⚡️萨提亚·纳德拉：无先验假设 x Latent Space 跨界特别节目，微软 Build 大会

让我们欢迎 swyx、saragawa、alad Gil，以及微软的董事长兼首席执行官萨提亚·纳德拉。

Please welcome swyx saragawa alad Gil and chairman and chief executive officer of Microsoft, Satya Nadella.

你好，萨提亚。

Hello, Satya.

能来到这里我特别兴奋。

I'm so excited to be here.

欢迎收听《No Priors》与萨提亚·纳德拉的跨界特别节目。

Welcome to a crossover episode of No Priors in Lane Space with Satya Nadella.

恭喜你完成了这么棒的产品。

Congratulations on an amazing build.

不用客气。

No.

非常感谢，能和你们两位在一起我很开心。

Thank you so much, and it's great to be with both of you.

我经常听你们俩的播客。

I listen to both of you or both the podcast all the time.

能上这个节目真的很棒。

It's great to be on it.

非常感谢。

Thank you so much.

你一早上都在聊微软各个部门那些令人惊叹的发布，好像说了三个小时吧。

So you're just talking about these amazing announcements from across the Microsoft estate all morning for, I think, three hours.

你最深刻的体会或者收获是什么？

What is the, what's the most important reflection or takeaway you have?

对我而言，最重要的一个想法是，不如把这看作一个生态系统，而不是单一模型或单一平台。

I I'd say there are perhaps the biggest one for me is let's conceptualize this more as an ecosystem play as opposed to a single model or even a single platform.

我从小在微软长大，经历过至少四次重大转型，我一直相信，一个平台的价值，关键在于它能为平台之外创造多少价值，而不是平台自己能捕获多少。

At least for me, grown up at Microsoft, having seen whatever four major shifts, I've fallen into that camp where a platform is defined by fundamentally its ability to create more value about the platform versus what's captured in the platform.

所以如果你看看现在发生的事，今天上午的主题演讲讲的就是：无论你是原生AI公司，还是传统企业，都能作为平等的参与者参与进来，推出自己打造的AI产品。

And so if you view what's happening right now, I think this morning's keynote was how can any company, whether it's an AI native company or a traditional enterprise company, participate as a first class participant where they can point to AI they create.

对吧？

Right?

他们并不是不用别人的AI。

It's not that they don't use other people's AI.

当然会用。

Of course they will.

但对我来说，关键路径是什么？

But to me, what's the path?

成功的秘诀在哪里？

What's the recipe?

我该怎么下手？

How do I do it?

整个技术架构该是什么样子？

What does the stack look like?

需要用到哪些工具？

What does the tooling look like?

到底什么才是最有价值的？

What is valuable?

你具体是怎么做的？

How do you do that?

就这么简单。

That's it.

这正是我们的工作。

That's our job to do.

对。

Yeah.

生态系统战略非常复杂。

Ecosystem strategy is very complicated.

对吧？

Right?

你得自己开发一部分组件，再找别人合作另一部分，还要去支持它们。

Because you end up building certain components, partnering for certain components, supporting them.

你刚发布了这套庞大的模型组合。

You just announced this big suite of models.

能跟我们说说微软的训练策略吗？

Tell us a little bit about the training strategy for Microsoft.

是的。

Yeah.

我们做MAI模型的目标，正如穆斯塔法提到的，首先是打造一个优秀的技术传承，从高质量的数据预训练开始，做各种消融实验，确保每一步都扎实。

So so the thing that we wanted to do with the MAI models was to build, and as Mustafa talked about, first of all, a great lineage, right, starting with pretraining with very good data quality, doing all the abolations, making sure.

因为说实话，现在要构建一个干净的模型传承越来越难了，外面的东西太多，你必须把无关的都剔除掉，才能做出一个出色的预训练模型。

Because in in some sense, it's becoming even harder to build a clean lineage model just because there's so much stuff out there that you truly need to ablate out to be able to have a fantastic pre trained model.

事实上，很多开源模型的问题就在于，它们在一两个基准上表现不错，但实际应用中却不行。

In fact, that's one of the challenges of a lot of the open rate models is they look great on one benchmark or two, but they're not great on practice.

所以这也是为什么我们的FDE团队对这些MAI模型这么感兴趣——一个只有50亿参数的小模型，是怎么做到这么强的？

So that's why, in fact, even in our FDEs are pretty gotten really excited about these MAI models because how the heck can a small 5B model hill climb?

这其实又回到了我认为最关键的一点：努力找到那个认知核心。

And it goes back a little bit to what I think is ultimately the key thing to do, which is try to pursue finding that cognitive core.

