Context Engineering

Context engineering is the discipline of designing everything a model sees, not just the user-facing prompt: it encompasses system instructions, conversation history, retrieved documents, tool definitions, few-shot examples, and tool results that together shape model behavior. Prompt engineering focuses narrowly on crafting individual instructions, but in agentic systems the user's prompt is a small fraction of the total context; the majority arrives dynamically from tool results, retrieved data, and conversation state.

Every model has a finite context window measured in tokens, and treating that space as a budget is essential for effective agent design. You must allocate tokens across system instructions, conversation history, retrieved context, tool definitions, and the model's own reasoning and output, since exceeding the window causes silent truncation or errors while wasting tokens on irrelevant information degrades performance even within the limit.

The system prompt is persistent context the model reads before every user interaction: it sets the agent's role, behavior boundaries, output format, and decision-making rules, functioning as the agent's standing instructions that govern all subsequent reasoning. Well-designed system prompts specify what the agent should and should not do, include examples of desired behavior, and give the model a clear framework for handling ambiguous situations without asking for clarification.