Context engineering vs prompt engineering
Prompt engineering crafts the instructions you write to a model; context engineering manages everything the model sees — instructions, tool descriptions, retrieved documents, conversation history, and the budget that forces trade-offs between them. One is a writing skill; the other is an information-architecture discipline, and agents made the second one mandatory.
| Dimension | Prompt engineering | Context engineering |
|---|---|---|
| Unit of work | The instruction — wording, examples, structure | The window — what gets in, in what order, at what cost |
| Scope | Text you author | Everything assembled at runtime, much of it authored by others |
| Era | Single calls and chat turns | Agent loops with tools, retrieval, and history |
| Key failure | Vague instruction, weak examples | Right instruction drowned by noise, stale history, token overflow |
| Core skill | Writing and testing | Selection, ordering, compression, budgeting |
| Security relevance | Instructions can be argued with | Every context source is an injection path to manage |
The verdict
Treat prompt engineering as a subset that you still need — the instruction layer of a context you now have to engineer whole. For a single chat call, prompt work is most of the job. The moment a system retrieves documents, holds tools, or carries history across steps, the highest-yield work moves: a perfect instruction loses to a window cluttered with stale steps and unranked retrievals, and tool descriptions quietly steer behaviour more than your prose does. Budget your effort accordingly — instructions get you the first mile, but production agents live or die on what else you let into the window, in what order, and [what you refuse to trust once it is there](/guides/secure-agentic-ai).