对我来说，先从一个干净的传承开始，再搭建一套支持体系，让企业不仅能用它做通用模型，还能基于它打造自己的专业模型。

So to me, starting with a clean lineage, then creating that ability for companies to be able to use this, right, not just as a generalist, but to create their own specialist by building this hill climbing scaffold around it.

对吧？

Right?

所以不只是模型本身，还得有这个渐进式优化的框架，之后你才能开始构建你的RLE。

So it's not just the model, but you have a hill climb scaffold around it, then you will start building your RLE.

你会开始收集使用轨迹。

You will start collecting the traces.

最重要的是，你会有自己的私有评估体系，因为我们都知道，现有的那些评估标准虽然不错、有意思，但其实已经没那么关键了，毕竟全都能被刷满。

Most importantly, you'll have private eval because we know all the evals out there are good, interesting, but they're not really that critical at this point because they all can be maxed.

所以关键在于，每家公司都该有自己的私有评估体系。

So the point is each company will have its own private eval.

我觉得，围绕我们的模型打造这套端到端的平台体系，才是真正有意思的地方。

And so that end to end platform story around our models is what I think is interesting.

还有一件事，莎拉，既然你提到了，我觉得确实出现了一个新前沿。

And then the one other thing, Sarah, since you brought that up is I do feel there's a new frontier.

人们总在谈前沿，问你是不是在前沿上操作。

Like, people talk about the frontier and are you operating at the frontier?

有趣的是，只要加入一点时间维度，就能用上——比如我们展示的兰多湖演示就非常棒。

Interestingly enough, if you add a little temporality to it, you can use, let's say in in in fact, the Lando Lakes demo we showed was pretty cool.

我们用的是那个叫GPT-55的东西。

We used whatever GPT-55.

对吧？

Right?

然后你收集了一堆追踪数据，再用一个五B参数的推理模型，效果就提升了。

Then you collected a bunch of traces, and then you took a five b reasoning model and achieved higher.

这其实也是什么叫‘站在前沿’的另一个体现。

So that is another aspect of what it means to appear operate at Frontier.

是的。

Yeah.

我首先得恭喜你，短短两年内在微软内部建起了一个前沿创新实验室。

I think I first of all, have to congratulate you on basically building a Frontier NeoLab inside of Microsoft in two years.

我想知道，你们现在推的这一整套AI战略，是怎么规划的？

I'm wondering, you have all this AI strategy that you're rolling out.

我很好奇，你现在知道了哪些事情，要是能早点告诉两年前的自己就好了？

I'm wondering, what do you know now that you wish you would tell yourself two years ago?

等一下。

Wait.

或者说是两三年前？

Or two or three years ago?

和詹森的合作是三年前，MAI是两年前。

Three years for the Jensen partnership, two years for MAI.

对。

Yeah.

我经常反思的一点是，我最初接触这个领域，是因为那篇《Gaining Loss》的论文让我很兴奋，后来OpenAI那边来找我们合作，他们说：

I think the thing when that I reflect quite a bit, which is obviously, I got into all this when I got excited by this gaining loss paper and when even the OpenAI partnership came about when those folks said, hey.

他们会投入大量计算资源和Transformer模型，这确实帮了大忙。

We're gonna really throw a lot of computer transformers, and they've helped.

对吧？

Right?

我总是回过头想想，真没想到。

The thing that I always go back and say, wow.

这些东西的能力确实在不断提升。

These things do have capability that they're climbing up with.

说白了，智能跟计算量的对数有关，这个说法还挺靠谱的。

This this crude way of saying it is intelligence is log of compute kind of works.

不过我觉得，我们可能低估了把这些技术真正落地到现实世界中的复杂性。

Now, what I think we underestimated perhaps is the real world complexity of deploying these so that they actually deliver the value in the real world.

对吧？

Right?

基准测试的结果固然有趣也很重要，但真正的考验是，外面的人能不能用它们做出独一无二、只有他们自己才懂价值的事，而且这种价值是能清晰衡量的。

So the outcomes as measured by any benchmark is interesting, important, but the true eval is when people out there are able to do unique things that they only can value, and it's very measurable.

对吧？

Right?

我希望我们整个行业能更多地意识到这一点。

That I wish we had even add more in our consciousness, right, which is as an industry.

因为现在，当人们说‘天啊，我可不想被Tokenmaxx限制住’，这其实反映出我们整个行业还没意识到，我们每一步都在用token创造价值。

Because right now, I think when people say, wow, I don't want a Tokenmaxx, it's an artifact of us not having thought ourselves as an industry that we are using tokens to create value every step of the way.

所以我觉得，这正是我希望我们早些意识到的，不过好在我们现在终于走到了这一步。

So I think that's what I wish we had gotten there, but I'm glad we are here.

