Jeff Dean's 1-hour AI engineering lecture: from LLM basics to graph-orchestrated agent swarms
glenbeer · x · 2026-10-05
- A resurfaced 1-hour AI engineering lecture by Google's Jeff Dean, distilling 27 years of building AI at Google.
- Structure: 1:45 LLMs from scratch; 17:22 how to use AI models; 30:03 prompt engineering; 52:35 one human coordinating 100 agents; 1:02:40 where coordination actually lives.
- Core arc: Prompts → Agents → Loops → Graphs. The author argues most people stop at prompt engineering, while Dean spends the last 20 minutes on what stays true next year — same model, same tokens, but coordination changes everything.
- The quoted article "Loops and Graphs" explains how to stop babysitting agents and only approve the final step.
More from coding & agent
- 32,000 shards, one loop: procedural Three.js metamorphosis built with Claude Opus — techartist_ · 2026-10-05
- Cloudflare Agents SDK adds Pi Durable harness for crash-resilient long-running agents — mitsuhiko · 2026-10-05
- Codex team pledges daily improvements for 28 days or a full reset — lxfater · 2026-10-05
- OpenAI Agents API in 10 steps: GPT-6.1 Sol nears GPT-6 Astra at one-fifth the price — Roger_M_Taylor · 2026-10-05
- Engineer 'fired herself' from the loop, runs 24/7 agent fleet in Grok Bot — the talk Musk reposted — Roger_M_Taylor · 2026-10-05
- DIGI-FUZE: a retro console where AI fuses any two game cartridges into a new game — Flomerboy · 2026-10-05