#63 - 廉价的AI提高了错误判断的成本 - Florian Douetteau

我们正在进入一个世界，在这里，事物的真正成本不是最初的创造，而是管理这些“奶牛”的成本。

We are entering in a world where the real cost of things is not initial creation, the real cost of things is managing cows.

因此，Dataiku 专注于这一点，为企业提供一个强大的平台，让你能够控制所有让 AI 安全且成功所需的要素。

And so Dataiku is specializing in this, providing a robust platform in the enterprise where everything you need to make AI safe and successful is something you can put under control.

这样，你的数据就能得到控制，代理及其流程也能得到控制，你还能监控哪些代理真正完成了工作，哪些没有。

So that your data can be put under control, so that the agent and the agent processes can be put under control, so that you can monitor which agents do perform their work versus not.

我们希望为企业提供这个控制层，让他们能够真正按照自己的标准，而不是空谈和炒作，来实现 AI 的成功。

We want to provide this control layer for the enterprise that they need to use in order to actually, make AI really successful in their terms, not in the terms of talk and burn.

在本期节目中，我采访了 Dataiku 的首席执行官兼联合创始人 Florian Douetteau。这家公司十年前创立时，使命是让数据科学大众化，如今则专注于帮助全球许多大型组织安全地部署 AI 代理。

In this episode, I talk to Florian Douetteau, CEO and cofounder of Dataiku, a company that started its mission a decade ago around democratizing data science and is now focused on helping many of the largest organizations in the world to safely implement AI agents.

Dataiku 最让我印象深刻的一点是，他们帮助企业认清自身的现实，并创造出发挥自身优势和实现目标的解决方案，而不是像过去那样，让企业去适应所采用的软件。

One of the things that impresses me the most about Dataiku is that they help enterprises read their own reality and create the solutions that play to their strengths and their goals, as opposed to the old way of doing technology where companies had to adapt their business processes to whatever software they adopted.

Florian 谈到了他看到的当今企业在实施 AI 时采用的最佳实践和最令人印象深刻的商业成果。

Florian talked about the best practices and most impressive business outcomes that he's seen from enterprises implementing AI today.

他还谈到了现在招聘人才时他看重的三件事。

He also talked about the three things he's looking for when hiring people now.

当然，我们还聊到了他给领导者们的最佳建议，就是在人工智能时代如何最大化利用自己的时间。

And, of course, we talked about his best tip for leaders to get maximum leverage for their time in the age of AI.

这个问题我喜欢问每一位领导者。

The same question I love to ask every leader.

好的。

Alright.

那么，接下来就是我们和Dataiku的Florian Douetteau的对话。

So here it is, our conversation with Florian Douetteau of Dataiku.

好的。

Alright.

那么，Florian，请给那些不太了解的人简单介绍一下，Dataiku是做什么的，以及你在公司里的角色。

Well, Florian, tell a little bit for those who aren't familiar, tell a little bit about, what Dataiku does and your role with the company.

我是Dataiku的创始人和CEO。

So I'm I'm the founder and CEO of Dataiku.

这家公司是我十多年前创建的，使命是帮助企业通过人工智能取得成功。

It's a company that I created a little more than a decade ago, which mission is to help enterprise to find success with AI.

我们已经服务了相当一部分财富500强企业，帮助他们在AI方面取得成功。

And we are already serving a good portion of the Fortune 2,000 companies in terms of making them successful with AI.

这个使命很广泛，但我个人认为，有了AI，构建东西的成本正在趋近于零。

Well, it's a very broad mission, but the way I specifically see it is that with AI the cost of building things is turning to zero.

我们每个人把任何类型的想法变成实际的东西，都变得越来越容易。

It's easier and easier for any of us to turn any type of ID into something.

它可以是一份文档，可以是一种查看数据的方式，也可以是一种自动化工作的方法，等等。

And it can be a document, it can be a way to look at data, it can be a way to automate your work and so forth.

我们正在进入这样一个世界。

We are getting into that world.

但从企业的角度来看，管理和维护、控制以及治理所有被创造出来的东西，成本只会越来越高。

But from the perspective of the enterprise, the cost of managing and maintaining and controlling and governing everything being created will just get higher and higher.

我们正在进入一个世界，真正的成本不是最初的创造，而是管理混乱。

We are entering in a world where the real cost of things is not initial creation, the real cost of things is managing chaos.

所以Dataiku专注于这一点，为企业提供一个强大的平台，让你能控制所有让AI安全且成功所需的东西，让你的数据可控，让智能体和智能体流程可控，这样你就能监控哪些智能体在真正执行工作，哪些没有。

And so Dataiku is specializing in this, providing a robust platform in the enterprise where everything you need to make AI safe and successful is something you can put under control, so that your data can be put under control, so that the agent and the agentic processes can be put under control, so that you can monitor which agents do perform their work versus not.

我们想为企业提供这个控制层，让他们真正按照自己的标准来让AI成功，而不是只看烧了多少token。

We want to provide this control layer for the enterprise that they need to use in order to actually make AI really successful in their terms, not in the terms of token burns.

Florian，我们在这个领域也看到了一些实实在在的动向。

We're seeing some real moves in this area too, you know, Florian.

某种程度上，感觉世界正在朝你的方向转变。

Like in in some ways, see sort of the world moving in your direction a bit.

比如就在这周，微软发布了财报，市场反应非常热烈。

You know, even just this week, we saw, for example, Microsoft, report its earnings, and it had this amazing, response from the market.

微软也在关注如何让AI更容易部署、更安全、更稳健，很多方面都和你说的不谋而合。

And Microsoft, is really focused similarly on how do we make AI easier to deploy, safer, more, robust in many of the ways that that you're describing.

而对比之下，Meta则专注于如何成为OpenAI和Anthropic的竞争对手，打造世界顶尖的模型。

And then you had the contrast was Meta who's really focused on how do we become a competitor to OpenAI and Anthropic in building the world's best models.

但Meta发布财报后，市场并不买账，因为它花了太多钱想成为前沿实验室。

And it reported its earnings, and and the the market did not like the fact that it's spending so much money to try to become a a frontier lab.

所以从这个角度看，我们是不是看到AI的价值更多流向了那些把AI落地实践的公司，而不是那些在更高层次的理论、研究和模型上投入的公司？不过，模型本身也很重要，这也许是个两部分的问题。

And so in that sense, what are we seeing the value in AI accruing more toward the companies that are putting AI into practice versus the ones who are working at this sort of this higher level of theory and research and and models when, you know, the models are are so this is maybe a two part question.

模型是不是正在变成商品化？

Are the models becoming commoditized?

然后我们是不是看到价值更多地流向了那些把AI应用到实际中的公司和团队？

And then are we seeing the value accrue more to the companies and the the teams that are putting this to work in in the real world?

