What is context engineering?

Definition

Context engineering is the practice of designing, building, and maintaining the meaning layer that AI systems reason over — semantic models, ontologies, retrieval pipelines, business memory, and the evaluation harness that protects all of it.

How it differs from prompt engineering

Prompt engineering tunes the request. Context engineering tunes what the model knows about your business when the request arrives. A great prompt against a poor context still hallucinates. A modest prompt against a rich context tends to be correct, defensible, and reproducible. One is a craft of words; the other is a discipline of systems.

What a context engineer actually does

• Defines the canonical metrics, entities, and relationships the company agrees on.
• Owns the semantic layer or works hand in hand with the platform team that does.
• Designs retrieval — beyond vector search — so models fetch the right meaning, not just nearby text.
• Writes the evaluation harness that catches when meaning drifts.
• Sits between data, AI, and product — translates business questions into context the system can answer.

Why it is a real role, not a job title fad

Every meaningful AI failure in an enterprise is, at root, a context failure. Someone has to own that surface. Calling it nothing means everyone is responsible — which means nobody is. The book makes the case for the title and gives you the responsibilities, deliverables, and interview questions that come with it.

How to start without a new title

Write the ten questions your business asks most. For each, document the metric, entities, filters, and owner. That is your seed context layer. Connect one agent to it. You will see in days what most teams take quarters to learn.