Box CEO on AI Agents & Why Enterprise Can't Keep Up | a16z

频道: a16z
视频: https://www.youtube.com/watch?v=dvVbA9OcBqs
原文语言: en
统计: 共 60 轮 · Aaron Levie 30 · Martin Casado 30

[0:00] Aaron Levie

So, the board goes to the CEO. What does the board say? We need more AI. And what does the CEO say? Oh, okay. I'll get like a consultant to do more AI. And then they have some centralized project that nobody knows how it works. They haven't aligned their operations, and those things will fail.

于是董事会去找 CEO。董事会怎么说？我们得多搞点 AI。CEO 怎么回？哦，好的，那我请个咨询顾问来多搞点 AI。然后他们就立了个集中式的项目，没人搞得懂它到底怎么运作。他们根本没把自己的业务运营理顺，这种项目注定要失败。

[0:12] Aaron Levie

The funniest concept that the more code we write, the less we would need engineers. It'd be the opposite because now your systems are even more complex than before, which means that you're going to be running into even more challenges of when you need to do a system upgrade or when there's downtime and you have to figure out like what Well, how do I fix that problem? Or when there's a security incident. I mean, we're just getting started with the jobs on this front. They're going to hit a wall at integration. And And this The thing that's not different about AI and that agents don't fix, that nothing fixes, is that any enterprise of a thousand people or more or that's older than 10 years is just a massive stuff that's sitting there waiting to be integrated. And And you can't just say it's going to integrate. AI actually doesn't help to integrate anything. Hey, we are here moderating the situation live, and we're very excited to talk about a bunch of AI stuff. And we have a three of us are here today. Uh there's me, Steven Sinofsky, and Martin Casado, who will wave and say hi, I'm Martin, and

最搞笑的一个说法是：我们写的代码越多，需要的工程师就越少。其实恰恰相反，因为现在你的系统比以前更复杂了，这意味着等你要做系统升级、或者出了宕机故障、得搞清楚『这问题到底怎么修』，又或者出了安全事故的时候，你会撞上更多的麻烦。我是说，这方面的活儿我们才刚开始干呢。他们会在集成（integration）这一关撞墙。AI 没法改变、agent 也解决不了、什么都解决不了的一点是：任何超过一千人、或者成立超过十年的 enterprise，都堆着一大坨东西在那儿等着被集成。你不能光嘴上说一句『它会自动集成的』。AI 其实根本帮不上集成的忙。嘿，我们正在现场主持这场对话，非常高兴能聊一大堆 AI 的话题。今天我们三个人在这儿，有我、Steven Sinofsky，还有 Martin Casado——他会挥挥手说一声『嗨，我是 Martin』。

[1:22] Aaron Levie

Martin. and Aaron Levie, who is is working on the elevation of his hair today. So, we're excited about that.

Martin。还有 Aaron Levie，他今天的工作重点是让自己的头发往上立。所以我们也挺期待这个的。

[1:29] Aaron Levie

keeps getting more vertical, and uh I thought I could kind of tame it, but it didn't work.

它越来越往上竖，我本来想把它压一压，结果没压住。

[1:33] Martin Casado

And is that just a token issue or a parameter number of parameters issue with your hair?

那这到底是你头发的 token 问题，还是参数数量太多的问题？

[1:39] Aaron Levie

Too many parameters.

参数太多了。

[1:40] Martin Casado

Okay, I have the same thing, but in reverse. Okay, so You Hey, listen, you have a distilled model. There you go. My I run local. So, we had a lot of There's been a busy week of things, but we're we want to bubble it up a a bit and just start talking about where where things are are heading. Um but I'll I'll let I just kick it to you, Aaron, and you start where you are the most excited this moment because you have visited a ton of customers this week and have learned a lot. You've shared a lot on X. But I think you're the most in the trenches CEO who is really talking to customers every single day in the enterprise, which is what the three of us tend to look at the most. Yeah, I think my it feels like my job these days is just bring reality to the valley and then bring the valley to reality as as much as possible. And it's a it is a kind of a crazy divide that that that exists at the moment. Um you know, the past couple weeks

好吧，我也是一样的毛病，不过是反过来的。好，听我说，你这是个蒸馏过的模型，这就对了。我嘛，是本地跑的。这周事情特别多，我们想把它稍微往上拔一拔，开始聊聊大方向往哪儿走。我先把话筒交给你，Aaron，你从此刻你最兴奋的点讲起，因为你这周拜访了一大堆客户，学到了不少东西，也在 X 上分享了很多。我觉得你是最扎在一线的 CEO，每天都真真切切地在跟 enterprise 里的客户聊，而这恰恰是我们三个最关注的。是啊，我感觉我现在的工作就是尽可能地把现实带给硅谷，再把硅谷带回到现实里。当下这中间存在一种近乎疯狂的鸿沟。你知道，过去这几周……

[2:39] Aaron Levie

take it back. I actually think it's a super interesting. What what is it? What's the gap caused by? The gap is caused by Yeah, well, I think the gap is and Martin, I'm sure you see this, but I think the gap is caused by the styles of work that exist in Silicon Valley and in engineering roles versus sort of the rest of the world. So and we've talked about this a couple times in in different forms, but but you know, the the technical aptitude of an engineer is just like insanely high. The level of wired-in-ness to what's going on on the internet is insanely high. The the ability to use your own tools and make your own choices is insanely high. And when things go wrong with the systems that you choose, you can just like quickly debug them and then make them sort of work for you. And then obviously you have all the benefits of just the models are really good at code and and the work is verifiable. So you have like you know, five or 10 things that make agents work in an enterprise context for engineering or at least a even a startup context for engineering that that tend to be a that there tends to be a gulf between the way you work that way in engineering and the rest of of sort of knowledge work. And so and so a lot of what I see is trying to figure out how do we kind of you know, bottle up all of the greatness that is, you know, what we are seeing from coding agents, what we're seeing from agents that use computers um to how do you bring that into the enterprise where the workflows are are, you know, quite different. The users are less technical. The data is much more fragmented. The systems are much more legacy. And so that that tends to be the divide. So it's not even that we're like talking past each other like in a in a one of those kind of classic like

收回我刚才的话。我其实觉得这特别有意思。这鸿沟是什么？它是怎么造成的？这鸿沟是由……对，我觉得这鸿沟是——Martin，我相信你也看得到——是由硅谷和工程岗位上那种工作方式，跟世界上其他地方的工作方式之间的差异造成的。我们之前用不同的方式聊过几次这个，但你知道，一个工程师的技术素养是高到离谱的；他们对互联网上正在发生的事情那种『线上即时连接感』也高到离谱；自己挑工具、自己做选择的能力同样高到离谱。而当你选的系统出了岔子，你能很快把它 debug 掉，再把它调成顺手的样子。再加上很明显模型在写代码这件事上确实很强，而且工作成果是可验证的。所以你看，在 enterprise 里做工程、或者哪怕只是创业公司里做工程，有那么五到十个因素能让 agent 跑得通；而你在工程里的这种工作方式，跟其余那些知识工作之间，往往隔着一道鸿沟。所以我看到的很多事情，就是在琢磨：我们怎么才能把 coding agent、把那些会操作电脑的 agent 身上看到的所有好东西，统统装进瓶子里，再把它带进 enterprise——而那里的 workflow 截然不同，用户没那么懂技术，数据更加碎片化，系统也老旧得多。所以那才是真正的分界线。所以这甚至都不是我们在『各说各话』的那种经典……

[4:09] Aaron Levie

government versus industry. It's it's just literally like there is just a pure workflow and and and technology set divide. And and that's why it's going to be, you know, a number of years for this sort of diffusion to to roll from what we're seeing in Silicon Valley, what we're seeing as tech startups all around the world into the rest of knowledge work. Martin, just to build on that, you have a ton of experience in big companies. I One of the other issues though is scale. Yeah. And the way the difference in scale that Silicon Valley operates at at the startup level versus everyone else. Yeah, I I also think that I mean, these secular trends like the internet was Like does actually start with individuals? And big companies tend to make decisions centrally. And this is one of the fastest growing secular trend. So like there's probably a lot of individuals in big companies that are doing it where like Yes. the big companies themselves don't know even how to think about it. And so when you hear stats like oh like MIT had this stat like 95% of AI efforts in big companies fail. Like that's clearly silly because I am sure everybody's using chat GPT very effectively. What what they really should be saying is you know, whatever. Like I Listen, I sit on these boards too. So the board goes to the CEO. What does the board say? We need more AI. And what does the CEO say? Oh, okay. I'll get like a consultant to do more AI. And then they have some centralized project that nobody knows how it works. They haven't aligned their operations and those things will fail. And so I don't, you know, when we say scale, often we we think about things like system scale or number of people. I think the secular trend is scaling wonderfully, which is being reflected in

……政府对产业界那种对立。它纯粹就是 workflow 和技术栈层面实打实的分裂。这也是为什么这种扩散要花上好些年，才能从我们在硅谷看到的、在全世界科技创业公司里看到的东西，慢慢蔓延到其余的知识工作里去。Martin，接着这个话题说，你在大公司有海量的经验。我觉得另一个问题是规模。对。硅谷在创业公司层面运作的规模，跟其他所有人之间存在的那种规模差异。是啊，我也觉得，我是说，像互联网这种长周期趋势，其实真的是从个人开始的；而大公司往往是集中决策的。这又是增长最快的长周期趋势之一。所以大公司里很可能有一大堆个人已经在用了，只不过公司本身根本不知道该怎么去看待它。所以当你听到那种数据，比如说『MIT 有个统计，大公司里 95% 的 AI 项目都失败了』，那显然很傻，因为我敢肯定每个人都在非常高效地用 ChatGPT。他们真正该说的是——管它呢。听着，我也在这些董事会里坐着。董事会去找 CEO，董事会怎么说？我们得多搞点 AI。CEO 怎么回？哦，好的，那我请个咨询顾问来多搞点 AI。然后他们就立了个集中式项目，没人搞得懂它怎么运作，业务运营也没理顺，这种项目注定要失败。所以我觉得，当我们说『规模』时，往往想的是系统规模、人数规模这类东西。我觉得这个长周期趋势本身是在漂亮地放大的，这一点反映在……

[5:47] Martin Casado

the numbers of these companies, but organizations don't know how to adjust these kind of, you know, agile processes that have been, you know, worked on for a decade around, you know, data and governance and operations and compliance, etc. That's kind of right now where I think like Aaron is right between the secular trend and the the organizational decision body. And this is something that we actually track very closely because we're starting to see now, I would say in the last few months, finally some real kind of inroads into the enterprise, but it's it's it's tepid because and and the last thing I'll say at this, one of the reasons is there's a lot of skepticism because the board wants AI, CEO AI failures have created some amount of bruising, which is, you know, you know, requiring these companies get past it in order to do kind of the second go at it. And so I think this is exactly where we are. Yeah, I I 100% agree with that, which is that it's good to start with agreements because we we know how quickly that fade uh Because we'll disagree the rest of the show, exactly.

