Term
Context Engineering
Context engineering is the discipline of deliberately curating and maintaining everything inside the LLM context window — system prompt, tool definitions, RAG hits, memory, conversation history.
Context Engineering — explained in more detail
Prompt engineering optimises the instruction sent to the model. Context engineering steps up one level and treats the whole context window as a scarce resource: which system prompts, tool descriptions, retrieved documents, notes and prior turns does the model see, and when? Anthropic defines it as an iterative strategy for managing all available tokens — selection does not happen once, but at every step of an agent. With longer agentic runs and 1M-token windows, the question “what belongs in, what doesn’t” matters more than “how do I phrase the prompt”.
Example / Practical context
In Claude Code you can watch context engineering in action: the agent writes a TODO list to external storage rather than keeping it in context; tool outputs are deliberately trimmed; files are loaded “just in time” with Read instead of all up front. Anthropic’s guidance: keep system prompts at the right level of abstraction (neither too rigid nor too vague), design tools to be self-explanatory and robust (every unclear tool description pollutes context on every call), and offload memory outside context, loading it only on demand.
Distinction from related concepts
Prompt engineering optimises input phrasing — context engineering curates the whole context across multiple steps. Retrieval-augmented generation (RAG) is a tool within context engineering: a way to selectively pull external content into context. Memory management overlaps heavily but focuses on long-lived state stores across sessions.
Entdecke mehr
Effort level and deep thinking: two independent axes for AI tasks
Effort scales breadth, deep thinking scales depth. When each setting makes sense — with three clear examples and one rule of thumb.
GlossarChain-of-Thought
Chain-of-Thought (CoT) is a prompting technique that asks the model to spell out its reasoning in intermediate steps — boosting accuracy on multi-step tasks.
LexikonPrompt Security — Injection, Leaking, Guardrails
How prompt injection, prompt leaking, and jailbreaks work — and which defenses (guardrails, spotlighting, sanitization) actually help.