LLM Inference Interview Questions, issue 7, Aug 5, 2026

The Summarization Paradox

Senior AI Infrastructure Engineer interview at Anthropic, and the interviewer asks:

You enabled prompt caching on a 100-step agent trajectory expecting a 5–10x cost drop. In production you’re seeing barely 1.3x, and most per-step content is cache-missing. What’s actually happening, and why does the order of your context decide whether caching works at all?

Don’t say: Caching must be broken, or the provider isn’t caching enough of my prompt.

How your cost-saving context condenser is secretly fighting your prompt cache, and why accepting an expensive hard reset is cheaper than constantly re-editing history.

The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.

Read it on Substack

Get the next one

Free on Substack. Unsubscribe in one click.