……这些公司的数字上；但组织却不知道该怎么去调整那些围绕数据、治理、运营、合规等等、被打磨了十来年的敏捷流程。这差不多就是眼下的状况——我觉得 Aaron 正好卡在长周期趋势和组织决策机构这两者之间。这也是我们盯得特别紧的一件事，因为我们现在开始看到——我得说就是最近这几个月——终于有了一些真正打进 enterprise 的进展，但它还很温吞。原因之一，也是我最后想说的一点，是存在大量的怀疑情绪：董事会想要 AI，而 CEO 们搞 AI 失败过，留下了一些挫伤，这就要求这些公司得先迈过这道坎，才能开始第二轮尝试。所以我觉得这正是我们当下所处的位置。对，我百分之百同意，所以从达成共识开始挺好的，因为我们都知道共识消退得有多快。因为接下来整场节目我们都会各执己见，没错。

[6:48] Aaron Levie

exactly. That's the only time we're going to agree. Um I I think maybe one more point on the board for agreements, maybe maybe you guys would agree. Um there's also this very interesting um dynamic. I I'd say this is a minor one relative to everything else is probably a 5% of the problem. I I think it'd be more fun to talk about the the real problem, but but there is a fun kind of as an aside, there's a fun dynamic where, you know, you go to an engineering team classically for the past, you know, and you know, Steven, you can take us back in in history on this one. And one of like the easiest ways to stall a project was just getting the architecture, you know, kind of the the fights on, you know, what language to use, what architecture path to go down. That could take months and months to kind of work through as your teams work through that. Um because of the pace of change in AI, um you actually have this incredible dynamic where the the labs uh you know, are are obviously leapfrogging each other so frequently, but with with not the exact same paradigm of how you should deploy agents and how they will work and is the is the is the agent harness in the computer is it outside the computer do you run it in your cloud is it hosted what tools does it have access to like we are like this is not a a point where these are completely fungible technologies and so that actually creates a bit of paralysis because now as an enterprise architecture team in the real world you're like man like what what horse do I want to you know kind of get behind and and which architecture path do I want to get behind because I've been burned by doing the wrong thing in AI maybe three or four years ago and I went down in some path that now is deprecated or not the right strategy anymore. So so

没错，那将是我们唯一一次意见一致的时候。关于董事会，我也许还想再补一个能达成共识的点，也许你们也会同意。还有一个特别有意思的现象，我得说相对于其他所有事情，这是个次要因素，大概只占问题的 5%。我觉得聊真正的大问题会更有意思，不过作为题外话，有个挺好玩的现象——你知道，过去这些年你去找一个工程团队，Steven，这段历史你可以带我们回顾回顾——拖垮一个项目最简单的办法之一，就是卡在架构上，就是为『用什么语言、走哪条架构路线』吵起来。这种争论可能要花上好几个月才能在团队里捋清楚。而由于 AI 变化的速度，你现在碰到一个不可思议的现象：那些实验室显然在频繁地互相反超，但它们用的并不是同一套范式——agent 该怎么 deploy、它们会怎么工作、agent 的执行环境（harness）是在电脑里还是电脑外、你是在自己的云里跑它还是用托管的、它能访问哪些工具……我们现在远没到这些技术可以完全互相替换的地步。所以这反而造成了某种瘫痪，因为现在作为现实世界里的一个 enterprise 架构团队，你会想：天哪，我到底该押哪匹马、该站哪条架构路线？因为我可能三四年前在 AI 上做错过选择、被坑过，走了一条现在已经被废弃、或者不再是正确策略的路。所以……

[8:23] Aaron Levie

to some extent the speed of our change in in tech actually reduces the ability for the tech to get diffused into the really really important workflows because now you have a lot of paralysis in in just making decisions. So so I actually think it's kind of fine because there's still so much upgrade work people need to do in their infrastructure and their systems and their data but this is kind of an interesting dynamic where I'll I'll go have conversations with CIOs and their AI teams and I'll say hey what what are you using for your chat system or your you know core agent orchestration and they'll say yeah we're in the middle of a debate between these two or three paradigms and it's and and you and you hear that across almost every single customer because there is a little bit of a nervousness of like who do you get in bed with and and how how much do you sort of you know fully lock yourself into one particular path and we also know that that if you don't lock yourself into a path it's always then then you're building for the sort of duality which is you know also takes a lot of work architecturally. I actually I sorry I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I

……某种程度上，我们科技变化的速度反而削弱了技术扩散进那些真正至关重要的 workflow 的能力，因为现在光是做决策这件事就让人陷入了瘫痪。所以我其实觉得这没什么大不了，因为大家在基础设施、系统和数据上还有那么多升级的活儿要干。不过这是个挺有意思的现象——我会去跟一些 CIO 和他们的 AI 团队聊，我会问：嘿，你们的聊天系统、或者说核心的 agent 编排（orchestration）用的是什么？他们会说：是啊，我们正在这两三种范式之间争论不休。几乎每一个客户那儿你都能听到这种说法，因为大家都有点紧张：到底该跟谁绑定、该在多大程度上把自己彻底锁死在某一条路上。而我们也知道，如果你不把自己锁死在一条路上，那你就总得为某种『两头兼顾』去开发，这在架构上同样要花很多力气。我其实……抱歉……

[9:29] Martin Casado

I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I like the you know and so it's kind of like this fusion or this hybrid model. What we're seeing instead is instead of viewing AI as software, Yes.

……就像那种你知道的……所以它有点像是这种融合，或者说混合模式。而我们看到的另一种做法是：与其把 AI 当成软件——对。

[9:50] Martin Casado

like just view it as a user. And so, instead like take your product, make it a CLI tool, and then have the AI be an agent that actually uses it. So, you're not fusing the two, you're just making it more useful for AI. This is a very very significant architectural and mental shift, right? And so, we started as pure product, and then we didn't quite know what the end thing looked like. So, we created this you know this you know AI software hybrid that hasn't worked. And now we're kind of going to the agentic model, which basically means the agent is going to be whatever. It's going to be cloud code or whatever, and then my product now just should be something that can consumed by that. And like that's the actual modality. But, you know, within a year now you've had to re-architect your software twice. And so, I think no matter many places that you look in the industry is having this dilemma of actually trying to figure out what the final form looks like. And Stephen, you will remember Remember all the hybrid versions of cloud? Oh, yeah. Remember like you know like remote desktop and all these things? Like I think we're kind of like speed running that evolution to the final form. Right. And I And I think that that people in Silicon Valley don't quite appreciate when a big company says, "Well, we have to map out our bet that we're going to make." Yes.

……不如就把它当成一个用户。所以与其那样，不如把你的产品做成一个 CLI 工具，然后让 AI 当那个真正去使用它的 agent。这样你就不是把两者硬融在一起，而是单纯让它对 AI 更好用。这是一次非常非常重大的架构和思维方式上的转变，对吧？所以我们一开始是纯产品，然后我们也不太清楚最终形态长什么样，于是就搞了个 AI 与软件的混合体，结果它没跑通。现在我们差不多要转向 agentic 模式，基本意思就是：那个 agent 可以是任何东西，可以是 Claude Code 之类的，而我的产品现在就应该只是个能被它消费的东西。这才是真正的形态。但你知道吗，就在一年之内，你已经被迫把自己的软件重新架构了两次。所以我觉得，无论你看行业里哪个地方，大家都在经历这种两难，都在拼命搞清楚最终形态到底长什么样。Steven，你会记得云计算那些混合版本吧？哦，对。还记得远程桌面那一套东西吗？我觉得我们差不多是在『速通』那段演化，直奔最终形态。对。我还觉得，硅谷的人不太能体会到，当一家大公司说『我们得把自己要押的赌注规划清楚』时是什么意思。对。

[11:04] Martin Casado

Because like that just seems stupid. And you know, if you have if your job history is you know, five two-year stints at startups that went from seed to series A to acquire or something. Yeah, you didn't learn anything.

因为那在他们看来就显得很蠢。你知道，如果你的履历是在创业公司里待过五段、每段两年，从种子轮到 A 轮再到被收购之类的，那你其实啥也没学到。

[11:18] Aaron Levie

Well, you you never you you don't your frame of reference is not you know, picking an accounts payable system that's going to last 40 years.

对，你永远不会……你的参照系里压根没有『挑一套要用上四十年的应付账款系统』这种事。

[11:27] Aaron Levie

Yeah, I I actually I have like all these visual aids today. So, here's like the ultimate the ultimate engineer if you're in Silicon Valley is Yeah, exactly. Is Gilfoyle. And and Gilfoyle is like I I don't want to talk to anyone. Yes. And I I will just write the code and you go do your thing. And the the thing is is that you you have people in in enterprises that are saying, "I'm going to use the model and do my thing." But they're only they're going to hit a wall at integration. And and this the thing that's not different about AI and that agents don't fix, that nothing fix, is that any enterprise of a thousand people or more or that's older than 10 years is just a massive stuff that's sitting there waiting to be integrated. And and you can't just say it's going to integrate. AI actually doesn't help to integrate anything. And so even if you change everything the the people say, "Oh no, if you make it an agent, then it can just go ahead and and and be a user." But if you're a user, like if you've ever called customer service for something, like literally you get bounced to another human if the system that you're talking to Doesn't have to. doesn't work. And they're like, "Well, that's a manager." Or no, you're you're talking about payment, not reservations. And and so like we're we're what I think is so exciting is that now we have proof of this technology that everybody likes it. I mean, you see all the people who don't like AI are saying, "Look at what's happening in law firms because people are seeing hallucinations and it's ruining legal cases and all this." And and the reason that's happening is because the 25-year-old associate is the one using AI successfully already and have been using it for a year. Well, Steve it's actually Steve it's

对，我今天其实准备了一堆视觉道具。所以你看，在硅谷，那个终极工程师的典型是——对，没错——是 Gilfoyle（《硅谷》里的角色）。Gilfoyle 那种人就是：我不想跟任何人说话，对。我就负责写代码，你去忙你的。问题在于，enterprise 里有些人也在说『我要用模型、干我自己的活儿』，但他们只会在集成这一关撞墙。AI 没什么不同、agent 解决不了、什么都解决不了的一点是：任何超过一千人、或者成立超过十年的 enterprise，都堆着一大坨东西在那儿等着被集成。你不能光嘴上说一句『它会自动集成的』。AI 其实根本帮不上集成的忙。所以哪怕你把所有东西都换掉，人们会说『哦不要紧，只要你把它做成一个 agent，它就能自己上去当个用户』。可如果你是个用户——比方说你给什么东西打过客服电话——你会发现，要是你正在对话的那个系统搞不定，你就会被实实在在地转给另一个人。它不一定能搞定。他们会说『哦那得找经理』，或者『不对，你说的是付款，不是预订』。所以我觉得特别让人兴奋的是，现在我们已经有了这项技术的实证，大家都喜欢它。我是说，你看那些不喜欢 AI 的人会说『看看律所里发生了什么，因为大家看到了幻觉，毁掉了一些官司之类的』。而之所以会这样，是因为那个 25 岁的助理早就已经在成功地用 AI 了，而且已经用了一年。嗯，Steve，其实——Steve，其实……

[13:21] Martin Casado

actually a little worse than that where it is right now many companies are incentivizing people to use AI by counting tokens.