我非常坚信，至少对于企业用例来说，模型正在并且将会变成商品化，而价值将集中在光谱的两端。

I I have a fairly firm belief that, at least for enterprise use cases, models are becoming and will become commoditized and that the value will accrue on both ends of the spectrum.

你需要基础设施来运行这些模型，所以几乎要深入到数据中心、能源等等，这些需求会不断增长，就像推理标签一样，但从非常底层的角度来看。

You need infrastructure to run those models, so almost down to data center, energy and so forth, where the demand will grow and grow and grow and grow, like the inference tag, but from the very low level perspective.

而价值将集中在光谱的另一端，这不仅仅是创造和采用，而是真正能够在你的企业中控制AI的创造和采用。

And the value will accrue on the other end of the spectrum, which is not even creation and adoption, but the ability to actually have the control over the creation and adoption of AI in your enterprise.

因为很大程度上，正如我所说，创造本身正在变得商品化。

Because to a large extent, as I said, the creation itself is getting commoditized.

你可以把任何ID提示成某种东西。

You can prompt any ID into something.

你可以把任何ID变成一个应用或一个代理。

You can turn any ID into an app or into an agent.

但真正的问题是，你怎么做才能让它真正落地、长期运行，并且作为生产环境中的产物来维护？

But the real question is like, who do you do it so that you can actually keep it and run it for real and support it as a production artifact?

对企业来说，这才是关键。

And for enterprise, that's what matters.

这就像，你自己周末随便写个代码玩一玩，和——天哪，我们每个人现在可能都干太多这种事了，而且玩得挺开心——之间的区别。

That's the difference between, I don't know, something you vibe code on your own over the weekend and well, god forbid, like, everyone everyone of us, I guess, is doing that too much these days, and we're having a lot of fun doing this.

但我自己随便写的东西，我其实没打算长期维护，也没打算让除了我好朋友以外的人真正去用。

But all of the things I vibe code, I don't really intend to maintain them or to have anyone but my very close friends to use them for real.

我挺惊讶的，我的情况正好相反。

I'm surprised I have like the exact opposite.

他们理想中的状态是，东西只建一次，然后能以很高的信任度和安全性去维护。

They would like ideally to have things only being built once and be able to maintain them with a very high level of trust and safety.

所以我觉得，至少对企业来说，价值就在这里。

And so I think that's where the value would accrue at least for the enterprise.

好的。

Okay.

你知道吗，Florian，现在有很多公司来找我，说他们是要来构建安全AI代理的。

You know, Florian, there are a lot of companies that come to me now and say, like, we're here to, build safe AI agents.

他们是要来帮助企业做代理和AI的控制平面的。

We're here to help enterprises with a control plane for agents and for AI.

所以这种话我听得太多了。

And so I I hear this so often.

很明显，这说明市场上的需求相当大。

Clearly, means that the demand out there is quite significant for it.

但这也意味着大家都看到了这个需求，很多人都在争着成为那个帮助企业做这件事的公司。

But it also means that everyone is sort of seeing the need, and a lot of people are are racing for to be that company to help enterprises do it.

对吧？

Right?

还有小企业，实际上所有人都是。

And small business and and really everybody.

但是你的公司已经成立十多年了，就像你说的。

So but your your company's been around for, you know, like, over a decade, as you said.

那对你们来说是什么样的呢？

And what does that look like for you all?

你们是从哪里开始的？你们的旅程起点在哪里？

Did you did you start where did you start the journey?

嗯。

Mhmm.

那你们是怎么走到今天这一步的？

And and how did you end up here?

这里。

Here.

那么，在那么多试图提供帮助的公司中，Dataiku到底做了什么特别与众不同的事情呢？

And then, like, what is it that that Dataiku does that's sort of really different and distinctive in all the companies that are trying to help?

我觉得有两点，而且我认为我们在这方面确实比较独特。要真正帮助大型企业，尤其是那些组织架构有一定复杂度的企业，不是随便什么企业，你需要遵循两大原则。

My two picks there is that, and I think we're pretty fairly uniquely minded at least from that perspective, is that in order to actually help large enterprise in particular for real, I'm not talking about any type of enterprise, but let's say an enterprise with a certain level of complexity in the organization, you've got two big principles.

首先，你的控制平面需要在开放性和控制力之间取得很好的平衡，这是第一个原则。

First, your control plane needs to have this great balance between openness and control, which is one principle.

第二点是，在现实中，事情是分层的，就像蛋糕有多层一样。

And the second one is, well, in real life you've got multiple layers to the cake.

让我来说说这个。

So let me get to that.

首先，开放与控制意味着，你不能只是给大型企业提供一个控制平台，然后告诉他们一切都按这个模板来，用我们的模型运行，这样就能保证AI安全。

First, openness and control means that you can't just provide the control plane to a large enterprise by telling them this is how everything is built, fit into this template and run with our model, that's how you will make AI safe.

每个企业都有多种现有技术，业务遍布多个国家，有组织上的限制，有些用户你想给他们很多自由，有些用户则需要严格控制。

In every enterprise, you've got a multiplicity of existing technologies, you operate across multiple countries, you've got organizational constraints, you've got some users to which you want to give lots of freedom, and some others where you want to provide lots of control.

所以你需要真正帮助企业找到开放与控制之间的平衡。

And so you need to actually help the enterprise build that balance between openness and control.

这其实就是问题的关键。

That's actually what the problem is.

也就是帮助他们把开放与控制映射到他们的实际需求上。

That's helping them to map openness and control to what they actually need.

第二点是，事情是分层的，就像蛋糕有多层一样，要实现控制，你需要控制数据，因为数据是喂给智能体的，也是问题的一大部分。

The second thing is that there are multiple layers to the cake, as in to get to control, you need to control the data because that's what feeds into the agent and a good part of the problem.

你需要能够对业务中的一些关键流程进行非常严格的控制，但不是所有流程。

You need to be able to have very tight control on some key processes of your business, but not all of them.

你不必重新发明轮子。

You don't have to reinvent the wheel.

你还需要对业务中发生的一切有一个全面的了解。

And you need to have also a view on everything that's happening in your business.

这些智能体在哪里？

Where are those agents?

你如何大规模地管理它们？

How do you manage them at scale?

谁在做什么？

Who is doing what?

所以你有不同的方式来审视这个问题，所有这些都需要参与到控制中，如果你只提供其中的一小部分，你并不能解决问题。

So you have different ways to to look at it that all need to participate to to control, and you you don't solve a problem if you only provide a a tiny portion of this.

非常好。

Very good.

好的。

Alright.

很高兴你提到控制，我想稍微深入聊一下，因为我们一直生活在这样一个技术世界里，如果你是企业，你的主要任务就是选对供应商。

I'm glad you mentioned control, and I'd to double click on that for a sec because we've lived in a in a world with technology where you your your main thing if you're an enterprise is picking the right vendor.

