People say prompt engineering became context engineering. What do you actually put in the context for a request, and in what order?
What they are really testing: The 2026 framing of the core skill: assembling the right information budget per request. Tests knowing the components, the ordering that exploits caching and attention, and that curation beats stuffing.
A real interview question
People say prompt engineering became context engineering. What do you actually put in the context for a request, and in what order?
What most people say
drag me
“You include the system prompt, relevant documents and the conversation history, giving the model as much relevant context as possible so it has everything it needs.”
"As much as possible" is the tell that they have not operated this: context is a budget where more is often worse, ordering matters for both cost and attention, and nothing here suggests they have ever measured what a component contributes.
The follow-ups they ask next
Why does irrelevant retrieved context hurt rather than just cost money?
The model grounds on what it is given: plausible-but-irrelevant passages compete with the right ones and produce confidently wrong syntheses. Precision at 5 beats recall at 30 inside the prompt.
A rule in your system prompt keeps being ignored mid-conversation. What do you try?
Restate critical constraints near the end of the context, closest to generation, keep them short and imperative, and where possible enforce them outside the model entirely, schema or filter, rather than escalating prompt emphasis.
What the interviewer is listening for
- Orders stable-first and can say why in cache economics
- Knows dilution and lost-in-the-middle, curates to the reranked few
- Ablates components against an eval rather than accreting
What sinks the answer
- More context treated as monotonically better
- No awareness that ordering affects cost or attention
- Never measured what any component contributes
If you genuinely do not know
Say this instead of freezing. Reasoning out loud from what you do know beats silence every single time, and a good interviewer is listening for exactly that.
“Context is [a per-request budget]: order [stable block first for the roughly 10x cache discount, critical rules in the system block, freshest material last], select [the reranked best few chunks, windowed history plus summary, examples that earn their place], and [ablate against the eval, because most contexts carry passengers].”
Keep going with prompting
Foundation
What actually belongs in a system prompt, and what is it unable to guarantee?
Mid
Your team has 30 prompts across 6 features, edited by 5 people. How do you manage prompts like production code?
Junior
How do you decide how to chunk documents for a RAG system, and what goes wrong with naive chunking?
Junior
Your LLM feature must return JSON that downstream code parses. How do you make that reliable?
Junior
What is prompt injection, and why is it a bigger deal once your model can use tools?
Junior
Your LLM feature works in the demo. What will it cost in production, and where do the surprises come from?
Knowing the answer is not the same as recalling it under pressure
Sign in to send the questions you fumble to spaced recall, so they come back right before you would forget them, and learn the concepts behind them with hands-on labs.
Start free