Anthropic engineer just released a 2-hour workshop on "Graph Engineering" for agentic systems: “80% of our engineers are using self-improving loops. Now everyone is building agentic graphs.” • 00:00 - Introduction to RAG & Graphs • 06:39 - Core of "Graph Engineering" (state, nodes) • 14:29 - 3 feedback... loops of Graph agents • 23:06 - Agent evaluation with Graphs • 36:29 - Agent cycles in graphs • 1:15:22 - Agentic RAG & agent context • 1:41:20 - Evaluation datasets based on Graphs This 2-hour workshop will replace 10 paid courses on agentic engineering. Watch it today, then learn how to become a Graph Engineer in the article below.show more

haha this talk was actually from 2 years ago (before I was at Anthropic)! :)

the shift from agent loops to agent graphs is happening fast

This is true, and Anthropic is always ahead of each new development. By the way, are you building graphs yourself?

not yet, still learning the architecture before building one. what stack would you recommend starting with?

This guy should be blocked from your feed!

Here’s the source. A year old stuff.

State management is where toy agents become production systems.

states / nodes / edges all matters in my opinion. thats the foundation of Graphs !

80% adoption makes this worth watching

Yeah bro, this is the best masterclass on RAG and graph engineering I’ve seen so far.

What exactly is “Graph Engineering”? This is just a LangGraph tutorial. There are thousands of them. Nothing new here. Definitely not an evolution of Loop engineering.

curious what changed internally at anthropic that made them ship this public instead of keeping it inside

The more people educate about building agents, the more they use Anthropics products and give them more money. That’s the main reason, in my opinion.

graphs everywhere

It’s the basic layer of agent orchestration. That’s the reason

your articles gets more attention than Dario himself XD

Lool. Thx bro. Happy to be useful for my audience! Btw, have you read the latest on graphs?

copying many of your posts into Perisan, Your letters are crazy! gj, keep going

Damn, legend! Thanks for sharing alpha with a wider audience. Let’s keep shipping!

was he early at langchain?

Agent cycles in graphs is part that i want to listen

Great. Then this video is must watch for you bro

absolutely banger my man

So true, mate. I’m sure if you spend just 2 hours on this single watch, you’ll learn graph engineering forever. Just spend 2 hours - it’s worth it.

isn't this just workflows and each node is a task? similar to how ci/cd pipelines are created or any workflow if you've worked in an enterprise environment. We used Logic Apps in Azure, self-hosted n8n workflwos, and StepFunctions in AWS. Is the only difference now that each node uses inference along with tools, API calls, etc.

2 hours spent, and an insane amount of useful info gained. thnx Codez!

Are you out of your mind? What has to do with Anthropic. He talks about langgraph AgentState, state graph, ReAct, Routing, @tool, binding with models, function/tool calling from nodes. This is old which came out Oct 2025 but new since LCEL. What is Graph Engineering in this.

graph engineer is gonna be on resumes by next month and nobody will agree on what it means

配套中文信息图：六原则/词汇表/菱形模式，扫图上手 Graph Engineering。

this workshop looks way better than most paid courses

nice share

Wow, that's incredible to hear about the adoption of self-improving loops and agentic graphs! Sounds like a huge step for

@hanspreinfalk

anthropic team finally sharing some of the sauce... let's use it to build something solid. check out

2 hour workshop from anthropic on graph engineering. free. saves 10 paid courses. that's the actual alpha

I've been doing this manually by breaking PRDs into Prompt phases, then creating an alignment doc to manage parallel building with two agents. This looks way better! Thanks! By the way, this could be the Orchestrator (App) killer too! LoL 😆 everything is a killer now!

the self improving loop part is the bit people sleep on. every agent demo looks clean til state management falls apart around step 4. modeling it as a graph is the only thing thats made mine less brittle. saving this one

Thanks for sharing...

We are bringing up Kimi K2.7 locally for a focused smoke test at our company. Specialist agents will build a graph that keeps sources behind every finding. If the experiment is strong, it will guide whether we evaluate Kimi K3 next. Keep you updated.

@grok summarise thus

After that the only asset that compounds is an income stream running on AI while you sleep.

Can I do it on my $20 plan ?

Partially implemented in my own harness.

His name is "Anthropic engineer"?

The self-improving loops stat is the one to watch here. 80% engineer adoption doesn't tell you how many are running those loops in production vs. prototyping, and production graph agents have failure modes that don't surface in a 2-hour workshop.

This is Lance, formerly of LangChain, right? Love learning from his videos