因为一旦你选了那个供应商，或者按照传统方式选了那个供应商，你最终就会被迫改变很多自己的业务流程，对吧，去适应那个技术，适应那个供应商的技术。

Because once you pick that vendor or once you picked that vendor in in sort of the traditional way, then you would end up morphing a lot of your own business practices, right, to fit the technology, to fit that vendor's technology.

所以你必须谨慎选择，因为不管你选哪个，你最终都得重新调整自己的运营方式来适应那个技术。

So you had to pick carefully because whichever one you pick, it's gonna you're gonna end up sort of having to reshape the way you operate to to fit the technology.

就像，这就是过去几十年世界运作的方式，尤其是在软件领域，你知道的。

Like, that's the way the the the world has worked, especially in software, you know, for, the past several decades.

但我们现在正在进入一个世界，你可以不用把那些控制权交给选定的供应商，而是可以拥有更多的控制权。

But we're moving into a world where you can, instead of sort of giving over some of that, control to the vendor that you pick, you can have more control.

对吧？

Right?

你可以自己创建自己的软件。

You can you can create your own software.

制作软件的成本正在急剧下降。

The cost of making software is is dropping so dramatically.

所以当我听到你说控制权时，我理解的是——如果我说错了你纠正我。

And so when I hear you say control, what I hear and you correct me if I'm wrong.

我理解的是，我们进去和公司合作。

What I hear is you saying that we go in and we work with the companies.

我们帮助他们看清自己的实际情况，然后为他们现有的业务流程打造最合适的方案，而不是进去告诉他们，不，我们有我们的系统，你们得转换到我们的系统或模板上来，因为我们知道怎么运营。

We help them to read their own reality and then build the thing that's going to make sense for the business processes that they have in the way that are ideal for them, not go in and tell them, no, you have here's our system and we want you to come convert to our sort of system or template because, you know, we know how to we know how to run this.

据我理解，这是一个相当大的转变。

That's a that's a pretty big shift as I understand it.

我理解得对吗？

Do I have that right?

是的，这是一个相当大的转变，但同时也是——你提到的转变——我认为公司正在改变他们对采购软件或软件栈的思考方式。

Yeah, it's a pretty big shift, but it's also and you have combining shift, which is that, I think that companies are changing the way they think about, as you mentioned, procuring software or their software stack.

这不再只是选择一个首选供应商，作为记录系统和你们业务某部分的管理者。

It's less about picking a vendor of choice that are system of record and custodian of a part of your business.

关键是要么购买、要么构建最好的智能体，来改变你的业务。

It's about either buying or building the best agents that will transform your business.

我们客户的观点是，他们最终可能会购买90%的智能体，但会选择最好的软件供应商来做这件事，而不一定是为了销售或企业资源规划而保留的系统记录，而是希望在那里有更多的竞争和更好的性能。

And the view of our customers is that they might end up purchasing 90% of their agents, but like picking up the best software vendor doing this, not necessarily their custodian of recall for sales or enterprise resource planning, but like trying to actually have way more competition and smart performance there.

至于那10%真正需要自己构建的智能体，因为它们是针对特定业务的，他们确实需要能够自己构建。

And for the 10% of the agents they really need to build because it's specific to their business, indeed they need to be able to build by themselves.

另一个关于控制的问题是，确保他们以安全且可维护的方式构建，因为你可以通过偷工减料来快速构建，但问题是实际的维护成本、数据质量、评估循环，以及所有对重要流程至关重要的东西。

And the other question of control is making sure they build in a way that is safe and maintainable, because you can have the illusion of being able to build very rapidly by just by cutting things, but the issue is the actual cost of maintenance and the quality of the data and evaluation loop and like all of the things that are important for important processes.

这就是业务的现实。

And so that's the reality of the business.

一旦他们有了这个智能体的火花，比如10%是自己构建的，90%是通过高效采购获得的，利用各个供应商之间的竞争，他们的下一个问题是，我怎么知道发生了什么？

And once they've got this spark of agent, like 10% that they actually build themselves, 90% that they procure very efficiently, leveraging more of the competition between the various vendors, Their next question is like, who do I know what's happening?

我如何确保持续地挑战那些已经采购或构建的现有智能体？

How do I make sure that continuously I challenge the existing agents that are procured or built?

我检查它们是否真的在交付性能。

I check that they are actually delivering performance.

我要确认它们确实在履行职责。

I check that they are actually doing their job.

我要了解它们的成本，因为现在出现了一门新学问，不是管理人，而是管理智能体，让它们在你的业务中持续发挥绩效。

I understand their cost because you've got now a new science which would be about not managing human but managing agents and making them continuous team of performance in your business.

这就是现代企业。

So that's the modern enterprise.

而且，这不仅仅是关于ERP、CRM或其他什么系统的问题。

And indeed, it's not not only about ERP versus CRM versus whatever else.

对。

Yeah.

所以现在有很多公司会展示智能体的演示。

So there's a lot of companies out there that that will put together demos of agents.

对吧？

Right?

他们会直接向你展示所有东西，或者通过他们自己的视频、演示和概念来展示。

They will show you all the things in either directly to you and or sort of their own videos, their own demos, their own, concepts.

但是，是什么让你们与众不同？你们和很多公司合作。

But what distinguishes you all work with a lot of companies.

你知道，你提到过，你和财富1000强、财富2000强中的多少家合作。

You know, you mentioned, you know, how many of the Fortune 1,000, Fortune 2,000 you work with.

是什么让那些在演示中看起来很棒的代理，和那些企业真正信任、已经部署并正在做实际工作的代理区分开来？

What distinguishes sort of the the agents that look great in a demo versus the ones that enterprises can actually trust and are getting deployed and are doing, like, real work?

因为我觉得你可能对此有相当好的洞察。

Because I feel like you have probably have some pretty good visibility into that.

你需要对我们的理解方式有适当的控制，比如它是如何运作的。

You need to have the right level of control in terms of our understanding of like how it works.

我觉得有些事情虽然微妙但却是事实，那就是如果它是在做业务的工作，那么业务需要拥有它，而为了拥有一个系统，你需要理解它是如何运作的。

I think there is something which is a bit nuanced, but true, which is ultimately if it's doing the job of the business, the business needs to own it, and in order to own a system, you need to understand how it works.

所以我们看到大多数成功的方式是，通过拥有越来越复杂的代理系统，它们能处理边缘情况，考虑法规，随时间改进，有选择性记忆等等，但其中越来越多的行为是由业务中的主题专家理解和决定的。

And so the way we've seen most of the success is by having agentic systems that can be more and more complex, like they handle edge cases, they take into consideration regulations, they improve over time, they have a memory but in a selective manner and so forth, But like where more and more of those behaviors are actually understood and decided by the people in the business that are the subject matter experts.

