四、第 2 层：browser-use（AI 智能体框架，底层还是 Playwright） browser-use 是开源社区里一个知名的"AI 浏览器智能体"项目，思路是：让大语言模型 自己观察页面、自己决定下一步点哪 ，而不是你写死步骤。它底层同样依赖 Playwright（或同类浏览器控制）来真正执行点击。

它把 Playwright 的能力完整包装成了 MCP 工具，AI 可以调用它导航到页面、点击元素、输入文本、读取页面快照、执行 JavaScript，甚至监听浏览器的网络请求。 这个 server 最厉害的一点，是会给 AI 提供“可访问性快照”。

2026年9月，OpenAI在DevDay发布了一批面向Agent开发者的新工具，Computer Use从一个"能跑但慢"的实验性功能，变成了一个在多数任务上比普通人操作更快的系统。Ari Weinstein负责OpenAI Computer ...

The researchers engineered mice to develop almost none of their own cerebralcortex, that's the outer layer that's sometimes ...

Around the end of 2024, I had the opportunity to build a performance testing environment for a web application.For the load ...

Engineering-focused Playwright automation framework built with TypeScript, featuring reusable fixtures, Page Object Model, authentication/session management, cross-browser execution, test data ...

This is a boilerplate/template for a Playwright-Typescript framework for web UI, API, mobile emulation, DB, and visual testing. Docker image, SonarQube, Lighthouse, GitHub Actions setup with Slack ...

AgentR today launched Webcmd, free open-source browser infrastructure that lets AI agents learn a website once and reuse that knowledge on ...

Have you ever had an AI build a site for you, only to find the design looks like something you've seen somewhere else? The ...

The race to build AI agents that can operate a computer has been defined by a single benchmark: can the model beat a human at clicking ...