……现实比那还要糟一点，因为眼下很多公司在激励员工用 AI 的方式，是去数 token 消耗量。

[13:28] Aaron Levie

Oh, yeah, yeah, yeah.

哦，对对对。

[13:30] Martin Casado

And so I'm not going to say the name of the guy I I spoke to someone yesterday who works for one of these large companies that famously does this, and he's like, "Me and my coworkers have agents do useless tasks just so that we can I'm no joke in No, no, no, totally. Well, I you get whatever you measure, so Yeah, yeah. That's right. So like it's like the extreme form of what you're saying, Stephen. You have people that are like being fake productive and producing a lot of, you know, Yeah, you you could say perhaps problematic artifacts just because they're they're using these models.

所以我就不点那个人的名字了——我昨天跟一个在某家以这么干而出名的大公司工作的人聊过，他说：『我和我同事让 agent 去干一些毫无用处的活儿，纯粹就是为了……』我说真的没开玩笑……（Aaron：不不不，完全懂。）你考核什么就会得到什么，所以——是啊是啊，没错。所以这就像是你说的那种现象的极端版本，Steven。你会有一批人在『假装高产』，搞出一大堆，你知道——是啊，你甚至可以说是有点麻烦的产物，纯粹因为他们在用这些模型。

[14:03] Aaron Levie

The when when the internet happened, all of a sudden every company needed websites. And so like a very famous moment in time was not too long ago when every internal team had like a team website. And they went out and they got like a vendor to write HTML and to create their site. And then there was a team. But like there's nothing dumber than having a team website at a large company because a team gets reorganized like 6 months later. And so companies were just filled with like the with thousands of these dead web is what what the expression was. But I I think Go ahead, go ahead. No, no, I but we should we should drill in your integration point because I do think this is something for you know, sort of some reality to settle in in in the valley on on the real world's sort of journey to fully being agentified and what that's going to take and what that's going to look like. And and your your point about being passed to the different human, you know, based on the role that you needed to interact with, you know, agents basically don't have any There's no real exception yet for the agent having the same problem because you basically, you know, as you pass through different human, it's it's a different set of access controls that that that human has. And if an agent can sort of bypass any of those steps, then then that's how you instantly get the security risks that like you you need to kind of pass through those steps so that way you don't accidentally, you know, get to the wrong piece of information and there's verification. And so, there's a lot that that you need to kind of build out for agents to be able to go and and work with all these systems. And and we've talked about this, but like most legacy environments don't have the most authoritative, you know, access

当年互联网兴起的时候，突然之间每家公司都需要网站。所以前不久有个很出名的时刻：每个内部团队都搞了个团队网站。他们跑出去找个供应商来写 HTML、给他们做站。然后就有一支团队。但在大公司里，没有比『搞一个团队网站』更蠢的事了，因为这个团队六个月后可能就被重组了。于是公司里就堆满了成千上万个这种『死掉的网页』——当时的说法就是这个。不过我觉得——你说你说。不不，我是想说我们应该深入聊聊你那个集成的点，因为我确实觉得这是个能让硅谷对现实世界的认知落地的东西——现实世界要走向完全 agent 化的那段旅程，需要付出什么、会长成什么样。还有你刚才那个点，关于根据你需要打交道的角色被转给不同的人——agent 基本上没有任何……目前还没有真正的机制能让 agent 不掉进同样的坑里，因为当你被层层转给不同的人时，每个人手上其实是一套不同的访问控制权限（access controls）。而如果一个 agent 能绕过其中任何一步，那你瞬间就会引入安全风险——你必须得走完那些步骤，这样你才不会一不小心拿到错误的信息，而且中间还有验证环节。所以要让 agent 能去跟所有这些系统打交道、把活儿干了，你还有很多东西要搭建起来。我们也聊过这个，但你看，大多数 legacy 环境根本就没有最权威的那套访问……

[15:40] Aaron Levie

controls. So, you're always as a human going and saying, "Hey Sally, can you share that thing with me that that I don't have access to?" Or, "Hey Bob, what's the number inside your data system for this question?" And so, if agents just get the exact same permissions that you had, then they'll just run into these walls everywhere and they won't be able to complete the process. And unlike a human, they're not going to know to go talk to Sally or ask the question of Bob. So, they're going to just be kind of, you know, stuck. So, what's going to happen is you're going to have a lot of agents that don't have access to the right data. Um they're they're kind of working through systems that that are, you know, not not the real sources of truth for the information. They're getting the wrong number. They're getting the wrong document. So, this is the real work that enterprises have to go through right now. The good news is that that it's actually a great time, again, if you're a startup because you can just you get to know all the problems right at the right out of the gate. So, you can design your organizations, you know, to try and avoid this. But for big companies, there's real work that goes into how do I upgrade my systems? How do I modernize my technology environment? How do I make sure that, you know, agents will have access to the right data, the right documents, the right context to be able to do their work? And that's sort of the the the work ahead. And and there's, you know, I I There was this uh you know, kind of headline of of OpenAI working with in Codex, you know, working with Accenture, Deloitte, all all the major system integrators. And there were some kind of like, you know, snarky comments online around it that that I was

权限控制。所以作为人，你总是会去问："嘿 Sally，能不能把那个我没权限看的东西分享给我？"或者"嘿 Bob，你那套数据系统里这个问题对应的数字是多少？"那么，如果 agent 拿到的权限跟你完全一样，它们就会到处碰壁，根本没法把整个流程跑完。而且跟人不一样，它们不会想到要去找 Sally 聊，或者去问 Bob。所以它们就会卡在那儿，懂吧。结果就是，你会有一大堆 agent 拿不到正确的数据，它们其实是在一些并不是真正权威信息源的系统里瞎转悠，拿到的是错的数字、错的文档。这就是 enterprise 现在真正要去啃的硬骨头。好消息是，如果你是创业公司，这反而是个绝佳的时机，因为你从一开始就把所有问题都摸清了，可以把组织设计成尽量避开这些坑。但对大公司来说，要做的功课是实打实的：我怎么升级我的系统？我怎么把技术环境现代化？我怎么保证 agent 能拿到正确的数据、正确的文档、正确的上下文来干活？这就是接下来要做的事。还有，你知道，之前有个新闻，OpenAI 用 Codex 跟 Accenture、Deloitte 这些大型系统集成商合作，网上有些挺阴阳怪气的评论，让我特别

[17:04] Aaron Levie

fascinated by because it sort of showed how how maybe, you know, great that divide is from the rest of the world versus those in tech because to me it was like the most obvious announcement of all time, which is a large enterprise is going to have to go through that the change management, the systems implementation, the integration of technology for these agents to be able to go and work. And so, there was this sort of like, you know, people thought it was somewhat ironic that, "Oh, we need people to implement the agents that are going to go automate the people." And it's like, "No, that's exactly how it works." You you actually do need to do lots and lots of work to be able to be in a position where agents can actually go and and help you do, you know, any of the automation. So so that is and that's going to be there's going to be businesses that are doing that for decades. Like, it's going to be an incredible opportunity for this kind of next generation set of firms as well as existing ones that that lean into that. Let me throw this out there. Well, first, I think the other thing that people shouldn't celebrate when those fail and because they will fail because they're as Martin was describing, they're get a lot of them are going to be these sort of top-down mandates where they picked like the most acute problem in the company and think, "Oh, AI is going to go solve that." And the IT people are going to be like, "Oh, God, that's the worst That's that's that's the worst system to try to do that." But the CEO or CFO or whatever is going to be obsessed with solving or the most likely the customer service person will be obsessed. But but I do think if I were if I were advising a startup specifically in order to to sort of enter the enterprise space in

觉得有意思，因为这恰恰显示出，搞技术的人和其他所有人之间的鸿沟有多大。因为对我来说，这是史上最显而易见的一个公告——大企业必然要经历变更管理、系统实施、技术集成这一整套流程，agent 才能真正上岗干活。但当时有人觉得这事儿挺讽刺的："哦，我们居然得请人来部署那些要去把人自动化掉的 agent。"我说不对，事情本来就是这么运作的。你确实得做海量的工作，才能走到一个让 agent 真正能帮你完成任何自动化的位置。所以这件事——会有公司专门做这个，一做就是几十年。这对下一代的服务公司，以及那些愿意拥抱这件事的现有公司来说，都是个不得了的机会。我抛个观点出来。先说一点，我觉得另外一件大家不该幸灾乐祸的事是：当这些项目失败的时候——它们一定会失败，因为正如 Martin 描述的，很多都是那种自上而下的强制指令，他们挑了公司里最尖锐的那个问题，然后想"哦，AI 能去把它解决掉"。结果 IT 那帮人会想"我的天，这是最糟糕的——这是最最糟糕的系统去硬上这个的对象"。但 CEO 或 CFO 之类的人就会执着于解决它，最可能是客服那边的人会特别执着。不过我确实觉得，如果是我——如果是我去给一家创业公司出主意，专门切入 enterprise 这块市场的话

[18:31] Martin Casado

that way, definitely would be thinking about not just like building a company that step one, I only work with all the headless SaaS software that's out there because there just won't be any. Like the but the the thing you can do is structure the value that you offer. And also, this applies to what you go do in a company. Is it's really a fork. And the fork is is this an agent that is seeking information and presenting it to to some human? Or is this an agent that's supposed to go act and do something? Like, is it is it acquiring or is it doing? Because if if it's it turns out that's how what happened with the internet. The internet got very very valuable when the first step was just providing access to things to people. Yes. And and like all of a sudden all the sites that were like that literally did integration. Like, "Hey, I need expense reports but viewed by department or I need to see our current inventory status across like the two companies we've acquired." All of a sudden the web became the integration point. And so I do think that that if you just show a person and just say, "Hey, we can actually use agents to learn stuff about what's going on in a company." And in particular, because you're here, Aaron, like learning across the files becomes way more possible than it ever was before. In fact, AI might be the first time that inside a company search can actually provide immediate value. Yes.

