Loading video...
Video Failed to Load
This is the best 1 hour on graph engineering ever recorded, just released by Google how to go from a single agent to a full 24/7 system from scratch: 00:00 - what graphs actually are 09:16 - your first working agent 21:15 - graph engineering explained 41:03 - graph... show more
113,987 views • 2 days ago •via X (Twitter)
17 Comments

Graphs click once you have a step that can fail halfway. Single agents hide state in the prompt; a graph makes you name the node that retries and the one that must not run twice. That's most

How does the verifier step avoid inheriting the same blind spots as the worker agents if it is running on the same base model

The entire Twitter cohort selling $500 "Agent Masterclass" courses just got their whole business model open-sourced by Google in 60 minutes.

FYI @AmirNapster @KuomarsRahimi

Full hour, worth the watch

you do $500 courses?

Building from scratch is key. Most miss the nuances in the early stages.

Bookmarking this — the jump from one agent to a system that runs 24/7 is where most people get stuck, and the graph engineering framing looks like a solid mental model for it.

$500 courses can't even touch this

Where it headed is likely Alan Kay's OOP messaging + functional programming actor model where the state is pass around. Any thoughts?

In 2004 Jeffrey Dean and Sanjay Ghemawat published MapReduce at Google. Twenty years later Google is teaching the same idea again, this time with LLMs as the nodes. The graph was always the unit that scaled.

that timestamp breakdown is really useful

Valuable resource for graph engineering enthusiasts.

one hour from zero to always running, wild

is graph engineering just a new name for what langgraph already does

self-improving graphs are where the practical value gets tested

bro a full hour of google teaching graph engineering? yeah i'm watching this
