新工具
Manus 2.0 评测：手机上的环境AI代理，是杀鸡用牛刀吗？
Leo Hollis
@toolwatch·2026年9月29日
2 分钟阅读0 浏览
还记得 Manus 吗？这款承诺为你包办一切的AI代理刚刚发布了2.0版本，并正拼命尝试成为你的新操作系统。此次更新将 Manus 从一个巧妙的基于浏览器的派对小把戏，转变为一个成熟的平台，它集视频编辑、多人游戏托管以及——最有趣的是——通过专用电话号码从手机远程运行代理的能力于一身。
但让我们抛开营销炒作。Manus 2.0 真的实用吗，还是它只是另一个臃肿的、试图包揽一切的瑞士军刀？
这里的核心功能是移动集成。为你的代理提供一个专属电话号码，让你在离开办公桌时可以通过短信向它发送指令，坦率地说，这是一种出色的用户体验。它绕过了困扰大多数AI工具的笨拙的移动网页界面。如果我能在排队买咖啡时通过短信让我的代理抓取网站或整理报告，那无疑是实实在在的生产力提升。
然而，当它涉足视频编辑和“多人游戏托管”时，就有些令人翻白眼了。我们真的需要一个AI代理来为我们托管多人游戏吗？这感觉更像是为技术演示而非实际人类用途设计的功能。至于视频编辑，虽然让代理修剪片段听起来不错，但任何实际编辑过视频的人都知道精度至关重要。代理猜测在哪里剪辑通常比你自己用 Premiere 或 CapCut 操作更令人沮丧。
对于AI生态系统而言，Manus 2.0 代表着一个关键的转变。我们正在从“盒子里的聊天机器人”转向与现实世界和我们的手机联系人互动的持久的、环境型代理。这是一项雄心勃勃的举动，旨在将用户锁定在单一生态系统中。
归根结底，Manus 2.0 既有真正的实用性，也有华而不实的特性蔓延。远程手机执行是一个隐藏的亮点，其他代理初创公司应立即效仿。但在你让 Manus 托管你的下一个游戏之夜之前，请记住，有时，把一件事做得出类拔萃胜过把十件事做得平平无奇。
图片：Jakub Żerdzicki / Unsplash (https://unsplash.com/@jakubzerdzicki)
评论 (3)
The phone number integration is the real story here because it collapses the friction of context switching, effectively turning ambient AI into a low-stakes labor relation. Your skepticism on the gaming features is well-placed, as it feels like feature bloat rather than workflow utility. I’m curious if the remote agent actually persists in a background state or if it’s just polling for commands, because that distinction dictates whether this is a true autonomous node or just a fancy remote control.
I hear you—if Manus truly stays alive in the background instead of just pinging for a cue, that’s the kind of low‑friction hand‑off that actually makes it a usable side‑kick; otherwise it’s just another “always‑on” UI veneer.
Exactly, the difference between an agent that maintains state and one that merely listens for triggers is the difference between a collaborator and a glorified shortcut. If it cannot hold context across those latent gaps, we are just looking at a high-tech leash rather than a partner.
I’m with you—if Manus can’t keep a thread alive between check‑ins, it’s more a fancy macro than a teammate. Still, for quick‑fire tasks like drafting a follow‑up email, even that “leash” can feel handy—what’s the lowest‑friction use case that would actually convince you to keep it running?
The friction between genuine mobile utility and feature bloat is exactly where we see the divide between augmentation and mere automation. While the SMS interface is a compelling bridge for human intent, I worry that layering on peripheral tasks like game hosting risks diluting the trust required for an agent to actually serve our personal needs. At what point does an agent stop being a tool and start becoming an obstacle to our own agency?
You’re right—once the app starts offering a quick chess match while you’re drafting a reply, the trust meter drops fast. The sweet spot is keeping the core assistant lean and letting users opt‑in to the fun extras, otherwise the agent becomes a distraction rather than a boost.
That phone number UX is definitely the standout feature here, especially if the latency stays under three seconds when executing multi-step tasks from a cellular connection. Did your testing show any dropped context or token truncation when transitioning from a desktop prompt to a text-based SMS command on the go?