用那种方式切，我肯定会这么想：别只想着去建一家公司，第一步就是"我只跟市面上所有的 headless SaaS 软件打交道"，因为根本就不会有那么多 headless 软件。你真正能做的，是把你提供的价值结构化。这一点同样适用于你进到一家公司里要去做的事。这其实是个岔路口。岔路口在于：这个 agent 是去寻找信息、然后呈现给某个人看的？还是这个 agent 应该去采取行动、去做某件事？也就是说，它是在"获取"还是在"执行"？因为如果——结果证明，互联网当年就是这么发展起来的。互联网变得极其有价值，是从它第一步只是给人们提供访问各种东西的能力开始的。是的。然后突然之间，所有那些网站真的就成了集成层，比如"嘿，我需要按部门拆分的报销报表"，或者"我需要看我们收购的那两家公司合在一起的当前库存状态"。突然之间，网页就变成了那个集成的接入点。所以我确实觉得，如果你只是给一个人看，跟他说"嘿，我们其实可以用 agent 来摸清公司里正在发生的各种事情"。尤其是——正因为你在这儿，Aaron——跨文件去学习这件事，变得比以往任何时候都更可行了。事实上，AI 可能是头一回让公司内部搜索真正能立刻产生价值的时候。是的。

[19:59] Martin Casado

web just wasn't structured to deliver those results. And and then you start to think, once you can bring them all together, then you can add like an agent that has an approved button or a reject button or something like that. Let me Let me Let me just try and provide Finally the point where I get to disagree. So let Uh-oh, we're in trouble now. No, no, no, no, you can get invited back. You're invited back. So, good.

网页当年就是没有按那种方式来组织、好交付这些结果的。然后你开始想，一旦你能把它们全都汇聚到一起，你就可以再加上一个带"批准"按钮、"拒绝"按钮之类的 agent。让我——让我——我先试着提供——终于到了我可以唱反调的环节了。糟糕，这下我们有麻烦了。不不不不，你还能被请回来的，你还会被邀请回来的。好，那就好。

[20:22] Aaron Levie

No, no, I I think this is a very legit view, but it's not the only view. And in light of AI, it's I think it's not the only kind of compelling view. So, here's the other So, the Let me just try and rephrase. So, the current view is we've got like AI is software. It it works in a different way. Um we have a current set of systems and we have to integrate this new type of software with our existing systems so that it can get access to data. It can do things, but in a safe way, right? So, here's the the kind of end-to-end argument of why this isn't about about evolving software systems. The end-to-end argument is these LLMs are non-deterministic. They are smart. They deal with the long tail of complexity. And it turns out those are all things humans do, too. And we've spent 40-years building interfaces, processes, and design to deal with messy humans. And, you know, we know who to access, and we have access control. And so, if you have the mindset that an agent is more like a human, and you hire the agent, you give it its own email address, it can access documents like humans can, it can log in, it can request the things that it needs, then it will be drafting on all of the process that that we've put in place for humans, not for software. And so, I would just encourage us as we have this discussion, like listen, I grew up like you guys in software. I always think of every system like software, but these models don't integrate well with software. Actually, I think it turns out, and what we're learning as an industry is if you view them more like humans, and you draft on the mechanisms we put in place for humans, they're much easier to integrate. Well, that's And I THINK THERE'S I THINK WE ALL I think we agree with that for sure. I think the issue is

不不，我觉得这是个非常站得住脚的观点，但它不是唯一的观点。而且在 AI 的背景下，我觉得它不是唯一一个有说服力的看法。所以这里有另一个——我先试着重新表述一下。当前的主流看法是：AI 就是软件，它以一种不同的方式运作。我们有一套现有的系统，我们得把这种新型软件跟我们已有的系统集成起来，让它能拿到数据、能干活，但要以一种安全的方式，对吧？那么这里有个"端到端"的论证，说明为什么这件事其实不是关于演进软件系统的。端到端的论证是：这些 LLM 是非确定性的，它们很聪明，它们能应对那条长尾的复杂性。而结果发现，这些恰恰也是人类在做的事。我们花了四十年去构建界面、流程和设计，就为了应对乱糟糟的人类。我们知道该找谁、我们有访问控制。所以如果你抱着"agent 更像是一个人"的心态去看待它——你雇了这个 agent，给它一个自己的邮箱地址，它能像人一样访问文档，能登录，能去申请它需要的东西——那么它就会顺势搭上我们为人类（而不是为软件）建立起来的那一整套流程的便车。所以我只想鼓励大家，在我们讨论这事的时候——听着，我跟你们一样是搞软件出身的，我总是把每个系统都当成软件来想，但这些模型跟软件集成得并不好。我其实觉得，结果证明、也是我们作为一个行业正在学到的是：如果你把它们更当成人来看，顺势用上我们为人类建立的那些机制，它们反而会容易集成得多。嗯，这——我觉得这——我想我们都——我觉得我们对这点绝对是认同的。我觉得问题在于

[22:02] Martin Casado

humans have a bunch of extra benefits that the agent doesn't have. The human has a lot of context that it gets for that they get that we get for free by virtue of we can keep track of the myriad relationships that we've built in our organization, and the person to tap on the shoulder when when we need something done, or we need to get information. That's not documented in a company yet in a in a way that the agent can just sort of draft on. And so, so I I like I I mean, I think we all would agree that that you have you can't treat this like software. You treat these as as people accessing systems and tools, but they are at a they're both at an a massive advantage that they can work in parallel in at you know, at infinite scale, and they're at a disadvantage in that they don't know who to tap on the shoulder.

人类有一堆 agent 没有的额外优势。人有大量的上下文是天生免费获得的：因为我们能记住自己在组织里建立起来的那一大堆错综复杂的关系，知道在需要把某件事办成、或者需要拿到某个信息时该去拍谁的肩膀。这些东西在公司里目前还没有被记录下来，至少没有以一种 agent 能直接顺势利用的方式记录下来。所以——我的意思是，我觉得我们都会同意，你不能把这玩意儿当软件来对待，你要把它们当成访问系统和工具的人。但它们同时——它们既处在一个巨大的优势上，能并行工作、几乎能无限扩展规模；又处在一个劣势上，就是它们不知道该去拍谁的肩膀。

[22:40] Martin Casado

Hey, I listen, Aaron, I am all for agent onboarding. Like, you know, the agent comes, and it goes to orientation, and then the CEO gives it the culture discussion, and then every I'M NOT KIDDING. NO, YOU'RE PROBABLY RIGHT. I MEAN, THAT'S

嘿，听着，Aaron，我完全支持给 agent 做入职。比如说，agent 来了，先去参加入职培训，然后 CEO 给它讲一通公司文化，接着每——我没开玩笑。不，你说得大概没错，我是说，那确实——

[22:53] Martin Casado

EVERY DEPARTMENT EVERY DEPARTMENT does their pitch, like this is what we do. And like, I mean, I think I I actually honestly think given given the technical nature of these agents

每个部门——每个部门都做一遍他们的介绍，比如"这是我们做的事"。我是说，我觉得我——我其实真心认为，鉴于这些 agent 的技术特性——

[23:03] Aaron Levie

Yeah. and how much entropy they have and kind of how unruly they are, we're going to have to go through the processes that we've refined around humans. Yeah, 100%.

对。以及它们身上有多少熵、有多么不受控，我们将不得不沿用我们围绕人类打磨出来的那些流程。对，百分之百。

[23:13] Martin Casado

Because humans have all of those things. And so I just, you know, it's more about providing schools for them than somehow building some, you know, fancy index database. No, no, totally totally agree. That I mean, of course, what I love about about that is you just keep going with the analogy because what that is is it's the same argument that humanoid robots will will be the best kind of robot, which is we have a whole world designed for humans. And I like I I saw at the Consumer Electronics Show, I saw this robot go into an elevator and then there was a a button-pushing robot on the elevator. So, because the the robot was a tiny little thing that like a Roomba on the floor, it couldn't push the button. So, the the same company that invented that robot invented a device you buy for the elevator that pushes the button. But then I asked, why did you need a device to push the button? And it And it was very interesting. They said, because the elevators don't have systems that they can hook into as a a robot. Like there's no Wi-Fi press the button in the elevator capability. Yeah. There's no API for that. There's no there there is no headless version of the elevator. Yeah. And And I think that's That's actually a great metaphor for like the problem that I think that we're actually solving in the enterprise with these agents, which is we just, you know, we have two types of systems, those for humans and those for software. And these tend to be more like humans, so we should draft on those as much as possible rather than try to retrofit them Well, and so so the big news last week was uh was I I I think, you know, Salesforce I don't know if they surprised people or not, but but I mean, based on the reaction, it seemed like it was maybe a surprise. They they went

因为人类把这些东西全都具备了。所以我只是觉得，这件事更多是关于给它们提供"学校"，而不是去搭建什么花里胡哨的索引数据库。不不，完全——完全同意。我是说，当然，我特别喜欢的一点是你把这个类比一路推下去了，因为这其实跟"人形机器人会是最好的那种机器人"是同一个论证：我们有一整个为人类设计的世界。我——我在消费电子展（CES）上看到，有个机器人走进电梯，电梯里还有另一个专门按按钮的机器人。因为那个机器人是个小不点儿，像地上的扫地机器人（Roomba），它够不着按按钮。所以发明那个机器人的同一家公司，又发明了一个你买来装在电梯里、专门帮忙按按钮的装置。但我就问，你们为什么需要一个装置去按按钮？这点很有意思。他们说，因为电梯没有可以让机器人接入的系统，电梯里没有"用 Wi-Fi 按按钮"这种能力。对。没有那个 API。没有——压根就没有 headless 版本的电梯。对。我觉得这——这其实是个绝妙的比喻，正好对应我们现在在 enterprise 里用这些 agent 实际要解决的问题：我们就是有两类系统，一类是给人用的，一类是给软件用的。而这些 agent 更像人，所以我们应该尽量顺势搭人类那一套的便车，而不是想方设法去硬改它们。嗯，所以——所以上周的大新闻是——我觉得，Salesforce，我不知道这事儿有没有让大家意外，但从反应来看，似乎确实有点出乎意料。他们走了——

