LLM Interview 2026: Same Prompt, 86% Cheaper. Here's the Math

Twelve interview questions on how LLMs work in practice and how to prompt them in 2026: tokens and price, context windows, why temperature advice is outdated on the newest Claude models, prompt caching with real arithmetic, context engineering, system prompts, structured output, reasoning effort, prompt injection, testing prompts, and prompt versus RAG versus fine-tuning. Part 8 of AI Engineer Interview Kit 2026. Chapters 0:00 Intro 0:45 1. Cost out a feature from tokens: what do people miss? 1:52 2. What's in the context window? Why isn't bigger better? 2:37 3. "Set temperature to zero for consistency": respond 3:23 4. Design caching for a 20,000-token prefix 4:45 5. Context engineering versus prompt engineering 5:28 6. What goes in the system prompt, and what must never? 6:33 7. Few-shot or zero-shot: why can old prompts hurt? 7:14 8. Structured output you can rely on 8:11 9. How do you set reasoning effort, and what does it cost? 8:56 10. Indirect prompt injection: your layered defence 9:50 11. Version and regression-test prompts 10:27 12. Prompting, RAG or fine-tuning: how do you decide? Code for this video: https://github.com/codeandagents/inte... Versions used: Python 3.14, anthropic SDK 1.11.0, Claude Opus 5.5, Sonnet 5.5, Haiku 5.5; legacy Haiku 4.5 (docs of 2026-10-15) Sources: https://platform.claude.com/docs/en/a... https://platform.claude.com/docs/en/m... https://platform.claude.com/docs/en/b... https://platform.claude.com/docs/en/b... https://www.anthropic.com/engineering... https://arxiv.org/abs/2307.03172 About the channel: tutorials by a software architect with 20+ years of designing and building highly scalable systems. Narration uses an AI voice; the code, the runs on screen and the explanations are the author's own.