智能体 AI 时代：谷歌反重力引擎驱动下一代产品

Agentic AI Era: Google's Anti-Gravity Harness Powers Next-Gen Products

本期速览 · Overview

谷歌 AI 工作室的 Logan 探讨反重力智能体引擎如何统一谷歌产品，赋能智能体编程和消费级智能体。

Logan from Google AI Studio discusses how the anti-gravity agent harness unifies Google's products, enabling agentic coding and consumer agents.

智能体 AI 仍处早期阶段；编码智能体领先，但科学和金融领域将紧随其后。
Agentic AI is in early stages; coding agents lead, but science and finance will follow.
谷歌的反重力框架统一了产品，使智能体 AI 成为新的主线。
Google's anti-gravity harness unifies products, making agentic AI a new through line.
编码领域的狭义超级智能已经到来，比等待通用 AGI 更具影响力。
Narrow superintelligence in coding is already here, more impactful than waiting for general AGI.

智能体式 AI 在大多数 Google 产品中仍处于爬行阶段，Gemini 应用和 anti-gravity 更接近行走。
Agentic AI is still in the crawl phase for most Google products, with Gemini app and anti-gravity closer to walk.
编码智能体是最成熟的智能体用例，但科学和金融等其他垂直领域将紧随其后。
Coding agents are the most mature agentic use case, but other verticals like science and finance will follow.
模型正在吞噬脚手架；我们现在所称的模型包括智能体框架和工具，而不仅仅是权重。
The model is eating the scaffolding; what we call a model now includes agent harness and tooling, not just weights.
Google 的智能体框架（anti-gravity）正在成为连接所有产品的新主线，而不仅仅是 Gemini API。
Google's agent harness (anti-gravity) is becoming a new through line connecting all products, beyond just the Gemini API.
Omni 是一个单一的世界模型，模糊了视频模型和世界模型之间的界限，从视频编辑开始。
Omni is a single world model that blurs the line between video and world models, starting with video editing.

智能体式 AI 对搜索和用户参与是正和的，并非最初担心的蚕食效应。
Agentic AI is positive-sum for search and user engagement, not cannibalistic as initially feared.
Google 的成功指标应该是最大化客户成果，而不是最大化产品上的眼球时间。
Google's success metric should be maximizing customer outcomes, not maximizing eyeball time on products.
大多数用户尚未准备好接受完全自主的 AI；他们希望保持主导地位。
The long tail of users is not ready for fully autonomous AI; they want to stay in the driver's seat.
编码领域的狭义超级智能已经到来，它比等待通用 AGI 更具影响力。
Narrow superintelligence in coding is already here, and it's more impactful than waiting for general AGI.
尽管模型公司不断扩张，初创公司的机会比以往更多；专注是它们的超能力。
Startups have more opportunity than ever despite model companies expanding; focus is their superpower.
今年通过编码智能体加游戏引擎而非世界模型，将能实现“氛围编码”制作视频游戏。
Vibe coding video games will be possible this year via coding agents plus game engines, not world models.