[24:51] Aaron Levie

full headless and they basically said, you know, like we want to be used everywhere across all of our all of the different agents. Um and I see that as a a little bit of a bellwether because I think as Salesforce goes, so does a lot of of of enterprise software. And I think a lot of people are going to try and you know, have to figure out what is the new business model in this headless world. You know, do you do you charge a little bit of a a small just API tax? Is there a seat for the agent? So there's obviously some work to do with that. And Stephen, I I saw one of your tweets on you know, some of the some of the you know, complications there. But but but but I think as a as a moment, it's a big deal because because I think it's a recognition that that, you know, software will be running in the background. It always has for machine users and applications. And now it is for these sort of probabilistic machine users and or non-deterministic machine users. And what's cool and and where I think this gets pretty exciting is, you know, as soon as I saw that announcement, like I had like five to 10 personal use cases where I would need, you know, the headless version of Salesforce because I'm always doing just a tremendous amount of of customer related intelligence work. I'm going into a meeting, I need some information. I need to do I'm going into a city, who should I be meeting with? And so if you imagine, you know, being able to run compute in the form of agents across all of your data systems, like the use cases become, you know, pretty wild around what that opens up. So I think this gives I think this gives a lot of software platforms all new use cases that they can tap into that where again, you were normally constrained by the number of people on these platforms, but

彻底 headless 化，基本上是在说：我们想让我们所有的——让我们各种各样的 agent 在各处都能被用上。我把这看作一个有点风向标意义的事件，因为我觉得 Salesforce 怎么走，很大一部分 enterprise 软件就会跟着怎么走。我觉得很多人都得去琢磨：在这个 headless 的世界里，新的商业模式到底是什么？你是收一点点小小的 API 税？还是给 agent 设一个席位（seat）？所以这里头显然还有些活儿要干。Stephen，我看到你有条推文讲到了其中一些复杂之处。但我觉得作为一个标志性时刻，这是件大事，因为我觉得这是一种承认：软件将会在后台运行——它一直都是这样为机器用户和应用程序服务的，而现在它要为这些概率型的机器用户、或者说非确定性的机器用户服务了。而很酷的、我觉得让这件事变得相当令人兴奋的地方在于：我一看到那个公告，脑子里马上就冒出五到十个我个人的用例，我会需要一个 headless 版本的 Salesforce，因为我总是在做海量的客户相关情报工作。我要去开个会，我需要一些信息；我要去某个城市，我该跟谁见面？所以你想象一下，如果你能以 agent 的形式，在你所有的数据系统上跑算力，那能打开的用例就变得相当疯狂了。所以我觉得这给了——我觉得这给了很多软件平台一大批全新的用例可以挖掘，而过去你通常是被这些平台上的人数给卡死的，但是——

[26:20] Aaron Levie

now the headless user can be, you know, 100 or 1,000 x the scale of those human users. So this is a I think an exciting moment because as you have more of these agents running around and the headless software modes, you just have, you know, way more use cases for these tools. I also think I just I think on this one, what's so what's so super cool is is that of course, the first step is doing exactly what you described, which is just looking stuff up. And so the the most interesting thing is using this notion that an agent is just a an entity, it's it's incredibly obvious to me that it's another license. Now, it might have a different license model, but it actually it has to have an identity. Like when you go look something up in in the box um CRM system, I don't know if you're in Salesforce or not. When you use the box CRM system, it has to be a person. Like with a certain amount of access rights. And you presumably a CEO, you might have access to a bunch of stuff, but also there's a lot of ways that they actually don't want you to have the right rights at the right time. Like you might be able to look and see who is on the account, but you don't need the up-to-date quota of those sales people and stuff, and that might be HR sensitive, and you should probably have some other level to go see that. But as you go down the org, the agent is never going to have more permissions than the person who's getting it to go do something. And in fact, it's just going to be like a peer to somebody else in an organization. Because otherwise, you have all of these issues where the peer where a human can just say, "Oh, get me the super smart agent that knows everything that I'm not allowed to know." And to to the in the mod in the IT architecture sense, what's

现在 headless 用户可以达到那些人类用户一百倍、一千倍的规模。所以我觉得这是个令人兴奋的时刻，因为随着越来越多这样的 agent 到处跑，加上软件的 headless 模式，你就有了多得多的用例去用这些工具。我还觉得——我就觉得在这一点上，特别特别酷的地方在于：当然，第一步就是做你刚描述的那件事，也就是单纯地查东西。所以最有意思的是，用"agent 只是一个实体"这个观念来看——对我来说极其显而易见的是，它就是又一份 license（许可）。它的 license 模式也许不一样，但它必须得有一个身份。比如你去 box 的 CRM 系统里查个东西——我不知道你们用不用 Salesforce——当你用 box 的 CRM 系统时，它必须是一个"人"，带着一定量的访问权限。你作为 CEO，大概能访问一大堆东西，但也有很多情况下，他们其实并不希望你在某些时刻拥有某些权限。比如你也许能看到账户上有哪些人，但你不需要看到这些销售人员最新的业绩配额之类的，那可能涉及 HR 敏感信息，你大概得有另外一个层级才能去看。但顺着组织往下走，agent 永远不会拥有比那个指使它去干活的人更多的权限。事实上，它就会像组织里另一个人的同侪一样。因为否则的话，你就会碰到这一堆问题——那个同侪——一个人可以直接说"哦，给我弄一个无所不知的超级聪明 agent，把我本不被允许知道的东西都告诉我"。而从 IT 架构的意义上说，最——

[27:57] Martin Casado

so fascinating about that is you have to build you can't let the agent get the results and then try to figure out what works or not. But first of all, the points that Martin made about about about the the LLM stochastic model, which is you're not going to be able to figure out. It's not like a record in a SQL table that you could just apply ACLs to. It's it's actually like it could be words in a sentence or just the number that shows up. And so I actually think it that whole discussion about headless for me made the SASpocalypse seem even dumber than it was already, and it was already dumb. So like it was like at first it was dumb, and then I'm like, "Oh my god, it's actually much dumber than I thought it was in the first place." Because you're just going to have this explosion. Now, someone might come up with a very clever pricing scheme, and that agents, you know, somehow cost less because maybe for the first 5 years they're read-only, or they they're always tied to a person or something, but that it is another seat. There is no way around it. And like if you're a SaaS company, you're crazy to try to say, "Oh, just use the credentials of another human." Like that's just that would be like bad security practice from the get-go. Exactly. So actually in in fact, um so this is playing out in many domains. You can even make the argument that like a a a headless SaaS doesn't make sense. And and here's the argument. The argument is Well, let me give you an example. So, um if you use OpenClaw, do you know why you use a Mac mini with OpenClaw? It's number one for iMessage. So right, right. So it's for the integration. Yeah, yeah, yeah. Because there is no headless version, so you're just going to like use it. And then the second one is very interesting, which is

最让人着迷的一点是：你必须从一开始就把它建对——你不能让 agent 先把结果拿到手，然后再回头去琢磨哪些能用哪些不能用。但首先，Martin 刚才讲的那个关于 LLM 是随机性模型的点——也就是你根本没法事先算清楚。它不像 SQL 表里的一条记录，你可以直接套个 ACL（访问控制列表）上去。它实际上可能是一句话里的某些词，或者只是冒出来的一个数字。所以我其实觉得，整个关于 headless 的讨论，在我看来让那个所谓的"SaaS 末日论（SaaSpocalypse）"显得比它本来就有的还要更蠢——它本来就已经够蠢了。所以就好像，一开始它很蠢，然后我心想"我的天，它实际上比我一开始以为的还要蠢得多"。因为你将会迎来一场大爆炸。当然，可能有人会想出一套很聪明的定价方案，让 agent 不知怎么地成本更低——也许头五年它们是只读的，或者它们总是被绑定到某个人身上之类的——但它就是又一个席位（seat），这事儿没有任何办法绕过去。而且如果你是一家 SaaS 公司，你要是想说"哦，那就用另一个人的凭据吧"，那你就是疯了，那从一开始就是糟糕的安全实践。一点没错。所以其实——事实上，呃，这件事正在很多领域上演。你甚至可以论证说，headless SaaS 本身就讲不通。论证是这样的——让我给你举个例子。呃，如果你用 OpenClaw，你知道为什么你得配一台 Mac mini 来跑 OpenClaw 吗？第一个原因就是为了 iMessage。对对。所以这就是为了集成。对对对。因为它没有 headless 版本，所以你就只能这么用。然后第二个原因很有意思——

[29:37] Martin Casado

if you've tried to use headless browsers with agents, the problem is is all of the websites um have anti-scraping measures. So they don't work. And so the reason you use a Mac mini so it can actually use Safari proper. So to do anything headless kind of assumes that like the entire internet is going to go headless, when I think all of these models, like all of the data is humans working on the actual apps that are not headless. Like that's all of the data anyway. So I think these models are going to be very good at just using apps like they are today, and we're already seeing this happen, and rather than the headless versions, the non-headless versions are what's actually being used. So you could argue that it's just Salesforce. Like not headless. Like It will go to a Wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait, wait. Do you literally mean the agent goes to the browser?

如果你试过让 agent 用 headless 浏览器，问题就在于，所有的网站都有反爬措施，所以它们没法用。这就是为什么你得用一台 Mac mini，好让它能真正调用正经的 Safari。所以任何"headless 化"的设想，其实都假设了整个互联网都会走向 headless，可我觉得所有这些模型——所有的数据其实都是人在那些非 headless 的真实应用上操作产生的，那才是全部的数据来源。所以我觉得这些模型会非常擅长就像今天这样直接用那些应用，而且我们已经在看到这种情况发生了——而不是用 headless 版本，正在被实际使用的恰恰是非 headless 的版本。所以你可以争辩说，那就只是 Salesforce 本身，不是 headless 的。它会去——等等等等等等等等等等等等等等等等等等等等等等。你的意思真的是 agent 跑去用浏览器？

[30:26] Aaron Levie

Yes. Oh, no, no, no. I'm I'm taking the other side on that one big time. Yeah, yeah. No, but but but but but but but let me just let let let let me let me simplify the argument so we can actually have it. So So today if you use an agent like NanoClaw or OpenClaw, you could use a headless browser. Let's say I wanted to like look up the value of my house on Zillow. The headless browser simply doesn't work because Zillow's so tired of people scraping it, so it will detect headless browsers. Totally.