所以一种思考方式是，当你从FrontierLab的角度来看，你有这种自我递归学习的代理，它们会随时间改进。

And so the one way to think about it is that when you take the perspective of FrontierLab, you've got this perspective of self recursively learning agents that will improve over time.

我觉得在大多数或许多企业的工作中，这种想法并不完全正确。

I think that in most or many enterprise endeavors, it's not exactly the right mindset.

你不应该抱着“闭环”的心态，也就是把人去掉，因为那并不是企业的最终目标状态。

You shouldn't be in the mindset of closing the loop, as in removing the human, because it's not actually the target end state for the enterprise.

你应该抱着这样的心态：把领域专家的专业知识转移和转化，让他们真正构建并拥有自己的智能体系统，这个系统会变得越来越复杂，但他们要真正理解其运作方式，这样才能对其负责，并找到正确的工作方式。

You should be in the mindset where you transfer and change the expertise of the subject matter experts so that they actually build an own agentic system that becomes more and more complicated, but like really understand how they work so that they can actually be accountable for them and find the right way to work on it.

实际上，如果你从事的是采购到付款、库存规划、欺诈检测或临床试验等领域，企业里会有人对欺诈被检测出来、生产按时完成、符合健康安全指南和要求等负责。

And in practice, if you work in, I don't know, procure to cash or inventory planning or fraud detection use cases or clinical trial and so forth, you have people in the business accountable for the fact that fraud has been detected, that production happens in time, that you meet the guidelines and requirements of health safety and so forth.

所以你需要保留这一点。

And so you need actually to keep that.

这就是大多数企业在深入这些业务场景时面临的实际问题。

And so that's the actual problem in most enterprises when you look into these deep business use cases.

这也是我热衷于此的原因，因为我认为如果实施得当，人工智能最终会在这些领域带来最大的益处。

And so that's what I'm passionate about because I think that's where AI would actually ultimately deliver the most benefit if implemented correctly.

你们那边也有这种氛围，你们在五月份做了一项调查，结果显示78%的首席执行官认为，如果他们处理不好人工智能，可能会丢掉工作，对吧？

You also have this atmosphere there where you all ran a survey in May that said 78% of CEOs, you know, think that AI could cost them their job, right, if if they don't get it right.

其实我记得我之前看到的数据是，到2026年，全球80%的CEO说如果AI搞砸了，他们的工作就有风险。

Actually, I I think that back I had it as 80% of c global CEOs say their job is at risk if AI fails, in in 2026.

你在外面看到的情况是这样吗？

Is that what you're seeing out there?

你有没有看到那种紧张感，就是领导者们知道他们需要AI战略，如果搞不好，对他们自己和公司可能会有很大的后果。

Are you seeing that level of tension of, you know, the leaders are they know that they or they feel like they need an AI strategy and that if they don't get this right, that, you know, there's there's sort of potentially big consequences for them and for their company.

这种紧迫感是不是也促使他们去做你刚才说的那些事情？

And has that created this level of urgency to do, you know, the things that you just talked about?

至少对你们来说，这是不是让你们能成为一个好的合作伙伴？

And and at least for for you all, does that let you be, you know, a good partner?

还是说你们也承受着很大的压力？

Or is there a lot of pressure on you all too?

我觉得压力是层层传递的，到处都是压力，这其实挺有意思的。

Meaning, I think there's like pressure cascading pressure everywhere, which is fun.

不过，要描述大公司CEO面临的战略困境，可以这样想：假设你是一家大公司，比如银行或者大型制造商，你的利润表里可能有大概20%的收入或成本是花在那些大型重运营上的，就像我提到的那些，是你业务的引擎。

But, like, one way to to to frame the, let's say, strategic potential dilemma for the CEO of a large company is that, imagine like you are any type of big company, like a bank or a big manufacturer and so forth, and you've got in your P and L, maybe, I don't know, 20% of your revenue or cost that are into big heavy operations, like the one I mentioned, like the engine of your business.

在这个核心业务里，你会有种直觉或者担忧，觉得可能有些竞争对手或同行能用AI改造它，把成本砍掉一半。

And in this engine, you've got this intuition or this fear that potentially some of your competitors or peers will be able to transform it with AI and slash the cost by two.

那20%就变成10%，等于他们多出10%的利润率，但你没有。

And the 20% becomes 10% becomes 10% of additional margin on their business, but not yours.

所以往前看，你会想，如果两年内我做不到，竞争对手的利润率会比我高10个百分点。

So if you move forward that, you say like in two years if I'm not able to do that, my competitors will be 10 points more profitable, but not me.

糟糕，那我饭碗就没了。

Oops, I will lose my job.

他们的公司市值会翻倍，而我的不会。

Their company value will double, not mine.

天哪，那我肯定得走人。

Oh shit, I will lose my job.

这才是真正的问题所在。

That's the actual issue.

从另一个角度看，他们会说，哦，我们要在公司里推行AI。

This is from the flip side, they say oh, we will implement AI in a company.

但要是我们在那些大型复杂项目里这么干，万一出了岔子，或者有什么地方跑不通，比如周三搞出个大AI事故引发丑闻，那我也得卷铺盖走人。

But if we do that in those large complex cases, and if we've got an issue, if we've got something not working, if we've got a big AI error on the Wednesdays that creates a scandal, I will also lose my job.

确实是这样。

That's true.

所以他们面临两难境地，或者说，他们到底该怎么评估AI带来的创新风险呢？

And so they have this dilemma, or like, who should they think about the innovation risk profile of AI?

好吧，什么都不做是一方面，但另一方面，他们也可能被AI关联的高风险压垮。

Well, don't do anything, but also potentially they can be overwhelmed by the level of risk associated to AI.

所以我在很多大型企业，尤其是银行、保险公司、制造商、公用事业公司以及医疗企业里，都能听到这种声音。

So that's what I hear in many, in particular large enterprises, or banks, or insurers, or manufacturers, or utilities, like healthcare companies.

如果你明白其中的利害关系，就会意识到这其实是一项高风险、高回报，但又不得不做的尝试。

If you understand the stakes, you understand that this actually is like high risk, high reward, necessary type of endeavor.

正因如此，最理性的做法是既要推进进展，又要建立好管控机制。这不仅仅是指安全模型或网络安全层面的控制，更是指如何以可控的方式，让AI在业务中越来越快地落地应用。

Which is why ultimately the rational way to think about it is that you need to progress but you also need to indeed have a control layer, not just in the sense of like a safe model or cyber security, but in the sense of really controlling how you can go quicker and quicker, but in a controlled manner, deploy AI in your business.

确实是这样。

Yeah.

很好。

Very good.

所以你刚才说的那些控制，很大程度上都跟数据有关，对吧？毕竟公司名字里就带着数据。

So a lot of that control that you're talking about comes down to data, right, which is where I mean, it's in the company's name.

据我了解，你们一开始就是以数据科学为核心的。

You all started, as I understand it, you know, data science at the at the core of the mission.

