Context Engineering
The discipline of deciding what an AI model sees on every call — instructions, retrieved data, memory, tool definitions, examples — and how to assemble them reliably as the workflow grows.
Production AI is not a prompt. It is a system of context, tools, permissions, traces, evals, and feedback loops.
What it is
Context engineering is the work of building the prompt the model actually sees. It's not 'prompt engineering' as one-off wording tuning — it's the systematic discipline of assembling instructions, retrieved context, memory, examples, tool definitions, and conversation history into a coherent input on every model call. It's where most production AI quality is decided.
Why it matters
The model is the same; the system around the model is what makes one team's AI feel sharp and another team's feel scattered. Context engineering is the cluster of decisions that determines whether the model has what it needs to give a good answer — retrieval design, chunk strategy, memory shape, prompt structure, example selection, tool surface.
How it works
Components: a stable system prompt (personality, scope, refusal patterns), retrieved context from a governed source graph, working memory and relevant episodic memory, examples selected by similarity to the current task, scoped tool definitions, and the user input itself. Each is its own slot in the context budget; each is testable; each is versioned.
Related resources
The total number of tokens an AI agent has available for instructions, memory, retrieved context, conversation history, and tool results — and how that budget is allocated across them.
Shrinking an AI agent's conversation history so the most relevant context stays in the model's window without exceeding the token budget — by summarizing, truncating, or selectively dropping turns.
The versioned, tested set of rules and templates that govern how prompts are assembled for an AI workflow — instructions, examples, formatting, refusal patterns, escalation language.
Before an AI agent can be useful to anyone, it has to be something — a coherent identity that holds up across users, sessions, and adversarial pressure. This is the research track that defines what that means and how to keep it stable.
The pattern where an AI agent retrieves relevant context from your data before generating an answer — instead of relying only on what the model learned during training.