AgentWeb 发布，让 AI 智能体无需浏览器即可完成网站任务：不读截图、不点按钮，直接调用预置动作。实测 47/47 个实时任务全部完成，覆盖 15 个政府门户、7 家保险公司及 Amazon、LinkedIn 等 25 个需登录网站，中位耗时 1.9s，整轮成本 $3。每个映射任务以带输入 schema、前置条件、权限边界和完成证明的工具形式暴露，任意助手均可调用。
A website is starting to look like a set of callable operations, which is exactly how an agent wants to see it.
AgentWeb just launched and it lets AI agents do website tasks without a browser: instead of reading screenshots and clicking buttons, the agent calls a ready-made action.
The numbers are actually solid. 47/47 live tasks across 15 government portals, 7 insurers and 25 signed-in sites like Amazon and LinkedIn, 1.9s median, $3 for the whole run.
AgentWeb exposes each mapped task as a tool with an input schema, preconditions, an authority boundary and a completion proof, callable from any assistant.
Introducing AgentWeb: reliable web infrastructure for generalized agents. 47/47 real world tasks on live sites. 1.9s median. $3 total. Request access: https://www.agentweb.us在 X 查看被引用的帖子
来源：Rohan Paul · x.com