Science：AI智能体40分钟完成人类60小时实验；Cursor秘密开发办公助手Sand挑战Claude | Science: AI Agent Completes 60-Hour Experiment in 40 Minutes; Cursor Secretly Builds Office Assistant Sand to Challenge Claude
Science：AI智能体40分钟完成人类60小时实验；Cursor秘密开发办公助手Sand挑战Claude
2026年7月11日，两条消息同时炸场：Science期刊报道AI智能体将60小时实验压缩到40分钟，Cursor被曝秘密开发办公助手直接叫板Claude。
一、Science期刊：AI已经会自己做实验了
2026年7月11日凌晨，Science期刊报道了一项重大突破：AI智能体可在40分钟内完成人类需60小时才能完成的实验工作。
这意味着什么？
- 人类研究员：60小时 = 7.5个工作日
- AI智能体：40分钟 = 不到一个 lunch break
- 效率提升约90倍
这项成果标志着AI从”辅助工具”向”自主科研主体”的转变。过去AI帮助人类分析数据、撰写论文；现在AI自己设计实验、执行操作、收集结果。
对科研领域的冲击是深远的：当一个AI智能体可以不间断地运行实验，研究周期从”月”压缩到”天”甚至”小时”，科研产出的速度将呈指数级增长。
二、Cursor秘密开发办公助手”Sand”
同一天凌晨，AI编程工具Cursor被曝秘密立项”Sand“项目。
目标：开发能写代码的AI办公助手，直接挑战Anthropic的Claude Cowork市场地位。
这意味着什么？
Cursor目前是AI编程工具赛道的头部选手。但它不满足于只帮程序员写代码——它想成为一个能处理办公文档、管理任务、辅助决策的全能助手。
竞争对手很明确：
| 维度 | Cursor Sand（曝光中） | Claude Cowork（已上线） |
|---|---|---|
| 定位 | AI办公助手 | AI办公协作平台 |
| 代码能力 | 强（编程工具出身） | 中等 |
| 文档处理 | 待确认 | 已支持 |
| 市场地位 | 挑战者 | 先发者 |
开发者社区反应强烈。有开发者表示：”Cursor如果能把编程能力和办公助手结合，体验可能比Claude Cowork更好——毕竟它的代码理解能力是一流的。”
三、两个事件背后的共同趋势
这两条新闻看似无关，但指向同一个方向：AI正在从”回答问题”进化到”执行任务”。
- Science的实验AI：不是帮人类想实验方案，而是自己动手做实验
- Cursor的Sand：不是帮人类补全代码，而是替人类完成办公任务
这就是2026年AI行业的主旋律：从Copilot到Autopilot。
Science: AI Agent Completes 60-Hour Experiment in 40 Minutes; Cursor Secretly Builds Office Assistant Sand to Challenge Claude
July 11, 2026: Two bombshells dropped simultaneously — Science reports an AI agent compressed 60 hours of experiments into 40 minutes, and Cursor was revealed to be secretly building an office assistant to take on Claude.
1. Science: AI Can Now Run Its Own Experiments
On July 11, Science journal reported a breakthrough: an AI agent completed experimental work in 40 minutes that would take humans 60 hours.
What does this mean?
- Human researcher: 60 hours = 7.5 working days
- AI agent: 40 minutes = less than a lunch break
- ~90x efficiency improvement
This marks AI’s transition from “assistance tool” to “autonomous research entity.” Previously, AI helped analyze data and write papers; now it designs experiments, executes procedures, and collects results.
The impact on research is profound: when an AI agent can run experiments nonstop, research cycles compress from months to days or even hours, and scientific output accelerates exponentially.
2. Cursor Secretly Develops Office Assistant “Sand”
On the same day, AI coding tool Cursor was revealed to be secretly developing “Sand” — an AI office assistant that writes code and directly challenges Anthropic’s Claude Cowork.
Cursor isn’t satisfied with helping programmers write code. It wants to be a full-spectrum assistant that handles documents, manages tasks, and supports decision-making.
| Dimension | Cursor Sand (in development) | Claude Cowork (live) |
|---|---|---|
| Position | AI office assistant | AI collaboration platform |
| Code ability | Strong (coding tool DNA) | Moderate |
| Document handling | TBD | Supported |
| Market position | Challenger | First mover |
Developer community reaction: “If Cursor can combine its coding prowess with office assistance, the experience could surpass Claude Cowork — their code understanding is top-tier.”
3. The Common Thread
These two stories point in the same direction: AI is evolving from “answering questions” to “executing tasks.”
- Science’s experiment AI: not helping design experiments, but running them itself
- Cursor’s Sand: not autocompleting code, but completing office tasks for humans
This is the defining theme of AI in 2026: from Copilot to Autopilot.
来源：Science、格物之声、腾讯新闻