你提到的那些应用场景里，有哪些是给客户带来最大价值的？

What are some of the use cases that you've seen that have created the most value for your customers?

我知道大家总在谈编程，我也很清楚，这方面的影响确实非常广泛。

Because I know that people talk a lot about code, and I think it's pretty clear that's something that's having very large scale impact.

除了编程之外，还有没有其他领域是你发现客户普遍受益的？

Are there other areas that you find in common that your customers are really benefiting?

是的。

Yeah.

我觉得没错。

I think yeah.

你说得对，现在确实已经做到了，不过有意思的是，Elad，我们居然还在讨论编程这件事。

To your point, obviously, is now got but it's interesting, by the way, Elad, to even talk about the coding.

对吧？

Right?

正因为代码这么有效，我们现在不得不重新设计开发环境。

Which is coding has worked so well that we now have to rebuild the IDE.

对吧？

Right?

这都是剪切操作。

It's cuts.

看到我们上线的东西，我简直惊了，居然有上百个智能代理会话。

To see what we launched is, oh my god, I have these 100 agent sessions.

这种认知负担全压回我身上，实在太大了，我必须得有个新界面。

I the cognitive load it transfers back to me as a human is so excessive that now I need a new UI.

顺便说一句，光靠聊天窗口根本行不通，所以我们才需要画布。

Oh, by the way, I like, the chat as the only artifact was also impossible, so that's why we need a canvas.

这其实引出了一个有趣的问题：软件和界面到底该用在哪儿？

So it's interesting for all the things about where is software needed or where is UI needed.

再想想代码吧，在完全由智能体驱动的世界里。

Repeat that even for code, right, in a fully agentic world.

不过话说回来，我们现在开始看到一种现象——协作中是这样，之前展示的自动巡航功能也是，关键在于很多人力其实都在做粘合性的工作。

But that said, one of the things that we are starting to see, we start seeing with co work, but even some of the work we showed with autocom autopilot on what with clause is a good one because if you think about a lot of human capital is doing the glue work.

对吧？

Right?

如果你能用长期运行、持久化的令牌智能体来增强这些工作，那么即使还是需要判断和粘合的任务，其扩展能力也会像编程一样得到极大提升。

If you now can augment that with Tokenagents that are long running, durable, then your ability to scale, even what is still judgment and glue work, gets amplified like coding does.

我敢肯定，六个月后，我们都会说：天啊，真没想到。

So you can I'm positive that six months from now, we'll all be saying, oh, wow.

整个晚上，一堆任务都在后台自动运行，都是我授权给那些自动驾驶系统去处理的。

Like, all through night the night, there was a bunch of stuff that all these autopilots that I have working on my behalf with my delegated authority, so to speak.

对吧？

Right?

就连我的身份都完成了一大堆工作。

I can given even my identity, did a bunch of work.

然后，我当然得让我的新ADE问一句：你都干了啥？

Then, of course, I'll need my new ADE to say, what did you do?

我到底有没有完成这项工作？诸如此类的问题。

Am I did I do this work and so on?

所以我觉得，流程的简化、任务的闭环，真正的价值就在这里产生。

So I think that's where compressing of workflows, completing of tasks, that's where I think a lot of the value gets created.

你提到一个特别有意思的观点：一边是真正写代码的智能体，另一边是围绕它的支撑系统。

I think you raised a really interesting which is there's the actual agent that's doing the code, and then there's a harness around it.

这个支撑系统就是运行环境。

And that's the environment.

也就是上下文环境。

That's the context.

作为开发者，你为代码智能体搭建的一切，都属于这个系统。

That's everything you're setting up as a developer around actually a coding agent.

对企业来说，这个支撑系统到底该是什么样子？

What is the harness for the enterprise?

在更广泛的生产力工作里，有没有类似的概念？你是怎么看待这个想法的？

Is there an equivalent concept for broader productivity work, or how do you think about that concept?

没错。

That's right.

某种程度上，你通过这个框架来定义模型、数据和工具，从而形成一个闭环。

So in some sense, you call the harness to define the models, the data, and the tools, and so that you have a loop across those three.

我们首先想确保的是，无论我们开发的哪个产品——比如 GitHub Copilot，或者之前展示的 Emdash 安全助手，甚至是科学发现工具——它们都不是孤立的。

And so what we are trying to, first of all, make sure is each of our products that we build, whether it's GitHub Copilot or the security copilot stuff we showed with Emdash, or even the discovery for science, it doesn't.

所有这些产品都是多模型框架，都支持工具调用，这样就能逐步展示功能，同时节省 token。

All of them are multi model harnesses with tools access so that you can do this progressive disclosure of tools even so that they're token efficient.

