正在加载视频...
视频加载失败
A Google DeepMind engineer just explained why even senior devs fail at building AI agents. This is the clearest explanation of how agents and loops actually work you'll find anywhere. People are paying $500 for courses that teach less than this 11-minute talk. Watch it, then read the step... show more
33 条评论
I watched that video and it changed my approach to building AI agents.
Glad to hear it, amazing
inference loops are where most teams break down
This is such a great explanation. Thanks for sharing this! It really clarifies how these agents work.
Watching it rn
This was insightful
This is awesome!
Stop building, start architecting loops.
the chads of the AI industry gives us more alpha than the courses
Most people are focused on the model. The real challenge is designing the system around it.
You need to try and it's better to do it right away.
We’re living in a non-deterministic world now, which i will note, is a much more human place to be.
Love seeing more people talk about AI loops. I've been thinking a lot about what happens outside the loop, how AI builds an understanding of the person it's working with. That's why @Jarvixlive caught my attention recently.
Good guide
Même constat chez NEREL — on build des agents Claude API en prod pour des groupes hôteliers n8n + Supabase + Claude = workflow complet en 30 secondes ⚡ On recrute un junior AI-native à Paris pour builder avec nous 👀 [email protected]
treating LLM calls like function calls = broken agents 💯
wild that it takes a deepmind engineer to explain what every failed agent demo was already screaming for a year straight
Paid courses pile useless jargon, hands-on takeaways from practicing engineers deliver far more value.
$500 courses teach less than this 11-minute talk. A Google DeepMind engineer just gave the clearest explanation of why even senior devs fail at building agents — and it’s free 🎯
11 minutes being worth more than a $500 course says more about the course industry than it does about the video.
will be interesting time to listen engineer from Google DeepMind
Wow, that's a fantastic breakdown. Definitely saving this one for later.
curious which part of loops specifically. most senior devs i've seen get the abstraction fine, the mess is always termination logic and state bleed between iterations. what does he say about that?
"Amazing. Every week there's a 'Google engineer explains why everyone else is wrong.' Somehow, software still gets built by teams—not Twitter threads."
This is super helpful! Thanks for sharing this explanation. It's great to see complex stuff broken down so clearly.
The failure cases are usually where the real lessons are
nice pick. will watch tomorrow
senior engineers with 10 years of experience are failing at this while 19-year-olds with claude and a weekend are shipping. that's what this talk is actually about
I keep pushing back on loop hype for one reason. A loop is only useful if I can inspect the boring chain: what triggered it, where it failed, what changed after the retry, and when a human got pulled in. Otherwise it is just automation theater.
Most senior devs still treat agents as chatbots that occasionally call tools. The ones that survive treat them as state machines with explicit contracts, budgets, and self-correction paths from the first commit.
great find, i will watch it today
The framing that "senior devs fail" usually means the failure mode is treating agents like deterministic pipelines. They're not. The loop behaviour under partial failure is where most production systems break, and no 11-minute talk covers that edge in enough depth.
the honest answer is most dev teams don't understand agent complexity until they're in too deep to turn back