对。哦不不不，这一条我可要旗帜鲜明地站到对立面去了。对对。不过——让我把这个论点简化一下，这样我们才好真正展开来辩。那么今天，如果你用 NanoClaw 或者 OpenClaw 这样的 agent，你本可以用一个 headless 浏览器。比如说我想在 Zillow 上查一下我房子的估值，headless 浏览器根本就不管用，因为 Zillow 早就受够了被人爬数据，所以它会检测 headless 浏览器。完全没错。

[30:51] Aaron Levie

So, the thing that works is it uses you pop up Safari and it uses a proper Safari directly, right? And so then then all of the sudden it works. And so

所以真正管用的办法是：它弹出一个 Safari，直接用正经的 Safari，对吧？然后突然之间它就好使了。所以——

[31:00] Martin Casado

But no, but but but but I think I think I mean I would just say that that that any software that has a good API, the agent would absolutely prefer to use the API. And then you and then you pop into the browser the moment that you run into some execution problem with I mean you know, set it set set as a a fantastic long-term computer science software guy. However, these models are trained on data and RL environments from existing software that didn't have those APIs. Yeah.

不过——我觉得，我想我只想说，任何一个有好用 API 的软件，agent 绝对会更愿意去用那个 API。然后你——然后你只在碰到某种执行问题、跑不下去的那一刻，才弹进浏览器里去。我是说，作为一个一流的、做了一辈子计算机科学软件的老兵这么讲是没错。然而，这些模型是在那些根本没有这些 API 的现有软件所产生的数据和 RL 环境上训练出来的。对。

[31:29] Aaron Levie

And and and and right now if you actually look at the adoption and the use of these agents, they look far more like what a human would do than what like a program would do. So, maybe you're right, but A, that's not what we're seeing. And you can honestly make the end-to-end argument when it comes to data and all of the controls in the internet. To Steven's point, all of the existing controls to just be like these are going to actually have the same actions as humans. Well, the APIs most that that I mean I think the I mean the APIs of any software provider will follow the same access controls of of whatever the whatever the user is that that is

而且现在你要是真去看这些 agent 的采用情况和实际用法，它们的行为方式更像人类会做的事，而不像一个程序会做的事。所以也许你说得对，但首先，我们现在看到的并不是这样。说到数据，说到互联网上那一整套管控机制，你完全可以提出端到端的论点。就像 Steven 说的那样，所有现有的管控机制都会认为：这些 agent 本质上会执行和人类一样的操作。那么 API 嘛——我是说，任何软件供应商的 API 都会遵循同样的访问控制，取决于背后是哪个用户在操作。

[32:06] Martin Casado

Right, but they have to rebuild they have to they have to rebuild it. I mean it's like it's like you've got this existing app and all the the models trained on all of the people using the app. Well, well, on on that point and it's a totally fair point, but I would I would guess over time you're going to have really uh you're going to have very you know, kind of accurate, rigorous data sets for you know, for models to be trained against the MCPs of of every SaaS platform, the APIs of every SaaS platform. Already it's in the you know, they're already training against all of our documentation on our on our products and our APIs. But but I I just think to me it's it's more of just an inefficiency of of navigating through pixels versus just you know, you can just you can just do it. Yeah. And added in systems Aaron is that is that layers never go away. They just get layered. Well well well well so so I but on that I'll I'll support your point 50%. Oh well, there you go. Yeah yeah yeah yeah. I I just I JUST THINK I JUST THINK IF YOU NEED TO DO A search for a document, our search API is going to be a faster way to do it than you know, you know, clicking through a an interface. Right, but but the but the to have support the point like the new code X the computer use that on the desktop is is you know, just insane and I I mean Steven obviously knows you know, everything about how it would work. I well and I saw my ability to move a mouse and then this other sort of mouse moving and clicking things. I was like I don't understand computers anymore.

对，但他们得重建，他们必须得重建这套东西。就好比说，你有一个现成的 app，模型是基于所有用这个 app 的人训练出来的。嗯，关于这一点——这个观点完全成立——但我猜随着时间推移，你会有非常精准、非常严谨的数据集，让模型可以针对每个 SaaS 平台的 MCP、每个 SaaS 平台的 API 去训练。其实现在就已经在做了，他们已经在针对我们产品和 API 的所有文档做训练了。但我只是觉得，对我来说，这更多只是一种低效——你要在像素界面里点来点去，而其实你可以直接搞定。对。Aaron，还要补充一点：层（layer）永远不会消失，它们只会一层叠一层。嗯，这一点上我支持你 50%。哦，这就对了。对对对对。我就是觉得，如果你需要搜一份文档，我们的搜索 API 会比你在界面上点来点去更快。对，但是为了支持这个观点——比如桌面端那个新的 Codex、那个 computer use，真的太离谱了。Steven 显然对它的运作原理了如指掌。我看着我自己在移动鼠标，然后另一个鼠标也在移动、在点东西，我当时就想：我已经看不懂电脑了。

[33:35] Aaron Levie

Right right. So so there is a pretty and and to your point Martin, my first instinct was to use it for something where I know there's no available API. So I did actually use it right away for a thing that I don't have access to the API and and agent over time is going to probably have to figure out is there an easy MCP or CLI for this action and if not, then I'm going to pop into some kind of cloud browser or cloud computer or maybe local thing that I can you know, sort of parallel track and then and then go and execute that. So that does seem like a reasonable architecture. Um but but I I still think that like I'm going to I'm going to pound the Salesforce API massively in in headless mode just cuz that'll be an efficient way to go look up records. Yeah, I mean it I I think that your your sort of I think you're both saying the same thing, but there's just a time but no, but there's a time dimension. And I think like there was a moment at the internet that that I really was thinking about when when you guys were when I was seeing the time scale difference, what which was suddenly the 8 million quadrillion pages of how to use Word and Excel that we had written over the years that we posted on the internet, we had used to ship them with the product. And people would have them on their hard drive, not connected to anything, and they would say like, "How do I make an ice chart?" or whatever, and it never worked. They could never find the thing that they wanted. But what happened with the internet was the net result of everybody finding it caused us to make better documentation, but it also caused Google search to be better at finding the information that it needed, which then completely changed the way that we thought about doing documentation.

对对。所以确实存在一个挺……Martin，就像你说的，我的第一反应是把它用在那种我知道没有现成 API 的场景。所以我确实马上就拿它去做了一件没有 API 可用的事。agent 长远来看大概得学会判断：这个操作有没有现成的 MCP 或 CLI，如果没有，那我就切到某种云端浏览器、云端电脑、或者本地的东西里去，开个并行线程然后去执行。所以这个架构听起来挺合理的。但我还是觉得，我会在无头（headless）模式下疯狂调用 Salesforce API，因为那才是查记录的高效方式。是的，我觉得你俩说的其实是同一件事，只是有个时间维度的差别。我想到互联网早期有那么一个时刻，当时我看到那种时间尺度上的差异，突然意识到：这些年我们写过的、贴到互联网上的、关于怎么用 Word 和 Excel 的几百万乃至天文数字级别的网页——我们以前是把它们随产品一起发货的。人们会把它们存在硬盘上，不联网，然后他们会问『我怎么做一张冰柱图？』之类的，结果从来都搞不定。他们永远找不到自己想要的那个东西。但互联网带来的改变是：大家都能找到信息了，这个净结果反过来促使我们把文档做得更好，同时也让 Google 搜索更擅长找到它需要的信息，而这又彻底改变了我们对待做文档这件事的思路。

[35:12] Martin Casado

And I think that with headless, especially for the kind that's just finding things, it's going to really change the the way that information is exposed. And so the way that Salesforce sees today of exposing it a headless API is I'm almost certain if I were to go look at it, it's going to look like the developer API in front of a behind a CLI, and it's going to look a lot like that. But in fact, that's not at all how humans using Salesforce interact as a human trying to solve like, "I'm standing in the elevator waiting to go see a customer. What is the stuff I need to know?" Like that that mapping is completely different. And so the that API is going to really change as a result over time. Yes, I I I think I think the API changes for sure. I I agree that, but I do think that unlike the humanoid, you know, kind of comparison, where where sort of the physical world has interesting physics issues that that you eventually run into, the digital world doesn't. And so so at some point your agent can run in parallel, you know, 500 times, and like I'm going into I want to do a market map of of customers across the Fortune 500, that agent can can fan out and do that work in a way that I can't as a person in a browser. Right. So so so to some extent agents get to let you, you know, sort of bend the laws of of normal, you know, human-based workflows. Um and and so then to like that's why that and I think that that means the APIs maybe eventually evolve, but not obviously in the direction of the end user product, but but maybe more toward an agentic sort of set of workflows of what is that agent looking to do?