同时，你还要提供非常丰富的上下文，因为过去两年我们学到的另一个关键教训是：真正神奇的地方在于，你得花大量精力去准备上下文层，让计划能高效执行。

And then you're feeding it with very rich context because that's the other hard lesson we have learned in the last two years is, oh my God, the amount of work you need to do to prep the context layer such that your plan can execute in the most efficient way is where the magic is.

就我们而言，我们有一个 GitHub 框架，现在正用在所有产品里。

So we have, in our case, we have the GitHub harness, which essentially we're using across all our products.

它已经在 Foundry 里上线了，而且是开放的。

It's available in Foundry and we're open.

你可以用你的Lama框架，或者任何开源框架，甚至你自己的框架，搭配你的工具、多个模型和上下文进行训练。

Like you can use your Lama harness, whatever, or you can use the any open harness or any harness of yours and train with your tools and multiple models and your context.

这就是我们的核心主张，因为现在很多人在说：‘如果你把框架、工具和模型一起训练，就能得到评估结果。’

And so that's the pitch because right now a lot of dialogue is hey, if I train the harness plus tools and the model together, you get evals.

而我们正在验证的，最好的例子就是我们为Emdash所做的工作，对吧？

And what we are proving out is, and the best example of that is what we did with Emdash, right?

因为上线后，它发现了Mythos没发现的漏洞和安全隐患。

Because when it launched, it found bugs or vulnerabilities that were not found by mythos.

这已经提供了实证支持。

And so there is existence proof.

我认为，一个多模态框架在实际应用中完全可以表现得更出色。

I would claim, that you can have a multimodal harness that can in fact be more, performant in the real world.

独立前沿实验室的训练理念是：我们会拥有这些模型，建立API业务，服务企业和初创公司，但真正驱动大部分收入的，还是我们自研的产品——无论是生产力工具、代码助手还是搜索功能。

So premise behind the training at the independent Frontier Labs is really we're gonna have these models and we'll have an API business and we'll support enterprise and startups but a first party product be it productivity or code or search drives the majority of revenue.

这和你刚才说的价值逻辑是不一样的。

That's a different value equation than you're describing.

我觉得在微软的生态里，如果真是这样，你能不能确认一下？因为你既有自有产品，也有赋能型产品。

I think with the Microsoft ecosystem, if that's the case, tell me if it's the case because obviously you have first party products and you have enablement products.

那在这个体系里，开发者扮演什么角色？他们需要哪些关键技能？又该如何获取价值？

What is the role of the develop like, what's gonna be hard and the set of skills and the value capture the developer has in that world?

对。

Yeah.

我认为，一个平台构建者如果特别成功，完全有可能同时推出自有产品。

So I think that there's always gonna be the case that someone who is super successful as a platform builder can also have first party products.

Windows 就是这样。

It was true with Windows.

在 SaaS 和云服务领域，我们和其他公司也都是如此。

It was true with the SaaS side and the cloud side as well with us and others and so on.

但关键在于，这不该成为其他人取得同样成功的障碍，对吧？

But the thing that is, it should not be a limiter to other people achieving that same success, right?

我认为这才是核心区别：这一次，围绕智能的网络效应之所以强大，是因为它们从数据中学习，而且不需要海量数据。

That I think is the core difference, which is the network effects this time around, around intelligence are such because they learn from data and not really lots of data.

你只需要看几个例子，就能明白什么东西是新颖的。

It's just a few samples that you have to see to understand what's novel about something.

所以关键就在于怎么去保护。

So that's why the game becomes how to protect.

所以我觉得，每家公司都有自己私有的评估数据，这可能是最重要的知识产权。

So that's why I would say every company having private evals may be the biggest IP.

对吧？

Right?

我一直在想这个问题。

I think about it.

比如，你有个私有的评估数据集，能用前沿模型去不断优化，又不会留下痕迹，这可能是知识产权最重要的来源之一。

Like, what's that private eval that you can then use even a Frontier model to hill climb on and not leak the traces, maybe one of the biggest drivers of IP.

换句话说，另一个资产标准就是：你拥有一个私有的评估数据。

So in other words, another asset test is you have an eval that's private.

你正在使用模型A。

You're using model a.

你能换成模型 B 再往上爬吗？

Can you switch it to model b and climb up?

如果能，那你就掌握了主动权。

If you can, then you're in control.

如果不能，你就没掌握主动权。

If you can't, you're not in control.

这时候，就连选择评估框架都变得至关重要了。

And that's where even the harness decision becomes super important.

对吧？

Right?

