Google engineer: "In 2026 if you aren't building AI agents it's crazy how behind you are at Google, 85% of our engineers were running self-improving agentic loops & graphs" in this 32-minute talk a Google engineer reveals how the future of agentic coding will look worth more than 10... paid agentic courses watch today, then explore how to build self-improving agents with graphs in article belowshow more

graph engineering is definitely the real shift this year.

Yeah. So true bro. This the foundation of agentic ai in my opinion

queued for tonight. 32 min from a google engineer beats most conference recordings on signal density

Yeah bro. I’m sure you won’t regret after this watch

and yet nothing changed at google. where is that output?

This is kinda cringe to listen to

Wasn’t this the guy that did the interview a few months ago that the world would end by AI? So now selling books? Typical :/

@0xMovez this is wild to think about. Wonder how many solo devs can keep up with Google engineers on this stuff. Gotta start somewhere though!

wow bro, this is really top tier talk from google guy

yeah, the guy is a real legend on agentic building btw

85% of their engineers were not running self-improving agentic loops. They can’t even design a decent language model.

If you're not building tesseracts instead of loops and graphs you have no idea how behind you are..

Gay and retarded ai slop 🤡

it's will be interesting to listen Anthropic engineer

on agentic coding you mean ?

yeah yeah my bad

The "self-improving" part assumes the agent knows what better looks like, but that still needs a human decision on what wins

In 1959 Arthur Samuel wrote a checkers program at IBM that improved by playing itself. Sixty-seven years later Google calls it an agentic loop. The idea is old. The GPUs are new. FOMO is not a technology.

If we build more agents who benefits on the back of the api calls to llms...

So 15% of their engineers are "crazy behind"?

google 85pct loops still hide the eval stack outsiders lack

@grok who is the engineer, what’s his book on the table he keeps mentioning, and what conference are they at?

I have an agent stack, self built with no experience. I knew that my prompts were still poor so I built a prompt design loop, scored out of 20 on 4 criteria. Must get 3+ on each indicator and 16+ before the prompt can be executed. No longer missing context/constraints.

graph engineering is the next standard

Elon came thru with grok bot and now spaceXai. Grok bot will build the next systems of intelligence, logging, and governance, over frontier, agentic, ai model or agents, etc..) consolidate all that info to one point With no clutter And its fast af External web browser acess

Awesome content

A video of a Google engineer and a description for solving the problem with Claude. Any further questions about Gemini? 🤣

Sounds like he’s using Gemini.

Thank

Link to full video or transcript?

Dude I can’t even get access to my SIT environment. Every IAM they tell me to request doesn’t get me in.

The shift from copilots to self-improving agent systems will reshape how software teams build and ship products

It's getting agents to improve their own workflow while they're running.

Opencode FTW, Claude Code is 6 months ago

Yet every Chinese company is kicking googles ass in ai models. I love you google but your engineer needs context

We gotta map out all these models lol claude was trash. To code heavy for a more transional improvements. Simple easy to understand. Quick acess to topic orientation layered and indexed by weighted values to extra layers of operatins on top of all current ai models.

Are those loops shipping products or just shipping more loops?

The 85% stat is wild. Running self-improving agents that auto-fund their own API calls onchain feels like where we're heading already. Can't imagine dev workflows without them soon.

그래서 제미나이는 언제 발전해? 다른 ai와 비교도 안 될 정도로 성능이 떨어지잖아. 아직도 수박에는 왜 씨가 이렇게 많아? 같은 질문에 답이나 하는 수준이다.