Claude最近最值得关注的新功能，是把“会聊天”推进到“会动手”。通过计算机使用（Computer Use）能力，Claude可以看屏幕、移动鼠标、点击按钮并输入文字，把一段段指令变成可执行的操作流程。下面按真实使用场景，拆开讲清楚Claude新功能到底能做什么、适合谁用。
Claude “Computer Use”: the key step from answering to executing
In the public beta phase, developers can guide Claude in the API to operate a computer like a person: first observe the screen, then decide where to click and what to type. It’s well-suited for multi-step tasks, such as configuring items one by one in a web admin panel, batch-filling forms, or completing a sequence of settings in a product console. It’s important to note that this capability is still experimental—it may occasionally click the wrong thing or miss a step—so it’s better used in a “semi-automated execution + human review” workflow.
For product and operations teams: Claude’s new feature works more like a “workflow bot”
In the past, you could ask Claude to write an SOP, but you still had to follow it yourself. Now you can have Claude execute the SOP, pausing at key checkpoints for your confirmation. For example, organizing support tickets, verifying page settings item by item, or moving a set of materials into CMS fields—Claude’s new feature can reduce the friction of repetitive operations. For better reliability, it’s recommended to define the success criteria for each step, such as “this step is complete only when a certain button turns gray,” which helps Claude execute more consistently.