所以，拥有一个开放的评估框架，让所有模型都能进来，用你的评估数据、上下文和工具来帮你优化，我觉得这正是 AI 原生初创公司、SaaS 公司，甚至每一家企业都需要的核心能力。

So therefore, having an open harness, letting all models come in, having your evals, your context, your tools help you hill climb, I think, is the skills that an AI native startup needs, a SaaS company needs, or every enterprise needs.

是的。

Yeah.

从某种真实的意义上讲，你就像当年的微软——先是操作系统公司，然后转型成云公司，也许第三幕就是你成为评估框架公司，或者 whatever，把你想要整合的这一堆概念都打包在一起。

I think in a very real way, you are Microsoft historically, as an operating systems company and then becoming a cloud company, maybe, like, the third act is that you're a harness or eval's company, whatever whatever the sort of conglomerate of concepts that you wanna put together.

我觉得，让每家公司都能拥有像前沿智能这样的能力——就是你之前说的那个词——这才是我们的使命。

I think enabling every company to have, like, Frontier Intelligence or what I forget the the exact term that you used is the mission.

对吧？

Right?

没错。

That's it.

这正是我们和你一起打造的平台承诺。

That is the platform promise that you build with us.

你会获得专属于你数据的智能。

You will get your intelligence for your data.

没错。

That's it.

对我来说，整个开发者大会的核心口号就是：每个人都能用前沿智能，像前沿一样运作吗？

That to to me, that is the there was one tag line for this entire developer conference is, can everybody operate at the Frontier with their Frontier Intelligence?

对吧？

Right?

我觉得这特别重要，不然我真不知道怎么实现稳定的平衡。

To me, that is so important because otherwise, I don't know how you achieve stable equilibrium.

对吧？

Right?

那我该怎么说呢？我的公司未来会有终极价值，因为我明白怎么在不断升级的平台上持续复利增长。

Which is how do I then go and say, my company is gonna have a terminal value because I now know how to continuously compound on top of what's a platform that gets better.

对吧？

Right?

比如当年Windows刚出来的时候，Adobe、Autodesk都做了自己的东西，就连Jensen说的那些也一样。

So when like, Windows obviously came out, Adobe built, Autodesk built, or even take what Jensen said.

我们做了DX，他则基于此开发了CUDA。

We built DX, and he built CUDA on top of it.

对吧？

Right?

我老跟Jensen说，唉，我这回真是捡了最差的那头。

I always say to Jensen, god, I got the shorn end of that.

对吧？

Right?

真希望我们当时能意识到这一点。

I wish we had recognized it.

但无论如何，能够打造一个平台层，让别人在此基础上扩展，构建自己的智能层——我觉得这才是关键。

But nevertheless, but that idea that you can build a platform layer that someone else can then extend out and build their own intelligence layer in this case, I think is everything.

对吧？

Right?

如果没有这个，那开开发者大会的意义何在？

Without it, why have a developer conference?

我大可直接让大家跪拜在一个模型面前就行了。

I can just come and have you all just worship at the altar of one model.

是的。

Yeah.

但那根本不是开发者大会。

But that's not a developer conference.

后台的时候，你们聊过什么是知识产权，或者公司真正的价值在哪里。

Back stage, you had a discussion about what is IP or what is the value in a company.

以前是看员工在公司待了多久，现在变成了另一样东西——也就是用智能代理解决实际问题的经验。

It used to be the length of human experience at the company, and now it's this other thing, which is the evals, the experience in applying agents to the company.

马库斯，你能再详细说说这个吗？

He I just want you to, like, flesh that out a bit, Marcus.

没错。

Yeah.

这个说法真好。

It's great way to frame it.

对吧？

Right?

归根结底，每家公司都会同时拥有两种东西：人类资本依然极其重要，因为人始终能发现那些持续存在的漏洞，这才是我们创造价值的根本。

Because, yep, at the end of the day, every company is gonna have both the human capital that is still gonna be super valuable because humans and their ability to find the gaps that exist at all times is going to be the way we all will create value.

对吧？

Right?

我 definitely 认为，即使代币价值上升，核心还是在于展现人类新的能动性和抱负。

So I'm definitely in the camp that this is going to be about expressing new forms of human agency and ambition even as token capital goes up.

对吧？

Right?

假设一家公司拥有大量代币和丰富的人力资本。

So let's say a corporation has lots of tokens and lot of human capital.

那问题来了，怎么把这两者结合起来？

The question is, how do you compound the two?

如果你看看团队，一边是很多代理在做事，一边是很多人类在做事，而它们之间的互动轨迹，恰恰是企业创造价值的关键背景。

So if you have a if you take in teams, I have a bunch of agents doing work and a bunch bunch of humans doing work and the traces between those, that is really important context of how that enterprise is creating value.

