Audio walkthrough of Chip Huyen's Chapter 5: Prompt Engineering. Reframes prompting as software engineering — prompts are code, they have versions, they need eval, they get refactored. Covers: prompt structure (system / user / assistant), few-shot vs zero-shot (when 3 examples beat 0, when 30 hurt), chain-of-thought (CoT) and when it actually helps vs just adds cost, prompt templates and versioning, eval-driven iteration, context engineering vs prompt engineering, prompt injection / jailbreaks, structured output (JSON mode), function calling.
Most practical chapter for an AI engineer's day-to-day. The distinction between 'I prompt LLMs' (every PM does this) and 'I'm a prompt engineer' (you ship reproducible prompts at scale) is the chapter's core thesis.
