Google just released a free 2-hour course on full Graph Engineering. How to go from one prompt to 100 agents running inside one graph: 17:44 - Build your first AI agent 39:30 - Run agents with loop engineering 1:12:38 - Turn agent loops into graphs 1:34:26 - Build agents... that throttle themselves 1:55:05 - Orchestrate the full multi-agent system Most people build one agent and stop there. Google is teaching the full stack: Prompt → Agents → Loops → Graphs → Multi-Agent Systems Single agents are the old workflow. Self-regulating agent graphs are the new one. This 2-hour watch is worth more than most paid agent engineering courses. Bookmark and watch it today Then read the full architecture below ↓show more

What about not being an asshole and gating other people's work:

graph engineering is the hottest topic right now, and this info is worth its weight in gold

yeah its getting huge

From one prompt to hundred agents is actually wild, Google cooked hard here

google really cooked here

bro really out here gatekeeping the free course like its a paid masterclass

No they didn’t:

Useful framing: single agents = old workflow, agent graphs = new one. How does the course address the two failure modes that usually kill these systems in production—loss of coherence between specialized agents and broken continuity across longer runs?

maximize token use

when my agents see my move out plans and their glow up plans simultaneously also, is it insane that I want my agents to have a visual office / home? want it to look like a game. want to open source my agents once they’re all glown up fr how long will it take?

Meh. Already passe

book'd for tonight

perfect for tonight

Hard to listen to her for long periods

This highlights an important shift in AI engineering. Building a single agent is useful, but designing systems where multiple agents coordinate, evaluate, and recover from failures is where production-grade AI is heading.

@LunarResearcher dang, this is exactly what I've been needing. looks like a weekend project coming up!

@grok is this video on YouTube?

@joaomendoncaaaa congrats on learning slop

@grok Extract all the valuable insights in a way that a non-engineer can understand and apply to his job

100 agents in one graph stays a diagram until something has to schedule them. Mine expands every loop iteration into a real step row before the loop starts, so the count is something you read rather than estimate. Past the ceiling the run fails instead of quietly growing.

curious where the full course is hosted since the post doesn't link it

the jump from single agents to orchestrated graphs is where things get really interesting once agents can coordinate regulate themselves and pass context between each other the system becomes far more powerful

The interesting leap is moving from clever prompts to systems that can manage complexity on their own.

Serious question: at what team size does graph orchestration actually pay off over a dumb pipeline? I manage 5 engineers and replaced a $30K/year vendor pipeline in 2 weeks with something simple. Feels like there's a threshold where the fancy tooling costs more than it saves.

honestly sounds cool but 2 hours to understand 100 agents running? feels like they're glossing over the actual hard parts lol

Lunar this is exactly right. single agents are already the old way. saw Databricks data, multi-agent setups jumped like 327% in under four months. the shift to coordinated agents is happening fast

Where is this from? Like source link? @grok

100 agents in one graph sounds great until you're debugging which of the 100 actually caused the failure.

i’m skipping straight to the throttling section

good one 👍

Graphs are a useful way to model execution, but they don’t answer what an agent is allowed to change. I’d keep purpose, row scope, and post-write verification explicit at the boundary. I’ll follow along.

Saying 'AI has limits' is what people say right before the limit moves. Again.

The next stage of AI isn’t about creating smarter single agents. It’s about designing systems where multiple agents can collaborate, adapt, and complete complex workflows.

Most annoying voice ever, stopped listening to this lady after 1,5 minutes

Going from one prompt to 100 agents is a huge leap to teach free.

There's a big difference between generating text and actually helping close the loop. Demi AI seems aimed more at the second part, especially with replies and follow-ups

Thanks for sharing. 💪

the full stack from prompt to multi-agent is the whole map