这就回到一个问题：别去训练通用模型，而该训练懂公司老底的代理。

Then that goes back to train not a generalist model, but to taint train the company veteran agent.

对吧？

Right?

这价值真的太高了，说白了，我觉得公司该把这种能力记进资产负债表里。

That is super valuable again again, right, which is when a company goes and says, it should in fact go on to the balance sheet is how I think about it.

对吧？

Right?

事实上，以前人类资本根本没法记入资产负债表，因为你没法捕捉那些隐性知识，但现在我觉得有了这些通过长期积累和行为痕迹学会一切的智能体，就能做到了。

That's in fact, there may be like, human capital was never possible to go put on a balance sheet because you didn't know how to capture the tacit knowledge, whereas now I think you can with the agents that have learned through the through time, through all the traces.

至少我们认为，未来就会是这样。

So that's what at least we think will happen.

我觉得美国证券交易委员会迟早得制定一套关于代币专业能力的会计准则。

I think the SEC is gonna have to have accounting standards for Token expertise.

你讲的是一个均衡状态，一种稳定的均衡，企业能积累这种复利价值，并看清自己的长期价值。

You're talking about the equilibrium state and a stable equilibrium where companies have this compounding value and can see terminal value for themselves.

另一个需要考虑的均衡问题是：有些应用和工作流程是某个行业或跨领域通用的。

Another challenge to be considered equilibrium of, okay, there are applications and workflows that are common to a vertical or horizontal.

这就像SaaS公司的兴起，微软也拥有大量SaaS产品。

And this was like the generation of SaaS companies and Microsoft has lots of SaaS properties as well.

而还有一些东西是每个企业独有的，正是这些让它与众不同。

And then there are things that are very specific to every enterprise that they're differentiated against.

我肯定你听过很多关于软件终结的讨论，毕竟现在这些工作流程生成起来太便宜了。

I'm sure you have heard much and participate much of the debate about the end of software because all these workflows are cheap to generate now.

你觉得未来，企业内部开发的智能体和供应商开发的智能体，它们的平衡状态会有什么不同吗？

Do you think the equilibrium looks different between what agents get built in enterprises versus in their vendors in the future?

对。

Yeah.

我觉得现在发生的是，我们过去用应用程序来捕捉工作流程的方式，其实挺特定的。

So I think what's happening there is, see, we had a particular way we captured, I would say, workflow in apps.

对吧？

Right?

因为我们当时是先设计数据模型。

Because we built a data model.

对吧？

Right?

我们把某个业务流程的一部分给结构化了。

We schematized some part of some business process.

然后我们写了一大堆业务逻辑。

We then built a bunch of business logic.

没错。

Yep.

接着在上面加了一层用户界面。

And then we put a bunch of UI on top of it.

对吧？

Right?

所以这就是每家 SaaS 公司的做法。

So that's what every SaaS company

再加一点配置。

And a little configuration.

在过去二十年里，这种模式

For the twenty twenty years, that

曾经是这样的。

was Yeah.

就这样了。

That was it.

有趣的是，现在你又得重新讨论这种垂直分层的问题了。

So interestingly enough, now you get to relitigate that vertical stacking.

对吧？

Right?

我觉得，比如每个SaaS应用下面的数据模型其实特别棒。

So I still think, for example, that data model that you build underneath every SaaS application is super good.

对吧？

Right?

干嘛非要重新发明一遍呢？

Like, why reinvent it?

我的总账就该是总账。

Like, my general ledger better be a general ledger.

我不需要每次都重新设计数据结构。

I don't need new schema creation.

不对。

No.

事实上，那个实体关系是个很好很稳固的结构，我希望能继续利用它。

In fact, that entity relationship is actually pretty good, robust thing that I want to feed.

而且你希望保持稳定。

And you want to be stable.

没错。

That's right.

当然。

Yeah.

业务逻辑也是同样的道理。

Then same thing with business logic.

对吧？

Right?

如果你看看我们有个叫 Power BI 的产品。

If you look at we have this product called Power BI.

对吧？

Right?

当时大家创建了无数个仪表板。

It was like dashboards galore people created.

这些仪表板背后其实是一个非常丰富的语义模型。

The beauty underneath that dashboard is a very rich semantic model.

对吧？

Right?

有人花了大力气去搭建仪表板，还做了各种计算指标。

Someone took the pain to create a dashboard and do all the measures.

而你希望这些就是你的业务逻辑。

And you want that that's business logic.

对吧？

Right?

我希望这些能直接为我所用。

I want that to be available to me.