现在说到数据，在企业层面，你管理它的方式、控制它的方式，还有数据主权，以及你如何灵活运用数据，这些都很大程度上决定了AI项目的成败。

Now data, you know, when it comes to enterprise, you know, the ways that you can the ways that you manage it, the ways you control it, you know, the the data sovereignty, the ways that you'll be able to be nimble, you know, with your data, that feeds into a lot of the success or or failure of, you know, of AI projects.

对吧？

Right?

那么，公司的基因是如何真正围绕数据来构建的呢？

So how has sort of the DNA of the company really focused around data?

这又如何影响了你们服务客户、帮助他们的方式呢？

How's that impacted the way that, you know, you kind of show up for for customers and and help them?

对。

Yeah.

确实，我们看到数据策略或数据质量是许多大型流程中的决定性因素之一，因为归根结底，在大型流程中，你的成本来自于大量的异常处理、人工劳动、需要交叉核对的事情，而实现自动化的唯一方法就是增加高水平的数据、可预测性和对正在发生的事情的理解。

Indeed, we see that the data strategy or the quality of the data is one decisive factor for many of those big processes, because ultimately, in big processes, your costs are coming from the fact that you've got lots of exception, manual labor, human labor, you need to cross check things, and the only way you can automate is by adding a high level of data and predictability and understanding of what's happening.

所以这确实需要一些工作，甚至相当多的工作。

And so it does need a bit of work there, quite a bit of work there.

所以我们以数据和数据质量以及更好的数据的角度思考的能力，将非结构化数据与结构化数据结合，从分析的角度谈论数据，即如何回顾过去，同时也从预测的角度，即如何利用过去的数据来可靠地预测接下来可能发生的事情。

So our ability to think in terms of data and data quality and better data, and unstructured data with structured data, to talk about data from an analytic perspective, as in how you look backward, but also from a predictive perspective, as in how can you use past data in order to firmly predict what could happen next?

我认为这就是我们所有智能体用例的DNA，而且确实不是每个用例都需要先处理数据才能进行转换。

I think that's DNA of everything we put into agentic use cases, and indeed it's not every use case where you need data first to transform it.

如果你的用例是关于加速网站页面的创建以优化SEO，那可能不是数据优先的方法，可能只是智能地管理内容。

If your use case is about accelerating the creation of pages on your website to optimize for SEO, it might not be a data first approach, it might be just about like managing the content in an intelligent manner.

但如果你想自动化库存规划，如果你想加速你所销售产品的支持，如果你是海军工业制造商，如果你在半导体行业工作并想自动化生产链的质量，所有这些用例实际上都是相当数据密集型的。

But if you're trying to automate your inventory planning, if you try to accelerate the support of your product, you're selling, if you're a navy industry manufacturer, if you work in semis and want to automate the quality of your production chain, like all of those use cases are actually fairly data heavy.

你有很多智能体用例，但最终智能体的好坏取决于数据的质量。

You have lots of agentic use cases, but ultimately the agents are only as good as the quality of the data.

而目前，数据无处不在。

And currently, data is everywhere.

有时候数据还在人的脑子里，或者还在Excel表格里，或者其他什么地方。

Sometimes data is still in people's head or data is still in Microsoft Excel or whatever else.

所以，是的，这是一个非常重要的挑战，就是把数据提升到智能体层面。

And so, yeah, that's a very important challenge, is like move the data up to the agentic layer.

当你和公司合作制定AI战略或智能体战略时，你是从数据开始的吗？

When you're working with companies on their AI strategy or their agent strategy, do you start with data?

这是不是他们必须首先搞对的地方，才能让其他一切顺利运转？

Is that the first place that that they have to get right in order for for everything else to work?

我在这一行干得太久了，知道如果只是为了数据而做数据，最后只会开一堆关于数据的会议，但什么也做不成。

I've been in this business for too long to know that if you just work in data for the sake of data, you just end up doing meetings about data, but you don't get anything done.

什么也做不成，因为如果你只从数据的角度出发，你就会想，哦，什么才是正确的数据？

You don't get anything done, because if you start everything with only the perspective of data, you're like, oh, what is the right data?

关于数据，什么才是对的？

What is the right thing about data?

关于数据模型，什么才是正确的思考方式？

What is the right thing to think about the data model?

结果就是，房间里坐着一群对数据很有耐心、也很懂行的人，然后到处画各种图表，但你不一定真正理解那些驱动数据决策的实际业务问题。

And you just get in the room people that are patient enough but also very brainy about data, and so you do lots of diagrams all over the place, but you don't necessarily understand, let's say, the actual business problems that drive the decision making you need about data.

你不可能得到完美的数据，所以你需要的是针对实际问题的完美数据，这样你才能把数据管理好。

You can't get perfect data, so what you need is actually perfect data for the actual problems you have, and that's all you can actually make it manageable.

所以我们的方法其实更多的是让客户和业务部门自己去发现用例，自己做出关于数据的决策，了解数据哪里不足，有时候数据缺失是因为你根本没收集这些数据，所以你需要在这个过程中弄清楚怎么解决或者绕开这个问题。

And so our approach is actually way more to enable our customers and the business to discover use cases by themselves and to make the choices about their data, to understand where the data is lacking, to understand when sometimes you've got a lack of data that is inherently because you don't collect the data anyway, so you need part of the process to understand how you will fix that or go around it.

所以这其实是人们应该思考的方式。

So that's actually the way people should think about it.

数据访问和数据质量很关键，但最好由实际业务案例来驱动。

Data access, data quality is key, but is better driven by an actual business case.

当你现在走进公司，走进你合作的客户那里，他们向你提出的挑战是什么？

As you go into companies today, as you go into the customers that you work with, what are the challenges that they're bringing up to you all?

他们会说，嘿，这是我们最需要帮助的地方。

That they're saying, hey, this is what we need the most help with.

那么你在帮助他们应对这些挑战方面，最成功的方式是什么？

And and how are you having kind of the most success in helping them, you know, deal with those challenge?

我觉得除了数据之外，最近有两个问题是我最关注的。

I think that apart from data, I've got two that I see top of mind these days.

其中一个挺有意思的。

One is very funny.

就是，也没啥大不了的。

It's okay.

就是“我的智能体在哪儿？”

It's where are my agents?

在我们服务的客户中，想象一下有14000家公司，其中很大一部分已经不知道自己公司里有多少个智能体了。

Among the customer we sell, so imagine 14,000 companies, a good chunk don't know anymore how many agents they've got in their business.

你随便问他们，有多少个？

Quickly tell you, how many?

二十、五十、两百、两千、两千一百，或者三十四个，但他们就是不知道。

Twenty, fifty, 200, 2,000, 2,100, so 34, but they don't know.

