Andrew Ng just dropped the best 2-hour course on Graph Engineering: from single agent to full automation 9:14 - your first agent 33:11 - loop engineering 1:02:46 - graph engineering 1:30:15 - agents that rewrite themselves 1:49:05 - full graph system two hours, and it replaces every agent tutorial... you bookmarked this year Prompts → Agents → Loops → Graphs most people will stop after the first agent and call it automation he saves the last forty minutes for the graph that runs it without him same model, same tokens, completely different week watch it today the step-by-step guide is below, save it while it is still early ↓show more

okay the “agents that rewrite themselves” bit got me 👀 saving this one

the timestamps are a lifesaver graph engineering still feels like the part most of us are nervous to hand off saved before the feed buries it

@res1dualedge love it. i often see people skipping the loop engineering stage and wondering why agents aren't adapting.

keep posting more gold like these

Creating graphs with code is such bullshit. Fucking think about it. You use some visual tool to draw a graph to explain how the system is supposed to work and then go: now, when it's clear, lets obfuscate it back into code. Are you fucking insane or what?

1:02:46 graph engineering section is dense in the best way

Honestly, this was worth stopping for. Really enjoyed it.

learned a lot from his stuff before, watching this one too

Graphs are so great, I made a cloud to make them form automatically for you.

Where can I find the link to this course?

@grok 自分を書き換えるエージェントって何ができるの？

Most people are still optimizing prompts while the bigger leverage is moving up a layer into loops, routing and graph design.

@grok find this video on YouTube

好

do i need to know coding to understand and implement this?

a single agent you can debug by reading the log. a graph without tracing, you find out something broke at step 6 an hour after it happened

Andrew is a wealth of amazing information

How do you make it really work for a company, not just a single player level? Did anyone build it for a real company?

the self-rewriting agents bit is the one thing that actually separates this from every other tutorial out there. everyone stops at loops and calls it a day

Very good

timestamped like this is the only way a course tweet is actually useful. 'graph engineering' tho, at some point the naming gets funnier than the content 🤔

The real unlock is adding observability with tracing on the graph layer, not just more agent logic.

you need to be Graphmaxxing

2 hours?? yeah there goes my afternoon 😂

he was almost president 🫡

The self-rewriting section is the one that gets messy in practice. Agents that edit their own prompts or tool graphs tend to drift unless you snapshot each version and diff behavior on a fixed eval set before letting the new one run.

timestamped breakdown is the useful part. 1:30:15 'agents that rewrite themselves' is where i'd want to see the actual eval, not the chapter title

That's awesome! Andrew Ng is a legend. Definitely going to check this out. Thanks for sharing!

graph engineering > prompt engineering. the 1:30 section on agents rewriting themselves is the part i'd actually watch. 2h is a lot but timestamped so fine

ANDREW MY GOAT