Andrew Ng just released a 2-hour course on full Graph Engineering. How to go from one prompt to 100 agents that loop, rewrite themselves, and run without you: 09:14 - Build your first AI agent 33:11 - Run agents with loop engineering 1:02:46 - Turn agent loops into graphs... 1:30:15 - Build agents that rewrite themselves 1:49:05 - Run the full graph system without you Most people are still building one agent and calling it done. Andrew Ng is already teaching everything after: Prompt → Agents → Loops → Graphs → Self-Improving Systems Single agents are the old workflow. Systems that improve and run without you are the new one. This 2-hour watch is worth more than most $500 agent engineering courses. Bookmark and watch it before everyone catches up. Then read how to run 1,000 agents from one prompt below ↓show more

ENS

the unattended graph section is probably worth the full two hours alone

yeah that section is gold

more agents only help if routing and evaluation are strong enough to catch the errors they multiply

routing and evals are everything

Blyat why he’s Andrew? He’s Ching Lu

nope, that's andrew ng

gold

thats true

Watching it tonight

perfect for tonight

that progression is the roadmap

yeah thats the roadmap

Graph engineering is where most agent projects either compound or break. Did Ng cover state management when graphs get nested more than two layers deep?

@grok give me the link of the mentioned course about Graph Engineering by Andrew Ng.

Loops create motion; graphs create inspectable state. The next maturity step is giving each node an evidence contract: inputs, outputs, evaluator, approval boundary, and artifact trail. Otherwise 100 agents just multiply unverifiable work.

A loop that can rewrite itself also needs a way to stop when the world changed underneath it. Do you model freshness or conflicting source state as a first-class exit, or only tool/error failures?

He’s the man. Took a bunch of his classes. What a legend

Her smile turns grace into a quiet power

Three pages of notes. The 1:49:05 timestamp answers a question I've been chasing for 2 months: how to make a render pipeline refine itself without manual intervention.

It’s probably best suited to content creators. Their workflow involves repeatable patterns. I can’t see using this in my work/life anytime soon

从 prompt → loop → graph 这条路径很符合实际体验。很多人卡在 loop 阶段就上不去了——agent 跑得动但没法 scale，因为 loop 之间的状态传递和错误恢复没人管。graph engineering 本质上是在给 agent 加拓扑约束，比裸跑 loop 可控太多

He has lost the capability to read something and research and then write essay on it. Who says researching and writing is boring and monotonous task 🤣🤣 Oh i forgot just because he is not good at it.

This is the actual ladder. Most people stop at “I built an agent.” The real work starts when you move to loops, then graphs, then systems that can improve without you sitting on them. Single agents are the old workflow. Self-improving systems are the new one.

good

Agentic workflows != Graph Engineering

Take a look at how works, you can do a lot of that there

what a time we are in now, writing loops is now an engineering field. there is literally nothing new about this course…

if you use AI agents and haven't even seen this video guide before I guess you just waste your time

Let's Loop n Flow!! 🎉✨

Very informative. But do you think it is like hype more than like a best practice?

@grok Can you provide me the link of course referred here?

从提示到自改写智能体，两小时课程把图工程讲透了

I’d spend most of the time on loop engineering.

the honest answer is most devs don't even know where to start with graph eng, never mind scaling it to 100 agents

Loops and graphs are the right unit to teach, and they are also where the bill quietly forms. A loop that resends the same system prompt and tool definitions every pass is paying full price for tokens that never changed. Prompt caching covers most of that, and batching anything nobody is waiting on covers a good part of the rest. Cheaper to wire in before the graph gets wide than after.