Context Engineering is the practice of dynamically assembling and managing all the information an LLM needs before each model call so the agent can behave intelligently, maintain state, and provide personalized responses. Unlike traditional prompt engineering, which focuses mainly on static instructions, context engineering manages the entire context payload sent to the model.
Core Idea
Large Language Models (LLMs) are inherently stateless. They only know what is provided within the context window of a given request. Context engineering solves this limitation by supplying the right information at the right time.Goal
Provide the model with:- No more information than necessary
- No less information than necessary
- The most relevant information for the current task