Loading video...
Video Failed to Load
Google engineer: "just uninstall your IDE, in 18-24 months all code is gonna be written by LLMs at Google 85% of PRs are shipped by agents. If you aren't building agents in 2026, it's crazy how bad you are" in an 17-minute talk, a Google engineer with 30 years... show more
11,720 views • 1 month ago •via X (Twitter)
22 Comments

The 85 percent PR stat is wild but directionally right. The real bottleneck is shifting from writing code to reviewing and verifying it. Prompt precision and agent orchestration are becoming the new code review skills.

in 18-24 the bottleneck will be the human in the loop, lol

Another brilliant Yegge speech, thx for the share Movez

this watch changed my view on future of coding bro !

this past week i've been obsessed with google's engineers, they're really killing it they make incredible repos and such useful lectures and research

Steve Yegge wrote book vibe-coding. In my opinion he is a God Father of vibe-coding

lmao then i can name like 5-6 of these godfathers

respect movez. codez's 14-step graph engineering roadmap pairs perfect as the follow-up path

thx bro, but i trully recomend to watch this stage with Steve, brilliant words here

yep, sure

@0xMovez wild times, man. I've been tinkering with AI agents myself. Curious to see how this pans out.

if almost everything even at google done with AI agents, average people should do it for sure

as Steve told, if you arent building agent's in 2026, its crazy how bad you are. Agree with him

Google engineer: just uninstall your IDE

85%这个数字,听着像演讲现场加了点杠杆。

The 85% agent PR stat at Google is exactly what Virtuals is tokenizing on Base right now.

skipping pod hype for this engineer’s talk is smart

85% of PRs by agents is wild

85 percent of PRs shipped by agents sounds abstract until you've wired one yourself. I exposed our Face AI API as a custom ChatGPT GPT Action through a real OpenAPI 3.1.0 spec, letting it accept an uploaded face image through ChatGPT's own file reference mechanism.

"85% of PRs shipped by agents" is doing a lot of work with no definition of "shipped" attached. Agent-drafted and human-approved is not the same claim as agent-shipped, and the talk doesn't say which one Google means.

if you're not experimenting with llms for code gen this year, you'll be outcompeted within two

Thanks for this
