32 · Context Engineering

Context engineering is the new prompt engineering: getting the right information, in the right form, at the right time. Better context leads to better reasoning, actions and outcomes.

How it works

1. 1Context is everything the model sees: user input, memory, retrieved data, tool results, system prompt.
2. 2A context assembler collects, structures and dedupes; a retriever adds RAG; an optimizer compresses and summarizes.
3. 3Apply context optimization: retrieval filters, re-ranking, summarization, compression and token budgeting.
4. 4Context types & storage: short-term (session), long-term (user/memory), episodic (working).
5. 5A feedback loop feeds observations and results back in.

Key points

• Principles: include only what's relevant, reduce noise, put the right info close to the model.
• Use the right form (tables > text) and respect the token budget.
• Token budget strategy: allocate system, memory and retrieved data deliberately.
• Context is your competitive advantage — engineer it like code.

Why it matters

As models get better, the differentiator shifts from clever prompts to well-engineered context. Feeding the right information in the right form is what delivers real business value.

Frequently asked questions

How is context engineering different from prompt engineering?

Prompt engineering crafts instructions; context engineering manages everything the model sees — memory, retrieval, tools and structure.

Why does token budget matter?

Context is finite, so allocating it well (relevant, compressed, structured) directly improves quality and cost.