正在加载视频...
视频加载失败
Anthropic engineer: “70% of our engineers are using self‑improving loops. Now everyone is building agentic Graphs" in 3-6 months, we’ll all be building graphs to orchestrate self‑improving agents. No more prompting.” in 30-minute lecture, Anthropic engineer shows how to build a self-improving agentic workflow from scratch. Worth more than... show more
44 条评论
I love this guy’s workshops on graph engineering. He worked at LangChain and is now shipping hard at Anthropic, so I’m sure Anthropic will shift toward graphs very soon.
Yeah. This is true. Happy he joined Anthropic team now !
haha, this video is from a few years ago (before I was at Anthropic) :)
the move from prompting to graph architecture feels inevitable, but evaluation will matter more than clever orchestration
Yeah, this is true. The right evaluation system for your agent is, in my opinion, the most important part of agentic orchestration.
Yeah I wonder why companies that profit from token burn would encourage practices that burn max tokens
We are in the graph part of the hype cycle (just like graphql).. Eventually everyone will figure out that this isnt needed and you can just use the basics
that video is from Jan 2025 which, in AI terms, basically makes it prehistoric there’s not a single concept in that video that is relevant anymore
great breakdown. do you think graphs become the default for smaller agents too, or only once workflows get complex?
In my opinion it’s worth using Graphs for bigger agentic setups
bro read 'graph architect' and immediately started updating his linkedin
loops just became nodes
Nodes are important parts of any Graph. Wait for the next drop on Codez, bro!
This is eerily similar to state machines. The concepts of nodes and flows are all very common speak there.
The shift from simple prompting to designing agent workflows is a fascinating evolution. Understanding how agents plan, learn, and collaborate will become an important skill for anyone building with AI.
great share my man! i just bookmarked it for later
thx browski ! always happy for you feedback.
Great to see that companies our moving our steps of context graphs which makes agent context aware. You can try out our MCP
is the 36:06 lecture from anthropic's own channel or a conference cut? bookmarked, alpha post!
It’s a workshop by Anthropic, with the Applied AI team at AWS and LangChain.
@0xMovez sounds like the industry's changing fast. need to catch up on these agentic graphs, feels like magic rn
The biggest change isn't that agents are getting better. It's that they're starting to improve the systems that build them. That's a very different feedback loop.
great video for improving skills on Graphs, thanks Movez !
Yeah bro, this is the best watch on RAG & Graph engineering so far.
bookmark
Hope its a useful watch for you Goaty !
Graph architecture is the natural next step after linear chaining
Yeah, this is true mate. Are you building Graphs for your agents ?
I must save this for improve my skills in Graphs
Yeah bro, this is the best thing I’ve watched on agentic engineering so far. Give it your time today.
The shift from prompting to designing systems feels like the natural next step. Workflows may become the real skill.
Prompt Eng to Context Eng to Graph eng... Ok.
graph architectures are changing the agent game so fast
AI corps needs more income so burn token while inventing your loop. Don't forget to stop loop otherwise you will brush dishes at antrophic...
Bunch of regurgitated slop. But you're just a bot amplifying so here's a cock.
Why I feel like we are at the stage where we are still trying/investing new way of leveraging AI - meaning nothing will last not even couple months.
Let the anthropic engineer do what they need and let us do what we need.
Yep, this is the part people skip. I've seen teams draw a graph on day one, then spend the week babysitting branches a tight loop could've handled. Timing the switch is half the game. I wrote more on this in 《The 100 Questions of Loop Engineering》:
anthropic's engineers are super smart guys whose opinion i'll always listen to
“Kendi kendine gelişen agentlar” bölümü videoda nerede anlatılıyor?
Self-improving agentic graphs is the part that flips the narrative. Once the model is responsible for its own prompts, every benchmark reframes — capability stops being what you ship, it is what the graph converges to overnight.
the loops vs graphs debate misses the actual bottleneck. whichever pattern you pick, the moment an agent does something irreversible like a payment, a deploy, a delete, you need a deterministic gate outside the model. built exactly that for an agent to agent payment flow recently, on chain escrow with a hard cap, and it's boring on purpose. keep the creative part probabilistic, keep the part that can't be undone deterministic.
@0xMovez could you provide the link to the original video please?
Is this just a new for of n8n?