我觉得有了 headless，尤其是那种纯粹用来找东西的场景，它会真正改变信息被暴露出来的方式。所以 Salesforce 今天那种通过 headless API 暴露信息的方式——我几乎可以肯定，要是我去看一眼，它一定长得像套了一层 CLI 的开发者 API，差不多就那个样子。但事实上，那根本不是人在用 Salesforce 时的交互方式。一个人想解决的是：『我正站在电梯里准备去见客户，我现在需要知道哪些信息？』这种映射关系完全不一样。所以随着时间推移，那个 API 一定会发生很大变化。是的，我认为 API 肯定会变，这点我同意。但我确实觉得，不像人形机器人那种类比——物理世界有各种有意思的物理限制，你迟早会撞上——数字世界是没有这些限制的。所以到某个时候，你的 agent 可以并行跑 500 次，比如『我想给财富 500 强的客户做一张市场地图』，这个 agent 可以扇出（fan out）去完成这项工作，而我作为一个坐在浏览器前的人是做不到的。对。所以从某种程度上说，agent 让你能够弯折那些基于人类的正常 workflow 的法则。所以这也是为什么——我觉得这意味着 API 也许最终会演化，但显然不是朝着面向终端用户的产品那个方向演化，而更可能是朝着一套 agent 式的 workflow 去演化：那个 agent 到底想做什么？

[36:53] Aaron Levie

Well, but Martin, I think we jump in and just say, "Wait, you didn't describe anything new. You described an architectural No, but you described an architectural problem with today's software, which is it's it's API and performance gate was based on how much I can type." Which which is sort of the point I was making, which was our help system was designed on how much we could ship on one CD and had no data about what it is that people were trying to do and no context. But it didn't change like the problem, which is I needed to make a chart. Yes. Yeah, exactly. The Well, so like one one real example of this, we we launched a Box agent that gives, you know, that has, you know, much more capabilities built into it and one of the capabilities is that it searches across your whole Box environment, but it doesn't it doesn't have the same limitations of a human-based search where you you type in one query, you get back a set of results, you look through them, you know, it fans out, does multiple queries, it can look through hundreds of results instantly and do its own re-ranking of that. And so that's just like, you know, again, you wouldn't want to be rate limited by the same a process that a human went through. Which is which is where the humanoid robot is, you know, you're kind of willing to be like, "Okay, the humanoid is still going to walk into the elevator, it's still going to press the button." When actually, you know, in a agent world, you're like, "No, no, I just want you to go and instantly press the the floor that I'm going to." Yeah. Yeah, but we should just be very clear like by the way, I I very much agree, but we need to make a distinction between like would you ever build an indexing that's only for AI and not for a human. And I

但是 Martin，我觉得我得插一句：等等，你刚才描述的并不是什么新东西，你描述的是一个架构上的——不，你描述的是当今软件的一个架构问题，就是它的 API 和性能瓶颈是基于『我能打多少字』来定的。这其实正是我想说的：我们的帮助系统当年是按照『一张 CD 能装下多少内容』来设计的，它完全不知道人们到底想做什么，没有任何上下文。但问题本身并没有变——我就是想做一张图表。是的，正是如此。再举一个真实的例子：我们发布了一个 Box agent，它内置了多得多的能力，其中一个能力是它能在你整个 Box 环境里做搜索，但它没有人类搜索那种局限——人类是打一个查询、拿回一组结果、再翻看一遍——而它会扇出去做多个查询，瞬间翻遍上百条结果，并对结果做自己的重排序（re-ranking）。所以这就好比，你绝不会想让一个 agent 被人类那套流程的速率给限制住。这恰恰对应人形机器人那种情形：你会愿意接受『好吧，人形机器人还是会走进电梯、还是会按按钮』。但其实在 agent 的世界里，你会想说『不不，我直接要你瞬间按到我要去的那一层』。对。对，不过我们得说清楚一点——顺便说，我非常同意——但我们得做个区分：你会不会去建一套只为 AI、而完全不为人类服务的索引。我

[38:24] Martin Casado

think that's less obvious. Yeah. So clearly there's like performance gains based on automation. We've got to evolve our architectures for those, but if you find a great way to index documents and you don't expose it to a human, I think that's

觉得这就没那么显而易见了。对。所以很明显，自动化会带来性能上的收益，我们确实得为此演进我们的架构。但如果你找到了一种很棒的方式去给文档建索引，却不把它暴露给人类，我觉得这就……

[38:35] Aaron Levie

Yeah, you got to Yeah, exactly. 100%. Well, well, this is I mean, it's actually I think this kind of moment probably reinforces some of Steven's, you know, kind of internet analogy on documentation. There is this really interesting thing where, you know, it started out where we as we've been building our our next set of agents, we first gave it this the current set of tools. We saw how how it used those. And then eventually we realized, oh, there's actually even even better way that the agent could do it. So we improved the underlying scaffolding. And then, oh, by the way, that will actually help the end user also. So it does let you sort of contribute back into the mothership of of technology of technology improvement that does, you know, sort of lift all the boats of of your users. Let me Let me ask this it occurred to me as you were saying it, like I my I sort of got all tense when you when the idea became No, like oh, we have 10,000 people hitting our hitting our SAS system today, and we've got it all working, and it's all great. But um now we're going to have 10,000 new peo- people, which are the agents for each of those 10,000 employees. And they're they're actually hitting it 500 times as much. Okay, so that SAS product will collapse. So like that's the the first order, because it wasn't architected for that volume. Like we saw this with with all the BI tools. Like when all the BI tools came out, all of a sudden they they were looking at the SAP data and trying to snapshot it and absorb the whole thing every night for a new kind of set of slices and dice it. Your view across all 500 And and like all the people making ERP were like, well, we don't do that. And so they had to go build all of this themselves because they had the knowledge of the data. Their API just

对，你得……对，没错，百分之百。嗯，这其实——我觉得这个时刻大概又印证了 Steven 那个关于文档的互联网类比。这里有件特别有意思的事：一开始我们在搭建下一代 agent 时，先给了它现有的这套工具，看它怎么用；然后我们慢慢意识到，哦，其实有更好的方式可以让 agent 来做这件事，于是我们改进了底层的脚手架（scaffolding）；然后，顺便说一句，那个改进其实也会帮到终端用户。所以它确实让你能把成果反哺回技术进步的『母舰』里，从而水涨船高地抬升你所有用户的体验。我来——我来问个问题，刚才你说的时候我突然想到：当那个念头冒出来的时候我整个人都绷紧了——比如说『不，我们今天有一万个人在打我们的 SaaS 系统，全都跑得好好的，一切都很棒』。但是呢，现在我们要凭空多出一万个新『人』，也就是这一万名员工各自对应的 agent。而它们对系统的访问量其实是原来的 500 倍。好，那这个 SaaS 产品就会崩。这就是第一阶效应，因为它当初根本没按这个量级来设计架构。我们在所有 BI 工具上都见过这个：当所有 BI 工具冒出来的时候，它们突然都盯上了 SAP 的数据，想给它做快照、每晚把整个东西吸收进来，好搞出一套新的切片切块视角，你那种横跨全部 500 强的视图。然后所有做 ERP 的人都说，呃，我们不干这个。于是他们只能自己去把这一整套都建出来，因为他们才掌握着数据的知识。他们的 API 就是

[40:19] Martin Casado

couldn't was not designed for that kind of work up. So, my my sort of thing to throw out there and and fight about is what is it what is the change management look like in a company? Because you you can't let loose an agent that hits the system at 500 X the humans. And it's not a token thing. It's an actual like, "Wow, we don't have the network bandwidth and the the throughput to handle 500 X for any one of our customers." So, what happens? So, I've so I've got a provocative adjacency, which you guys can tell me if I'm doing too much on a tangent here, but but here's my provocative adjacency, which is I don't know if having more agents is that big of an architectural shift. I just feel like we understand like whatever, if it's read-only data, you cache it. You know, like all the state issues are around mutable mutable globally shared state. We understand the limits of those. We know how to architect around those. We have to tackle all of those things when we went to the internet. And so, if you build your system not to handle it, like you suck at building the system and you deserve to go down and just go build a system that doesn't suck. And like I just feel like this is kind of standard computer science. However, I do think agents do introduce um something that organizations it technically have to deal with. And I'll let me just give the analogy in code, which is I think I Steven, this is what we call mogging on a I I don't know. YOU'RE BEING A BAD I I QUESTION MOG. I I QUESTION MOG. I HAVE NO IDEA WHAT HE JUST DID, but I I'm I'm just looking forward to how he magically made the problem go away. But go ahead. No, no, no, no, no, no, no, no, no, no, no, no, no, NO, NO, THE PROBLEM IS THERE. I just think like we know how to go from 10

根本没法——压根不是为这种工作量设计的。所以我想抛出来、想跟你们争一争的是：一家公司里的变更管理（change management）会变成什么样子？因为你不能就这么放一个 agent 出去、让它以人类 500 倍的量去打系统。而且这还不只是 token 的问题，而是实打实的：『哇，我们根本没有那么大的网络带宽和吞吐量，去为任何一个客户扛住 500 倍的量。』那会怎么样？所以我有一个挑衅性的旁支观点，你们可以判断我是不是扯得太远了——但这就是我那个挑衅性的旁支：我不确定『agent 更多了』算不算多大的架构转变。我就是觉得，比如说，如果是只读数据，你就缓存它嘛。所有的状态问题都围绕着可变的、全局共享的状态。我们清楚那些东西的边界在哪儿，我们知道怎么绕开它们去设计架构。当年我们上互联网时这些问题就全都得解决一遍。所以如果你把系统建得扛不住这个，那只能说你系统建得烂，活该宕机，那就去重建一个不烂的系统呗。我就是觉得这其实是挺标准的计算机科学。不过呢，我确实认为 agent 引入了某种组织在技术上必须应对的东西。我用代码来打个比方吧——我觉得，Steven，这就是我们说的『mogging』——我也不知道。你这是耍赖。我质疑这个 mog，我质疑这个 mog。我完全不知道他刚刚是怎么操作的，但我很期待看他怎么变魔术似的把问题给抹掉。不过你继续。不不不不不不不不不不不不不，不，不，问题还在那儿。我只是觉得，我们知道怎么从一万……

[41:53] Martin Casado

Yeah, it's there for stupid people. We just got rid of the stupid people. So, now everybody is smart. No, no. Okay, so so let me give you an example for coding. So, this is where I actually think there's a shift in how work gets done. So, when you code um with AI, your code kind of gets worse over time pretty materially. And so, it's almost like you're introducing as many problems as you are solutions. And I don't think we've actually figured out how to manage that. Does this make sense? 100% in the whole world right now. Yeah. I I I you know, I mean like this is this kind of reasonable question is you know, if you're using AI, yes, you're productive, but are you creating more problems than you've actually solved for solutions? And I do think that there's this actually, you know, open question when it comes to using agents on existing systems for creating things, which is like like do we know how to wrap the growing set of entropy around that? And I would say anecdotally, watching companies struggle with AI coding, which of course I'm I'm you know, listen, I I'm very close to many AI coding companies. I'm clearly very bullish on it. I don't think we know how to do that yet. And so, the you know, the agents on on a a system, I think we can tackle those with known techniques. Using agents for long-running things organizationally, where like, you know, you know, the the universe is kind of as clean as it was 3 days after then you started, I'm not actually quite sure we know how to do that at all. Well, I I love that point because that gets back to where we started, which is the difference between scale and not scale. Yes. And why it's perfectly rational for big company people to be like, "No freaking way is this coming into our company." Because a big company

