What is Context Engineering?
Also called Context management
Definition
Context engineering is the practice of selecting, organizing, and maintaining the information available to a language model during generation. It includes instructions, retrieved evidence, conversation history, and tool outputs.
The operation of Context Engineering is explained.
Anthropic describes context as a finite resource. Context engineering goes beyond wording a prompt to deciding which information enters the model’s working context and how it is retained or compressed during a task.
The relevance of Context Engineering is explained.
A SaaS documentation assistant may need the current product edition and permission model alongside a retrieved setup guide.
Including that context can prevent an answer that is plausible but inappropriate for the customer’s configuration.
The evaluation of Context Engineering is explained.
Representative tasks support evaluation of evidence selection, ordering, and compression. Summaries retain source qualifications and product versions. Additional material is not automatically better when it distracts from required facts.
Frequently asked questions about Context Engineering
Is context engineering just prompt writing?
Prompt writing is one part of the broader problem of managing all information supplied to the model.
Does context engineering change model training?
Managing request context does not itself retrain the underlying model.
Can context engineering prevent every hallucination?
Better context can support grounded responses but cannot guarantee error-free generation.