而这只是更深层次问题的起点。理想情况下，你希望能在这样一个控制平台上，了解每个智能体的成本、价值、行为、风险，以及谁在负责它，至少先做到这些。

And that's the starting question for a longer question, is ideally you want to be able to have this, in this control plane, the ability to understand for each agent what is their cost, what is their value, what are their behaviors, what are their risks, who is in charge of it, like to start with.

这是你在组织里看团队时会遇到的情况。

Something you have in your organization when you look at teams.

对于一个团队，你会有各个业务部门，你得清楚每个部门的成本、价值或者关键KPI是什么，还有哪些需要留意的定性行为指标，就像你审视自己的业务一样。如果你进入了一个世界，其中1%、5%甚至20%的业务实际上是由AI驱动的，我认为这正是我们正在进入的世界，你就必须去管理好这部分内容。

For a team, you've got all your business units, you understand what is the cost of a business unit, what's the value or the key KPIs of a business unit, if you've got any behaviors of qualitative things you should be aware of, like you inspect your business, and if you enter a world where one, five, 20% of your business is actually energetic, which is I think the world we are entering into, you need to actually manage that part.

坏消息是，你需要重新发明什么是“Agent管理”。

Bad news, you need to invent what agent management is.

好消息是，如果装备得当，Agent管理很可能比人类管理要简单一些。

Good news, it's likely that agent management is slightly simpler than human management if properly equipped.

我觉得作为一个物种，人类还没真正解决管理人的问题，这依然是个开放性的难题。

I think that as a species we haven't solved human management yet, really, it's still an open problem.

但代理管理或许是个能解决的问题，这是好消息。

Agent management could be a solvable problem, that's the good news.

所以这就是第一个问题。

So that's the problem number one.

第二个问题可能更具体一些。

Problem number two is maybe more into the weeds.

很多大型组织都面临这样的问题：他们想要改造业务流程，比如库存规划、产品质量、预测性维护、了解你的客户这类大领域，他们想重新定义流程，但与此同时，这些流程本质上很复杂，需要审计，而且每个国家和业务单元都有各种变体等等，所以他们想围绕这个来思考：我们到底怎么管理这一切？

Many large organizations have this problem where they want to transform business processes, as in like inventory planning, product quality, predictive maintenance, know your customer, like this type of big domain, and there, they want to redefine processes, but at the same time, inherently, those are processes that are complex, you need audits, you've got x variants of them per country and business unit and so forth, and so they want to wrap the red around like, who do we actually govern all of that?

事实上，我们到处都会有智能体，我们希望有集中的定义，但同时也要有灵活性，这样每个业务单元都能把事情做好，等等。

The fact that we will have agents all over the place, we want to have central definition, but at the same time flexibility so that things can get done for each of our business units and so forth.

所以我在市场上看到的大型企业的第二个问题，还是因为这是智能体规模化应用的真实问题，不只是说，哦，你做个概念验证，而是当你把它应用到组织的40个分支机构时，它在实践中真的有效吗？

So the second problem I see a lot on the market for large enterprise, again, because that's a real problem of scale of agentic, not just like, oh, do you do the proof of concept, but like when you apply it across 40 branches in your organization, or does it work in practice?

那些真正用AI推动变革的领导者，当你看到那些能够用AI和智能体推动变革的公司里的领导者和团队时，他们做对了什么？

The leaders that are really using AI to drive change, when you see sort of leaders in the companies that are able to, and the teams that are able to drive change with AI and with agents, you know, what are they doing right?

他们做对了什么？

Like, what are they getting right?

他们是怎么解决那些问题，真正取得业务成果的？

How are they able to, solve some of those problems and really get to business outcomes?

而且，如果你看到过一些真正令人印象深刻的业务成果，你有没有什么例子可以分享？

And, you know, if there are business outcomes that you've seen that are really impressive, you know, some of the, you know, any examples that you have to?

是的，我见过一些令人印象深刻的业务成果，比如你自动化了公司里的采购到付款流程，然后想办法释放了5000万美元的现金。

Yeah, I've seen business outcomes that are impressive, such as you automate some procure to pay process in your company and you find a way to leverage like 50,000,000 of cash.

你自动化了银行的后台流程，每天能处理的交易量就提升了4%。

You automate some backend process in a bank and you get like 4% increase in terms of your number of transactions you can process by day and so forth.

通过智能体，你确实能看到这些相当可观的业务成果。我不是在说那些简单的用例，比如客服之类的，而是在那些深度的用例中，你确实开始看到巨大的提升。

You see those business outcomes that are actually fairly impressive through agentic, and yeah, I'm not talking about, yeah, easy use cases such as, I don't know, customer support or whatsoever, but like in those deep use cases, indeed you start seeing these big uplifts.

我觉得共同点在于，你需要懂得如何避免被AI分散注意力。

And I think the commonality together is that you need to understand how to avoid the distraction of AI.

我觉得我看到的共同点是，你如何在组织里设置好，让你清楚地知道，选一个数字，一个、三个、五个、十个，如果你是大公司的话，那些你真正、真正、真正想掌握的关键用例，并且你明白为什么有战略上的理由去做这些。

And I think that's a commonality I see, which is how do you set it up in your organization so that you have a clear understanding of the, pick a number, one, three, five, ten, maybe if you're a big company, of key use cases you really, really, really want to own, and where you understand why there is a strategic reason to do it.

这不是分心，不只是为了学习，而是你真的需要去做。

It's not a distraction, it's not just for learning, it's like you really need to do it.

而且你要放下一个事实，就是你的组织里会有5100、500、1000个可能值得追求的AI相关想法，但你不希望它们变成干扰。

And put aside the fact that you will have in your organization 5,100, 500, 1,000 ideas associated to AI that could be worth pursuing, but that you don't want them to become a distraction.

所以你需要这种两层的方法，一方面，你在业务中给予很多自由，让人们能够创新，能够访问数据，发现所有这些用例，追求这些想法，在团队层面获得自己的生产力提升；但与此同时，你对那些真正改变业务的大项目有清晰的目标，并且可以从中心推动这些项目。

And so you need to have this two tier approach, which is like you give lots of freedom in your business so that people can innovate on one sense, that they can get access to the data, discover all of those use cases, pursue those ideas, get their own level of productivity at the team level, but at the same time you've got clarity on the big milestone you want to hit on big projects that are really transforming your business and that you can drive from the center.

我认为这种双运营模式，在这一点上，是企业目前避免分心、真正收获AI好处的一个关键。

I think that this this very dual operating model on that perspective is I think a key for enterprise at this moment to avoid the distraction and actually reap the benefit of AI.

很好。

Very good.

你知道，我当然是做讲故事这一行的，所以我和CEO们也都得会讲故事。

So, you know, of course, I'm in the storytelling business, and so one of the things I and and CEOs have to kind of be storytellers, you know, as well.

当你讲这个故事的时候，比如说，嘿，我们可以来帮你。

When you're telling the story about, hey, we can come in and help you.