所以我觉得SaaS商业模式的挑战在于，我们当初是这么打包的。

So I think the challenge of the SaaS business model is we packaged one way.

现在我们必须学会把这些东西拆开，再以新的方式组合起来，发现新的商业模式。

We now have to learn how to unbundle these things and rebundle in new ways and discover new business models.

对吧？

Right?

如果你看看现在微软365发生的变化，就是一个很好的例子。

If you look at it, what's happening today with Microsoft three sixty five is a great example.

对吧？

Right?

我们有个产品叫Work IQ。

We have this thing called Work IQ.

事实上，我们才意识到，天哪。

In fact, what we are realizing is, oh my god.

如果仔细看的话，其实这跟历史上某个情况很相似。

If you look at in fact, there's a historical parallel to.

对吧？

Right?

我们最早卖的是Exchange和SharePoint。

We sold first Exchange and SharePoint.

在Teams出现之前，我们还有Link Server之类的玩意儿。

And before Teams, we had a thing called Link Server and what have you.

当时我们以为，这些东西都会迁移到云端。

And we thought, oh, that's all gonna move to the cloud.

但我们根本没想到，用云服务器的人会增加十倍、一百倍。

But little did we realize that the number of people who will use servers in the cloud is 10 x, 100 x.

对吧？

Right?

因为以前根本没人买服务器。

Because people were not buying servers.

他们只是买个订阅服务。

They were just buying a subscription.

现在同样的事情正在微软365身上发生，因为通过Work IQ，我们把公司里最重要却一直没被当数据库用的数据暴露出来了——它以前只被我们的应用锁着。

The same thing is now happening with m three six five because with Work IQ, we have exposed what is perhaps the most important database in a company that never got used as a database because it is only captive to our apps.

所有邮件、Teams、Word、Excel、PowerPoint、SharePoint，全都在上面跑。

It was all email operated on it, Teams operated on it, Word, Excel, PowerPoint, SharePoint.

但现在，说真的，这真是我在Work IQ里觉得最酷的功能之一。

But now, like, this is one of the coolest things I get to do with Work IQ.

我会去GitHub的仓库里说，嘿。

I go to a GitHub repo and I say, hey.

上周我参加了好几个跟这个仓库有关的设计会议。

I attended a bunch of design meetings last week related to this repo.

你能把所有内容都整理出来，告诉我该做哪些调整吗？

Can you capture all that and tell me what changes I should make?

你想想这事儿。

Think about that.

对吧？

Right?

它真的能去查看所有这些会议记录，然后回来给你一份修改代码的方案。

It literally can go look at all those transcripts, come back with a plan to change your code base.

对吧？

Right?

以前你根本不可能想到能把 M365 用在这种地方。

Previously, you could never have thought of using m three six five for something like that.

所以在智能代理这个领域，现在创造价值的机会足足提升了十倍。

So the value creation opportunity now in the agent world is in fact 10 x more.

但这要求我们得有针对 M365 的使用场景，而且这种使用量可能比普通用户还要大。

But it does require us to have for example, there's gonna be usage around m three six five, right, which is gonna be perhaps more than even the end users.

我们甚至得重新设计架构。

And we have to even rearchitect.

以前用来服务收件箱或邮箱的系统，现在根本没法用来服务智能代理。

In fact, like what I used to serve an inbox or a mailbox cannot be used to serve an agent.

所以我们现在就在做这件事。

And so that's what we're doing.

我不相信这些领域有任何永久性的商业模式，但短期内，你对基于成果的定价、按令牌计费和企业套餐这几种方式，哪个更可能胜出有预测吗？

I don't believe in permanent business models for any of these domains, but in the near term, do you have a prediction between outcomes based pricing, token based pricing, enterprise bundles?

是的。

Yeah.

我是这么看的，我们不妨先看看按用户收费这种方式。

The way I think about this is always we've had let's even take the per user pricing.

按用户收费其实只是因为有人要制定预算，需要明确的支出预期。

The per user pricing is really an artifact of someone creating a budget needing certainty.

对吧？

Right?

因为这最重要。

Because it's the most important thing.

如果有人要报预算，那就必须得按用户来算。

If somebody wants a budget, they need a per user.

而按用户收费，本质上就是一组使用权限。

And per user is just a set of entitlements to usage.

对吧？

Right?

其实就是这么回事。

That's what it is.

所以第一个整合方式可能是，把一些使用量打包成按用户划分的套餐，然后卖订阅服务。

And so the way is, if the first bundling will be, take some usage, bundle it into per user stacks, and and sell subscriptions.

我觉得订阅模式会一直存在。

So subscriptions, I think, are gonna be there.

