AI Engineering 101 with Chip Huyen
Chip Huyen, author of O'Reilly's most-read book 'AI Engineering' and former core developer on NVIDIA's NeMo platform, joins Lenny's Podcast for a technical primer on building production AI products. She explains the core building blocks—pre-training vs. post-training, fine-tuning, RAG, and RLHF—in accessible terms, drawing an analogy to Sherlock Holmes decoding language statistics. The conversation challenges the AI hype cycle, noting that while pre-training gains may be slowing, post-training and inference-time compute are driving the next wave of improvements. A centerpiece is Chip's viral chart contrasting what developers think improves AI apps (chasing the latest model, agonizing over vector databases) versus what actually does (talking to users, better data prep, end-to-end workflow optimization). She argues data preparation—not framework choice—is the dominant lever for RAG quality, and unpacks the economics of the AI data-labeling industry, where many startups serve just a handful of frontier labs. On adoption, Chip is pragmatic: most companies can't measure productivity gains cleanly, and randomized trials inside firms show senior engineers gain the most from coding agents, while mid-tier engineers follow. She forecasts organizational restructuring—blurring product, engineering, and evals—plus rapid advancement in voice and multimodal AI, where latency, interruption handling, and disclosure regulations remain hard problems. Closing on a personal note, Chip shares she's writing her first novel and draws a parallel between predicting a reader's emotional reaction and predicting the next token.
Original sources
Podcast Lenny's Podcast
Episode:Al Engineering 101 with Chip Huyen (Nvidia, Stanford, Netflix) ↗
YouTube Lenny's Podcast
Video:Al Engineering 101 with Chip Huyen (Nvidia, Stanford, Netflix) ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.