我们可以帮你推动这个转型。

We can help you drive that transformation.

我们可以帮你实现这些业务成果。

We can help you get those business outcomes.

那你跟客户、跟其他领导讲的时候，是怎么让他们知道你能帮他们实现一些大目标的呢？

You know, what's the story that you tell to customers, to other leaders, to let them know that you could help them with some of their big goals?

我觉得我通常会先告诉他们，其实他们自己已经有解决方案了。

I think I want to, I usually start by actually telling them that they actually have the solution.

他们脑子里已经有解决方案了，就算自己没想到，团队里可能也有人知道该怎么解决。

They have the solution in their head, if not in their head, probably have people in their team that actually know the solution.

而且以现在的技术状况，他们其实很快就能把东西做出来，这是非常现实的。

And that with the state of technology, it's actually very realistic for them to actually build things quickly.

所以他们可能会低估自己在某些方面需要外部帮助的程度，也低估了自己团队或自身的能力。

And so they might underestimate their need to get external help of a certain nature and underestimate the quality of their own team or themselves.

这是第一点。

That's one.

第二点，我承认，要想规模化或者大规模做AI，他们不能太随意，不能以为只要给每个人无限的云资源，三个月后公司里一切都会变得美好顺利。

Second, I acknowledge that in order to scale or do anything AI at scale, they can't be cowboy and just imagine that by having unlimited cloud tokens to everyone, everything will be rose and work better in the company after three months.

我承认，如果他们那样做，很可能会造成相当大的混乱，大家都会搞不清楚业务到底在发生什么。

I acknowledge that if they do that, it'd probably be quite a bit of chaos and that everyone will go crazy in terms of not understanding what's happening in the business anymore.

这是什么？

What is this?

哦，就像是我自己瞎编出来的一个新应用，提供假数据，然后我们所有的业务决策都基于它。

Oh, it's just like some new app I hallucinated that is providing fake data and we make all of our business decisions based on it.

所以我告诉他们，他们需要做的是信任自己的业务，有一个清晰的策略，尤其是在他们需要推进的大项目上，赋能团队去做，但同时也要借助一些能力来控制数据层面、流程层面以及整个智能体组合的情况，确保他们持续朝着正确的方向前进，而这正是我们想要提供的。

And so I tell them what they want to do, I need to do is trust their business, have a clear approach, especially on this big project they need to move, empower their team to do so, but indeed leverage some ability to control what's happening on the data level, on the process level, on on the overall agentic portfolio to make sure that they are continuously going in the right direction, and that's what we want to provide.

所以如果他们能信任自己，他们就能和我们合作。

So if they can trust themselves, they can work with us.

如果他们觉得需要全面的控制，他们也可以和我们合作。

If they believe that control is required, like holistically, they can work with us.

如果他们两者都不认同，那也许我们真的不太合适。

If they don't believe either of that, well, maybe we're actually not a good fit.

有道理。

Makes sense.

好的。

Okay.

那招聘方面呢，Florian？

How about, you know, hiring, Florian?

你知道，即使在过去一年里，AI领域也发生了很多变化。

You know, there's a lot has changed even in the past year, you know, in the in the AI space.

这些工具能做的事情多了很多。

There's a lot more that these tools can do.

现在团队能做的事情比以前多多了，以前可能需要实习生或者初级员工来做的事情，现在AI也能做了。

There's a lot more that that, you know, teams can do that they might have needed, like an an intern or an entry level, employee to do.

我们一次又一次地看到，AI确实能完成一些这样的任务。

And we're we're seeing this, you know, again and again that AI can do some of those kind of tasks.

所以初级岗位的招聘，你知道，就遇到了这样的瓶颈。

And so entry level hiring, you know, has hit this kind of speed bump.

你们招聘的时候看重什么？

What are you looking for when you all hire?

你们会招初级员工吗？

Are you all hiring entry level folks?

更广泛地说，在招聘方面，你们寻找人才的标准有什么变化？

In in hiring in general, more broadly as well, how has what you look for when you hire, you know, changed?

还有，你想对那些想加入你们公司的人说些什么？

And, you know, what what would you say to to people out there that, you know, might wanna work for your company, right?

他们可以做些什么来准备自己，让自己脱颖而出？

What what can they do to prepare themselves and to stand out?

我觉得我在招聘时可能过于看重三样东西，以前我也看重这些，但现在可能没那么看重了。

I think that I probably over index when hiring on, meaning three things before, that I was like indexing on before but maybe less so.

就是某种好奇心、某种判断力和某种动力。

A certain type of curiosity, a certain type of judgment and a certain type of motivation.

动力这块很简单，因为AI太复杂、变化太快了，你需要那些对AI充满热情、极度自我驱动的人。

So the type of motivation is easy, which is like AI is so complicated and so ever changing that you want people that are passionate about it and let's say self motivated about AI to an extreme.

为什么？

Why?

因为你进入的是一个无法保证每次都能顺利发展的领域。

Because you're entering a world where you can't really guarantee that everything will go every time, like up to the right.

AI本质上就是这样。

By design, it's AI.

事情会变，市场会崩，市场会被重新定义。

Things will change, you will have market collapse, you will have redefinition of the market.

所以，你是不是那种比一般人自我驱动力强得多的人？

So are you someone that is more, like say, way more self motivated than the average?

关于好奇心这个问题，当然每个人都希望员工有好奇心、愿意学习，但现在这已经深入到很根本的层面了，就是说，你有没有自己动手用AI编过东西？

The question of curiosity, meaning of course everyone wants people that are curious and wants to learn in their business, but now it's like very down to the bottom of it actually, which is like, have you vibe coded something?

你最近一次自己动手创造的东西是什么？

What was the last thing you actually created by yourself?

如果你没有自己动手创造过任何东西，哪怕你是做销售的，那也挺奇怪的。

You have not, even if you're like a salesperson, if you have not created something by yourself, it's kind of strange.

如果你自己都不动手做事，那你为什么想来AI公司呢？

Why do you want for an AI company if you're not doing things by yourself?

为什么？

Why?

所以我觉得，在我们每天都被那些会彻底改变世界的技术包围的情况下，这种对技术的好奇心和接受度，如果你没有这种好奇心，对我来说就是个问题。

So I think that this level of curiosity of adoption of technology in the context where we are living every hour with technologies that will change completely the world, like if you don't have this level of curiosity, it's a problem for me.

最后一点是关于判断力，这可能是最难测试的，但这里的重点是，比以往任何时候都更重要的不是你能多快打字，多快组装IDE，而是你能不能有品味上的判断力，知道什么重要什么不重要，什么看起来真实什么不真实，以及你如何思考问题。

And the last one is the topic of judgment, which is probably harder to actually test for, but here the idea is that even than before it's not about how fast can you type, how fast can you even assemble ID, it's whether you can have some judgment of taste in terms of what's important versus not, what looks real versus not, and so forth, in how you think about problems.

