Skip to content
Home › Context Management & Reliability

Context Engineering

Context Management & Reliability

Definition

Context engineering is "the set of strategies for curating and maintaining the optimal set of tokens (information) during LLM inference" — the broader discipline of managing all information that enters the model's attention across multiple turns, of which prompt engineering is one subset.

Key points

  • Context engineering (verbatim): "the set of strategies for curating and maintaining the optimal set of tokens (information) during LLM inference."
  • Prompt engineering (verbatim): "methods for writing and organizing LLM instructions for optimal outcomes." Prompt engineering is a subset of context engineering.
  • Context = everything the model attends to: "system instructions, tools, Model Context Protocol (MCP), external data, message history, etc."
  • As agents run over more turns, the question shifts from "what prompt do I write" to "what configuration of context is most likely to generate the model's desired behavior."
  • Context is a finite resource with diminishing marginal returns: models have a limited "attention budget" drawn down by every token.
  • Guiding principle: find "the smallest possible set of high-signal tokens that maximize the likelihood of some desired outcome" — keep context "informative, yet tight." Curate "a set of diverse, canonical examples" rather than "a laundry list of edge cases."

Why it matters for the exam

Common gotchas

  • Prompt engineering ≠ context engineering — the former is writing instructions, the latter manages the whole token budget across turns. Don't treat them as synonyms.
  • "More context is better" is wrong: the attention budget is finite, so adding tokens has diminishing (and eventually negative) returns via Context Rot.

See also

Sources

Referenced by