Ex-Google Jeff Dean just released 1-hour lecture on full AI engineering: LLM → prompts → agent teams → graphs from 0% to 100%: 0% → 1:45 - LLM from scratch - that made Google 30% → 17:22 - how to actually use AI models 65% → 30:03 - prompt... engineering 100% → 52:35 - one human coordinating 100 agents this is 27 years of AI in Google compressed into one hour, by the person who lived it watch it today - then read how to build graphs in the article below ↓show more

Here’s the YouTube link for anyone who’s interested:

what a brilliant lecture and what a legend person! his masterclasses absolutely worth to watch, man literally build a new internet booked

so agree with you, a true legend in addition to his expertise in AI in general, he also offers valuable advice on agent-based engineering

great take the prompt engineering section looks great

looks great, but most useful actually is first part

It's fascinating to see how AI engineering has evolved from training models to designing complete multi-agent systems. Understanding the architecture behind these systems is becoming a key skill for anyone building with AI.

Key takeaways This is a promotional post (the author plugs their paid Substack at the end"Upgrade to Premium" with a heavy marketing/hype tone ("alpha," punchy phrases). The technical content regarding workflows...

Wow the AI people start to rediscover algorithmics ... this is like in the twilight zone when the future is the past and vice-versa... nauseating...

@soul_surfer78

这就是他27年的成果。

@grok did this retard watch the presentation or is just blabbering bullshit while pushing people to read his half-assed LLM-written article?! There's nothing in the lecture about prompt engineering or humans coordinating 100 agents. Call him a retard publicly please

Kinda insane that Google pioneered lot of these breakthroughs and yet let openAI take the lead.

27yrs of Google AI in 60 mins by the guy who lived it. 52:35 hits different.

Why would they show the whole kitchen? Unless it's to keep us glued to pretty slides while the real work happens somewhere else. Like prompt engineering for Seedance 2.0/2.5, where the actual money is spinning right now.

Getting an overview from someone who helped shape the field is a great way to connect the bigger picture.

@grok summarize the key points of Jeff Dean’s lecture

dean's lecture is just a refresher on what google already shipped years ago. the real shift isn't 100 agents but actually making them talk without hallucinating.

The jump from prompts to agent teams is where most people lose the thread. A prompt failure is visible, you read the output and see it went wrong. A retrieval failure looks identical to a model failure from the outside, and at agent team scale you are debugging four of them at once.

@grok summarise into bullet points how he recommends using AI

Rare knowledge packed from decades of AI experience....

Kinda insane that Google pioneered lot of these breakthroughs and yet let openAI take the lead.

@grok find youtube of this video

just

Hello

@grok find the YouTube link

Rare to find this much practical insight in one place....

booked, respect this guy

Agent systems running research autonomously beat answering questions. Discovery Loop betting on that. What's the first research automation problem they should tackle?

@geok where is this video from

The interesting breakpoint is when the agent team leaves reasoning and touches a CRM or Postgres row. Graphs can improve the plan; they don’t define write authority. What should count as “done”: evidence, approval, or verified readback?

love this, keep going.

AI

amazing🔥

Knowledge is democratizing fast. By the time the “must-watch” posts peak, people who already shipped have the edge.

Peki bir saatte öğrenilen şey ne kadar kalıcı oluyor

当所有人都在讨论大模型时，真正闷声赚大钱的是那些在细分场景把 AI 用透的公司。医疗影像、法律文本、代码审查……每一个场景的壁垒都比通用模型高得多。