, relevant documents, previous steps, available tools and examples. Context engineering decides which of these goes into the limited context window, and when.

This is crucial especially for AI agents working over many steps: too little context leads to errors, too much to distraction, cost and delays.