你在这里要解决的是，你不想招进来更多的人，他们只会生成五页AI垃圾内容，然后发给所有人，还觉得这样没问题。

What you want to solve for here is you don't want to get into your business more people that will just generate five pages of AI slop and send it to everyone believing that it's okay.

这其实很难监管，所以你需要招那种人，不管发生什么，哪怕你最后拿到五页从云端或ChatGPT生成的内容，也要把它精简成一页，专注于真正重要的东西，并且能分辨出什么是真正有价值的，什么不是。

This is actually very hard to police, and so you need to bring people that will be just like, whatever happens, even if you end up having like five pages from the cloud or chat GPT, get it to one and focus on what matters and understand what is love versus not.

所以这三样东西——判断力、动力和好奇心——在我看来，现在比两年前更重要了。

And so those three things like judgments, motivation and curiosity are, from my perspective, even more important now than two years ago.

非常好。

Very good.

你们公司还在招初级员工吗？

Are you still hiring entry level folks into your company?

是的，我们在招。

Yeah, we do.

我们在某些特定职位上会招。

We do in select positions.

我觉得，我的粗略猜测是，现在的市场环境决定了有很多人进入职场，尤其是工程领域，他们其实很有才华，但找工作却更难了。

Think that there is a My rough guess here is you're in a market where by design you have lots of people entering your career, in particular in engineering, that have actually lots of talent, enough harder time to actually find a job.

所以也许我有点愤世嫉俗，但我认为这仍然是在各个层级招人的理由。

So I'm maybe a bit cynical here, but I think that's a case for hiring at every level still.

非常好。

Very good.

而且你们也是全球性的公司，对吧？

And you all are global too, right?

比如，你们在哪些地方有办公室，团队都分布在哪儿？

Like, what are the places that you have offices and where, you know, do you have teams located?

你们的主要办公地点有哪些？

What are some of your kind of your your main locations?

是的。

Yeah.

主要地点包括东京、首尔、新加坡、迪拜、马德里、巴黎、伦敦、阿姆斯特丹、柏林和纽约。

Main location include Tokyo, Seoul, Singapore, Dubai, Madrid, Paris, London, Amsterdam, Berlin, and New York.

所以在Dataiku工作，确实可以在很多很棒的城市工作。

So yeah, working at Dataiku, can actually work in a variety of great cities.

我不知道哪个是我最喜欢的，但你觉得这些城市里哪个最好？

I don't know which one is my favorite, but which one is your favorite on those cities?

比如，你有最喜欢的城市吗？

Like, do you have a favorite city?

这些城市里我有最喜欢的吗？

Do I have a favorite of those cities?

我得说是伦敦。

London, I would have to say.

大概伦敦和纽约吧。

London and New York, probably.

这有点普通。

That's that's a little bit, boring.

这些城市我大部分都去过，不过，你知道，伦敦是个很人性化的城市，至少跟美国城市比是这样。

I've been to most of those cities, but, but, yeah, you know, London is a London's a very humane city, compared to The US cities anyway.

对。

So Yeah.

是的。

Yeah.

你呢？

How about you?

你花时间最多的是哪些城市？

What are what are some of the ones that, you spend your most time?

我的意思是，我也超爱伦敦。

Meaning I love London too.

我的意思是，那就是我的最爱。

Meaning that's mine.

对。

Yeah.

我觉得自己挺像伦敦人的，虽然你可能从我的口音里听不出来。

I feel like, I feel quite a bit like a Londoner, which, you could not tell by my accent, I guess.

当然。

For sure.

当然。

For sure.

这家公司是在巴黎创立的，对吧？

The company started in Paris, right?

我没记错吧？

Am I remembering that correctly?

对。

Yeah.

是的。

Yes.

没错。

Indeed.

很好。

Very good.

很好。

Very good.

嗯，Florian，我喜欢做的一件事就是，在节目最后，对所有嘉宾都问同样两个问题。

Well, one of the things that I like to do, Florian, is sort of end with the same two questions with, with all of our our guests.

第一个问题是，你知道，AI曾经有个承诺，说AI会帮我们做很多苦力活，就是那些我们不想做的事情。

And the first is, you know, there was this promise in AI that AI was gonna do sort of the a lot of the the grunt work for us, like the things we didn't wanna do.

然后我们就能把更多时间花在更高层次的任务上。

And then we'll be able to focus more of our time on, you know, the higher level bit kind of tasks.

也许这是真的，但不管怎样，我看到现在大多数领导者比以往任何时候都更缺时间。

And maybe that's true, but whatever, the case is, what I see from most leaders today is they're more stretched for time than ever.

他们节奏更快。

They're moving faster.

他们专注于如何最大化利用自己的时间。

They're focused on how can they get the maximum leverage for their time.

所以在我们结束之前，我的第一个问题是，你最好的建议是什么？

And so my first question as we wrap up is, you know, what's your best tip?

你如何最大化利用时间，对其他想做到这一点的领导者，你最好的建议是什么？

How do you get maximum leverage for your time and what's your best tip, you know, for other leaders who are trying to do the same?

我最好的建议之一就是，对，用AI来让自己变得更聪明，避免陷入认知懒惰。

My my my So one of my best tips for using AI is that, yeah, use it in a way that makes yourself smart in order to not get into cognitive laziness.

在过去一年的某个时候，我发现自己开始变得认知懒惰了。

At some point in the last year, I discovered myself that I was getting into cognitive laziness.

我发现自己还没怎么读书或者思考，就直接去问系统问题了。

I was like not reading as much or not thinking as much before asking the question to the system.

确实，我变得越来越依赖数字，然后我意识到，你必须真正保留自己独立思考的能力，能够连续思考几分钟，最好是几十分钟，而不是动不动就问系统，这样才能充分利用它。因为归根结底，长时间思考并产生想法这件事，AI目前还不太擅长，所以你需要保持自己的优势，而这需要你保持专注。

And indeed, I was becoming number on number, And I realized that you need to actually keep for yourself this ability to actually think for like multiple minutes, and ideally multiple dozens of minutes, without asking the system to make sure that you get the best of it, because ultimately the ability to think for a long time and to come up with some ideas is what anyway AI is not really good for yet, So where you need to keep an edge, and for that you need to keep your concentration.

我觉得这是一个建议。

I think that's one tip.

非常好，我喜欢这个。

Very good, I love that.

有意思，我其实有一个关于认知衰退的故事，很快就要讲到了。哦，是的。

It's funny, and I actually have a story that's coming up soon on cognitive atrophy and how Oh, yeah.

你知道，我们必须非常刻意地决定用AI做什么，以及我们愿意放弃什么。

You know, have to be, really intentional about what we use AI for and what we're okay sort of getting away from.

对吧？

Right?