微软想给 AI 的工作标上美元价值,但算得可信吗

微软的新工具可估算 AI 智能体的商业价值。
它将 AI 成本与人类劳动的价值进行对比。
估算出的经济价值并不一定意味着真实的节省。
我们都知道 AI 可能很烧钱。但这些支出能否带来真实的商业价值?这正是微软希望通过今天推出的 Agent 365 Business Analytics 来回答的万亿美元级问题。
你的智能体正在创造多少以实际美元衡量的价值?
如果这套方法奏效,那确实相当酷。微软正尝试把劳动力市场经济学映射到 AI 使用数据上,从而得出价值估算。
为此,它使用了真实的劳动力成本信息,包括来自美国劳工统计局的数据、美国劳工部的 O*NET 数据库,以及微软自家的 LinkedIn Occupational Graph。
这些数据与从 Agent 365 收集的数据汇总后,被用于生成分析,旨在回答:
单个智能体、模型和用户的成本是多少?
AI 正在执行哪些业务活动(涵盖 6,000 多个已定义的使用场景)?
这些活动通常需要多少人力工作?
这些工作以美元计算价值几何?
增加或削减 AI 支出在商业上是否合理?
目标是全面呈现 AI 支出与所创造价值之间的对比。
将 AI 作为劳动力/软件的混合体来衡量
软件会计方法由来已久且直接明了。我们衡量订阅费、每席位成本、算力成本、间接成本(如场地空间和其他基础设施)等。但 AI 的工作并不契合这一模式,使其价值极难量化。
相反,Agent 365 Business Analytics 既相对于人力成本来衡量开销,也从技术支出的角度来衡量。例如,它可能会审视一项任务,判断这项工作通常需要投入多少人力,然后按照当前普遍的人力费率给这份投入定价。
接着,它可能会把任务中实际由人类完成的部分与 AI 贡献的部分各占多少纳入考量,再加上 AI 贡献的估算价值。随后,它还会把 AI 的实际处理成本(无论是订阅费、Token 消耗、本地部署基础设施还是云支持)一并混入计算。
最终得出的是一个混合型的人力估算:它把人类贡献、替代人力带来的节省、AI 成本,以及可能存在的“AI 相对人力成本的节省”结合在一起。这听起来实在太奥威尔式了。但说到底,企业其实一直没能看清 AI 这本账。
啊,但估算价值并不等于实际节省
不过有个问题。所有这些数字都只是估算。在博客文章中,Redmond 明确指出:“辅助小时数是对 AI 帮助完成工作的方向性估算,而非实测的节省时间或财务回报。”
换句话说,Agent 365 Business Analytics 估算出的结果,未必与现实中的实际体验或支出相符。也有可能是:任务之所以交给 AI,只是因为手上有 AI 可以做,而不是因为 AI 在处理每一个 Token 时都在为公司节省人力成本。
换句话说,微软或许能告诉你 AI 辅助工作的经济价值,却无法确凿证明企业真的省下了钱。
我觉得这件事尤其耐人寻味:它出自那家做 PowerPoint 的公司。因为现在有了 Agent 365 Business Analytics,你完全可能做出看起来非常、非常惊艳的经济仪表盘,但这些仪表盘与现实生活毫不相干。
这并不是说这个产品没用。但它确实说明,你必须相当自律地对待如何解读结果,以及在决策中对其寄予多少信任。它只是一种可能的真相来源,而不是绝对真相的来源。
Microsoft 想成为 AI 的控制中心
Microsoft 数十年来持续投资,打造企业计算的核心基础设施。就公司的整体使命而言,AI 也不例外。
借助 Agent 365 及公司的相关产品,Microsoft 希望帮你:
监控来自多家厂商的 AI 智能体。
管理它们的身份、权限与安全。
追踪并限制 AI 支出。
评估智能体是否在创造价值。
判断哪些 AI 系统值得追加投资。
这意味着该公司未必需要向你提供 AI 模型来帮你运营企业。通过与 OpenAI、Anthropic、Salesforce、Google、Databricks、Snowflake 等厂商的产品集成,该公司甚至不必提供应用或 SaaS 能力。
但如果它能介入并管理你的底层管道、帮你控制成本,并充当你的 AI 劳动力的 HR 部门,Microsoft 就可能掌握你企业运营的核心一环。
资深特约编辑
从计算机科学教授转型为 AI 创新者的 David Gewirtz 在人工智能前沿领域推动技术进步已超过三十年,并荣获 Sigma Xi Research Award in Engineering。他撰写过一篇具有开创性的分析文章,帮助塑造了早期围绕 AI 伦理的讨论。他也是 AI 产品商业化的先驱之一,其中包括 AI 语言和基于知识的系统。他是 AI Editor 的设计者,这是一个用于解析、分类和识别新闻报道的实验性人工智能引擎。David 是 Association for the Advancement of Artificial Intelligence 的成员。他目前任职于 InfraGard 的人工智能威胁与缓解跨部门委员会(AI CSC),InfraGard 是 FBI 与行业领袖之间的合作组织,旨在保护美国关键基础设施。他著有 Where Have All the Emails Gone?,以及 How To Save Jobs 和 The Flexible Enterprise。查看完整简介
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(Microsoft wants to put a dollar value on AI’s work – but can you trust the math?)
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.
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.
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).
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.
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.
That’s not to say the product is useless. But it does go to the idea that you will have to be quite disciplined in how you interpret the results and how much stock you put in them for decision-making. It is one source of possible truth, not the source of absolute truth.
Microsoft wants to be the control center for AI
Microsoft has been investing for decades to build out the core infrastructure of enterprise computing. AI isn’t any different when it comes to the company’s overall mission.
With Agent 365 and the company’s related offerings, Microsoft wants to help you:
Monitor AI agents from multiple vendors.
Manage their identities, permissions, and security.
Track and limit AI spending.
Evaluate whether agents are delivering value.
Decide which AI systems deserve additional investment.
This means the company doesn’t necessarily have to provide you with the AI model to help you run your enterprise. With integration with offerings from OpenAI, Anthropic, Salesforce, Google, Databricks, Snowflake, and others, the company doesn’t even have to provide the apps or SaaS capabilities.
But if it can get in there and manage your plumbing, help you control costs, and be the HR operation for your AI workforce, Microsoft could own a central piece of your business’s operations.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
Please accept cookies to view the comments powered by Disqus.
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. He wrote a pioneering analysis that helped shape the early conversation around AI ethics. He was also a pioneer in the commercializing of AI products, including AI languages and knowledge-based systems. He is the designer of the Al Editor, an experimental artificial intelligence engine for parsing, classifying, and identifying news stories. David is a member of the Association for the Advancement of Artificial Intelligence. He currently serves on the Artificial Intelligence Threat and Mitigation Cross-Sector Council (AI CSC) for InfraGard, a partnership between the FBI and industry leaders for the protection of U.S. Critical Infrastructure. He is the author of Where Have All the Emails Gone?, as well as How To Save Jobs and The Flexible Enterprise. See full bio
原文为英文,正文由自动抓取并保留原文。
原文发布于 ZDNet(Microsoft wants to put a dollar value on AI’s work – but can you trust the math?),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。