按用户收费的方式也会继续存在。

Per user is gonna be there.

接下来最重要的趋势会是按使用量计费。

Then the next big thing will be consumption.

到时候人们会说，我想要按使用量付费。

So people will say, I want consumption.

也有可能人们会说，我连订阅或按使用量付费都不想付，只想为结果买单。

And it's also possible that people will say, I don't even want to pay for any of the subscriptions or the consumptions outcome.

但别忘了，大多数人一开始都喜欢按结果付费，直到真到了要付钱的时候。

But remember, most people love outcomes until they have an outcome.

因为一旦真要按结果付费，就等于在分你的利润。

Because once you have an outcome, it's like giving away royalty.

对吧？

Right?

我接触过不少客户，他们都说特别认可按结果收费的模式。

Like, I've I've talked to customers who love outcome based pricing.

我说好啊，我完全支持，结果他们立马惊呼：等等，这不行！

I say, I'm all in until they oh my god.

你这是什么意思？

What are you talking about?

你这是要分走我的成果啊。

You're sharing in my outcome.

不是这样的。

No.

不。

No.

我希望你回到按用户收费的模式，同时也要考虑按使用量计费。

I want you to go back to per user pricing, and I want you to consumption price.

对吧？

Right?

我觉得这场争论还会继续下去。

So I think that debate will go on.

所有这些商业模式都有其适用的时机和场景，而不是有一种模式能包打天下。

And all all of these business models have a particular time and a place versus one to rule them all.

如果你是SaaS供应商或者平台提供商，这种灵活性就特别重要——说实话，我们在GitHub上也遇到了同样的问题。

And if anything, if you're a SaaS vendor or you're a platform vendor, having that flexibility and quite frankly, we face this with GitHub.

对吧？

Right?

我们最近刚在GitHub推出了按用户收费的模式。

We just recently announced a per user pricing on GitHub.

当初我们推出 GitHub Copilot 时，是按用户收费的，那时候还没意识到代理工具的使用强度会这么大。

Because little we GitHub Copilot was constructed at a per user level before we understood even the intensity of usage of agents.

对吧？

Right?

那时候它只是开发者用来辅助写代码、完成任务的一个互动工具。

It was an interactive way for developer to use CodeComplete, maybe task.

没人想到会一下子启动上万个代理，整天不停地跑。

It was not like, oh, I launched 10,000 agents that are going on all day.

对吧？

Right?

所以现在的调整就是针对这个问题。

So that is what the adjustment is about.

现在我们明确一点：用户订阅制还会保留，但必须加上按使用量计费的机制。

So now that we really want, there will always be a per user, but there will have to be a consumption meter.

你对 SaaS 模式的长期可持续性怎么看？

How do you think about the durability of SaaS more generally?

我注意到，很多企业内部都有团队陷入了一种‘代理狂热’。

One thing I've observed is in a lot of enterprises internally, there will be teams that almost have agent euphoria.

他们对能构建的各种新东西兴奋不已，甚至想重新开发很多应用，或者跑去对他们的 SaaS 供应商说：我们不跟你合作了，我们要自己搞。

They're so excited about the explosion of things they can build that they're trying to rebuild a lot of applications or going to their SaaS vendors and saying, we're not gonna work with you anymore, or we're considering an internal project.

但我觉得，六个月到九个月后，其中一些人可能会回头说：其实我们根本重建不了所有东西。

And it seems like in six to nine months, maybe some of those people will come back and say, actually, we can't rebuild everything.

你觉得在这个领域里，什么才是真正能持久的，什么只是昙花一现？

How do you think about what's durable in this world and what isn't?

我觉得得等一个完整的预算周期走完，才能真正看清这种平衡是怎么形成的。

I think we have to go through one full budget cycle on this to really see the the sort of the emergence of the equilibrium.

归根结底，生成一个应用的边际成本其实很低。

Because at the end of the day, there's marginal cost to even generating the app.

对吧？

Right?

其实换个简单说法也行。

In fact, there can be even a simple way to say it.

如果自己搭建和维护的成本更高，那就应该直接采购现成的。

If you should always acquire something if the marginal cost of building and maintaining something on your own is higher.

对吧？

Right?

这其实是个可以量化的问题。

That should be like, it's a quantifiable quantifiable thing.

而且维护这部分特别关键。

And the maintenance part is important.

对吧？

Right?

你得记住，别忘了。

Even like, you gotta remember, hey.

现在AI能发现的安全问题，你得赶紧修复，别拖。

All the security stuff that now AI will find, you better fix them too fast.

当然，有代码助手能帮你，但那样会消耗大量令牌。

Of course, there's a coding agent to help you with, but then that burns tokens.