微软希望为 AI 的工作标上价格——但你能相信这些算法吗?

媒体报道 ZDNet AI 原文语言:英文

借助 Agent 365 Business Analytics,微软可能会重塑企业开支,但其中存在一个关键区别……

全文

Microsoft wants to put a dollar value on AI’s work – but can you trust the math?

With Agent 365 Business Analytics, Microsoft could reshape enterprise spending, but there's a critical difference between estimated value and money saved.

David Gewirtz

Senior Contributing Editor

Computer science professor turned AI innovator David Gewirtz has spent over thirty years driving advancements at the forefront of artificial intelligence and is the recipient of the Sigma Xi Research Award in Engineering.… Read full bio

David Gewirtz

Senior Contributing Editor

Read full bio

November 17, 2026, 6:05 pm ET

4 min read

Screenshot by ZDNET

ZDNET’s key takeaways

Microsoft’s new tool estimates the business value of AI agents.

It compares AI costs against the value of human labor.

Estimated economic value doesn’t necessarily mean real savings.

We all know AI can get expensive. But does that expense result in real business value? That’s the trillion-dollar question Microsoft aims to answer with today’s introduction of Agent 365 Business Analytics.

Also: Microsoft says its new AI coding model rivals Claude Opus 5.5 at 5% of the cost

How much value in real dollars are your agents producing?

If it works, this is some very cool stuff. Microsoft is trying to map labor-market economics onto AI usage data to create value estimates.

Also: Work IQ is Microsoft’s big bet on agent-first enterprise IT, and I have questions

To do so, it uses real labor-costing information, including data from the US Bureau of Labor Statistics, the O*NET database from the US Department of Labor, and its own LinkedIn Occupational Graph.

This data, along with data gathered from Agent 365, is aggregated into analysis intended to answer:

How much do individual agents, models, and users cost?

What business activities is AI performing (across more than 6,000 defined use cases)?

How much human work would those activities would ordinarily require?

What is that work worth in dollar terms?

Does increasing or reducing AI spending makes business sense?

The goal is to provide a comprehensive picture of AI spend against value produced.

Measuring AI as a labor/software hybrid

Software accounting is long established and straightforward. We measure subscription fees, cost per seat, compute costs, burden costs (like floor space and other infrastructure), and more. But AI work doesn’t fit that model, making it very hard to quantify value.

Instead, Agent 365 Business Analytics measures cost both relative to human labor and in terms of technological expense. For example, it might look at a task and determine how much human effort that task would normally take, and then price that effort based on prevailing common labor rates.

Then, it might factor in how much of the task was actually done by a human vs. how much was contributed by an AI, along with the estimated value of the AI’s contribution. Then it mixes in the actual AI processing cost (whether in subscriptions, tokens, on-premises infrastructure, and cloud support).

Also: These Claude Enterprise features let you track AI ROI – what’s new

What comes out is a hybrid labor estimate that combines human contribution, human-replacement savings, AI costs, and then potentially AI savings over human costs. It sounds positively Orwellian. But the bottom line is that companies really haven’t been able to understand the AI bottom line.

Ah, but estimated value isn’t realized savings

There’s a catch. All these numbers are estimates. In its blog post, Redmond specifically says, “Assisted hours are a directional estimate of work AI helped complete, not measured time saved or financial return.”

In other words, what Agent 365 Business Analytics estimates might not actually line up with real-world experience or expenses. It’s also possible that AI work might have been assigned because there was AI to do the work, not that the AI was saving company labor costs with every token processed.

In other words, Microsoft may be able to tell you the economic value of AI-assisted work, but it can’t demonstrate definitively that the company saved any real money.

Also: 5 reasons your company needs a chief AI officer in 2026

I find it particularly telling that this is from the company that makes PowerPoint. Because now, with Agent 365 Business Analytics, it’s entirely possible that you could produce economic dashboards that look really, really impressive, but are entirely irrelevant to real life.

这并不意味着该产品毫无用处。但这确实说明,你必须在解释结果以及在决策中对其信任程度上保持相当的纪律性。它是一种可能的真相来源,而非绝对真相的唯一来源。

Microsoft 想成为 AI 的控制中心

Microsoft 多年来一直在投资构建企业计算的核心基础设施。在公司整体使命方面,AI 并没有什么不同。

通过 Agent 365 及公司相关产品,Microsoft 希望帮助你:

监控来自多个供应商的 AI 代理。

管理它们的身份、权限和安全。

跟踪并限制 AI 开支。

评估代理是否提供价值。

决定哪些 AI 系统值得进一步投资。

这意味着公司不一定必须提供 AI 模型来帮助你运行企业。通过与 OpenAI、Anthropic、Salesforce、Google、Databricks、Snowflake 等提供商的集成,公司甚至不必提供应用或 SaaS 功能。

但如果它能够介入并管理你的基础设施,帮助你控制成本,并成为你的 AI 人员的 HR 运作部门,Microsoft 就可能拥有你业务运营的核心环节。

你可以在社交媒体上关注我的日常项目更新。务必订阅我的每周更新通讯,并在 Twitter/X 上关注 @DavidGewirtz,Facebook 上关注 Facebook.com/DavidGewirtz,Instagram 上关注 Instagram.com/DavidGewirtz,Bluesky 上关注 @DavidGewirtz.com,YouTube 上关注 YouTube.com/DavidGewirtzTV。

社区指南

请接受 Cookie 以查看由 Disqus 提供的评论。

David Gewirtz

高级特约编辑

计算机科学教授转型为 AI 创新者的 David Gewirtz 在人工智能前沿推动进步已有三十余年,并荣获 Sigma Xi 工程研究奖。他撰写的开创性分析帮助塑造了早期的 AI 伦理讨论。他也是 AI 产品商业化的先驱,包括 AI 语言和知识库系统。他是 Al Editor 的设计者,这是一款用于解析、分类和识别新闻报道的实验性人工智能引擎。David 是人工智能促进协会(AAAI)的成员。目前他在 InfraGard(FBI 与行业领袖合作保护美国关键基础设施的伙伴关系)的人工智能威胁与缓解跨部门委员会(AI CSC)任职。他是《电子邮件都去哪儿了?》以及《如何拯救工作》和《灵活企业》的作者。查看完整简介

David 的最新动态

Microsoft 想给 AI 的工作定价——但你能相信这些数学吗?

Microsoft 称其新 AI 编码模型以 5% 成本媲美 Claude Opus 5.5

想试用 ChatGPT Wallet?请先仔细阅读 OpenAI 的责任免责声明

/ latest

最佳早期黑色星期五 Apple 折扣:MacBook、AirPods 等

Google Wallet 现已登陆你的 iPhone——但缺少此功能

iOS 27.3 健康应用中的 5 项新 AI 功能——哪些 iPhone 可试用

免费获得 Pixel Watch 5 和 Buds 2A,搭配 Pixel 11 Pro XL 参加 Verizon 黑色星期五促销——如何资格认证

正文为 AI 译文。

参考来源

← 返回每日AI报告