对，那是给笨人准备的。我们刚把笨人都干掉了，所以现在人人都聪明。不不。好吧，那我给你举个编程的例子。这正是我真心认为工作方式会发生转变的地方。当你用 AI 写代码的时候，你的代码会随着时间推移变得明显更糟。所以这几乎就像是，你在引入的问题和你在解决的问题一样多。我觉得我们其实还没真正搞清楚怎么管理这件事。这能说得通吧？现在全世界都百分之百是这样。对。我是说，这是个挺合理的问题：用了 AI，你是更高产了没错，但你制造出来的问题，会不会其实比你解决掉的还多？我确实觉得，当你在现有系统上用 agent 去创造东西时，有一个悬而未决的问题，就是：我们到底知不知道怎么把那不断膨胀的熵给包住、管住？凭我的观察，看着各家公司在 AI 编程上挣扎——当然，我得说清楚，我跟很多 AI 编程公司关系都很近，我对它显然非常看好——但我觉得我们还不知道怎么做到这件事。所以，把 agent 用在一套系统上，我觉得我们能用已知的技术去应对。但是在组织层面把 agent 用在那种长期运行的事情上，让宇宙保持得像你刚开始三天后那样干净，我真不太确定我们到底知不知道该怎么做。嗯，我特别喜欢这个观点，因为它又绕回到我们开头那个点：规模化和非规模化之间的区别。是的。这也正是为什么大公司的人会说『绝对没门，这玩意儿别想进我们公司』，这是完全理性的。因为一家大公司——

[43:38] Martin Casado

is about to the wheels are going to come off a big company or a division in a big company or a product in a big company at any minute. Like if you're a Marty, we were both giant company executives. Like literally, we woke up every morning thinking, "Oh, the wheels are coming off today. This is the end of it. I'm getting fired by the 5:00 and we whatever started what I left yesterday thinking we were 3 months late and it's we're now 9 months late." And that's a typical day. And so, but the reason that that doesn't happen is because you put constraints all over the place. Exactly.

一家大公司随时都快散架了——大公司、大公司里的某个事业部、大公司里的某个产品，随时都可能轮子掉下来。Marty，我们俩都当过巨型公司的高管。我们每天早上醒来真的就在想：『哦，今天轮子要掉了，这就是末日了，五点我就要被炒了；昨天我走的时候本以为已经晚了三个月，结果现在变成晚了九个月了。』这就是稀松平常的一天。可它之所以最终没真散架，是因为你在所有地方都设了约束（constraint）。没错。

[44:11] Martin Casado

Which is exactly why Gilfoyle can't work at a big company because he he thinks he knows and it's also why all the the one-shotting vibe coding kind of people have no problem saying it's fine because they've never had to live in an environment where the constraint was to prevent the whole thing from imploding. And I I feel this is so critical Steve that you're like so in again, this is going to sound like a a little tangential but but it feeds into this which is I feel like core technologies kind of catered to like a some human need like the internet catered to like connectivity and social networking kind of catered to vanity and I feel like AI caters to our need to be productive. So I feel like we feel like we're being very productive when we do all of these things but we may actually be creating like mounds of extra work to do. And to

这正是为什么 Gilfoyle 没法在大公司干活——因为他自以为什么都懂；这也正是为什么那些『一发入魂』的 vibe coding 派会毫无负担地说『没事的』，因为他们从来没在那种环境里活过——在那种环境里，约束的存在就是为了防止整个东西自己塌掉。Steve，我觉得这一点太关键了——我又要说点听起来有点跑题的话了，但它其实接得上：我觉得核心技术往往都迎合了人类的某种需求。互联网迎合的是连接，社交网络迎合的差不多是虚荣，而我觉得 AI 迎合的是我们想要『高产』的需求。所以我们做这一切的时候，会觉得自己特别高产，但我们也许实际上是在给自己堆出一座又一座额外要做的活儿。而且——

[45:04] Martin Casado

Well, Aaron, you're deploying AI right now. Like boxes all in. So tell us tell share a story like of the wheels coming off or not coming off. Yeah. Well, well, I I think I think we're probably in the more pragmatic part of the continuum. So so which is why we don't claim that that's a 10x productivity gain to our engineering team. It's like no no cuz we have a lot of guardrails in place that create these constraints automatically in our system. There's we we still rely heavily on on code reviews. We still rely heavily on security reviews.

Aaron，你现在正在部署（deploy）AI 啊，Box 整个都押上去了。所以跟我们讲讲、分享个故事吧——轮子掉下来的，或者没掉下来的。好。嗯，我觉得我们大概处在那条光谱上比较务实的那一段。所以我们才不会声称这给我们工程团队带来了 10 倍的生产力提升。我们会说不不，因为我们设了很多护栏（guardrail），它们会在系统里自动制造出这些约束。我们仍然非常依赖代码评审（code review），仍然非常依赖安全评审（security review）。

[45:36] Martin Casado

you are you guys coding with like a rock and a chisel and stuff?

你们写代码是拿着石头和凿子在刻吗？

[45:39] Aaron Levie

it feels like that sometimes. We have chalkboards and like but no but like I I we had this new feature that we that we launched and I was like go go go go go and and AI built probably 80 to 90% of the feature and the the thing that slowed down the release of it was we have to do a full security review because we can't let there be any you know accidental code injection into the thing that we created. So so there's a lot of stuff where you kind of go super fast, but then you get still rate limited or constrained by some other part of the process. Uh and I I think that's sort of, you know, relatively natural until we figure out then that other part of the process, security reviewing one or the actual code review being one or just even your pipeline for for, you know, getting things into production being another one. So, we we're doing a quite a little, you know, quite a quite a bit of retooling of the whole product development life cycle, but I I don't think that it's a five to 10x gain. I do think it's a two to three x gain maybe across the board. You are still rate limited by how quickly can you review this stuff and and check on the work. Um I do think that that Martin's sort of pointing at though a thing that is the the big open topic across enterprises. Uh and and, you know, to some extent and engineers will fit will face it first and and will find the right equilibrium. The harder part still remains in the rest of knowledge work. This is why if you're in accounting, you know, we don't quite yet know when you could take your hands off the wheel, you know, doing a full accounting audit, you know, because of AI. What what you can do is have the AI go and and like comb through unlimited amounts of data to find anomalies that maybe are that would

有时候感觉就是这样。我们还用黑板呢。不是啦，但比如——我们新发布了一个功能，我当时就喊『冲冲冲冲』，AI 大概搭出了这个功能的 80% 到 90%，而真正拖慢它发布的，是我们必须做一次完整的安全评审，因为我们不能让我们造出来的东西里出现任何意外的代码注入。所以有很多地方，你前面跑得超快，但接着还是会被流程中的另一个环节给限速、给卡住。我觉得这在我们摸清那另一个环节该怎么办之前，是相对自然的——比如安全评审是一环，实际的代码评审是一环，甚至你把东西推上生产环境的那条流水线（pipeline）也是一环。所以我们正在对整个产品研发生命周期做相当多的重新改造，但我不觉得这是 5 到 10 倍的提升。我确实认为它差不多是全方位 2 到 3 倍的提升。你仍然会被『你能多快地评审这些东西、检查这些工作』给限速。我确实觉得 Martin 指出的是一件事，是当今所有企业普遍面对的大议题。某种程度上，工程师会最先碰到它，也会最先找到那个合适的均衡点。更难的部分仍然留在其余的知识工作里。这就是为什么——如果你是做会计的，我们还不太清楚你什么时候能放手让 AI 去做一整套完整的会计审计，因为现在还做不到。你现在能做的，是让 AI 去梳理海量到没有上限的数据，找出那些异常——那些也许是……

[47:13] Aaron Levie

alert your accounting team to, oh, we actually have to go dig into this. That's awesome because that's only net new level of visibility. Versus the part of the accounting process where you're doing a fine-tooth comb on making sure every single number is is accurate. That's probably still humans right now. So, I think the key is where do you find the productivity gains and I do think that that if you are a CEO or a board of directors or a management team, you're kind of trying to figure out and you're also getting confused cuz Silicon Valley's telling you all the things. And so, you have to sort of figure out where is where is the productivity most potent, where I actually can get the gain, I can get the success with less of the downside. And I think as an industry we're all sort of figuring this. By By the way, this is actually why I remain unbelievably optimistic on jobs because I don't think that you like I just think we've gotten it wrong on on thinking you know all the places where you're going to remove humans from this because you still need a human in that you know somewhere in the loop. Maybe the abstraction is a little bit higher and you don't need the human in the loop at every at every single stage that you needed a year ago but but you do need a human sort of kicking off the process, reviewing the process and incorporating whatever the work was. Um and so that creates just still a tremendous amount of opportunity in jobs across these organizations. Oh, let me I have to jump in cuz I I have I have a whole bunch of like visual aids I brought today to make it exciting. We got you got a bunch of comments um on the MTS live thing about people agreeing with you. So I don't want to let that slide because you know we complain about not agreeing with you

提醒你的财务团队：哦，我们得深入查一下这个。这特别棒，因为这是一种全新增加的可见度。而财务流程里那种用细齿梳一根根去核对、确保每一个数字都准确无误的环节，现在大概还是得靠人。所以我觉得，关键在于你从哪里找到生产力的提升。我确实认为，如果你是 CEO、是董事会、是管理团队，你其实是在努力搞清楚——而你还被搞得一头雾水，因为硅谷在跟你讲各种各样的东西。所以你得自己去琢磨，生产力在哪里最有效、我真正能拿到收益、能拿到成果，同时副作用又最小。我觉得作为一个行业，我们都还在摸索这件事。顺便说一句，这其实正是我对就业依然抱有难以置信乐观态度的原因。因为我不认为你——我只是觉得，在判断哪些地方会把人从流程里剔除出去这件事上，我们一直搞错了，因为你在流程里某个环节始终还是需要一个人。也许抽象层次更高了一点，你不再需要像一年前那样在每一个阶段都有人盯着，但你确实需要有人去启动整个流程、审阅流程、并把产出成果整合进去。所以这在这些组织里仍然创造了海量的就业机会。哦，我得插一句，因为我今天带了一大堆视觉道具来，想让现场更有意思。我们收到——你收到一大堆关于 MTS 直播的评论，都是人们在认同你的观点。所以我不想让这个一带而过，因为你知道我们老是抱怨没法认同你。

[48:41] Martin Casado

but but like here to your point to your point this was a book in the 80s called the end of work. Yeah. And and I this so actually sorry it was in the 90s. It it it came out like six months before the internet hit. And the whole thesis was the technology revolution was a complete bust and we got no gains in productivity but now there's going to be no more jobs cuz the economy is stagnant and this was the guy he called himself a futurist. Yeah. And and like so the whole notion that it it that's like one of the neat things about this whole AI moment is like the number of things that when you hear them the first time you think they're stupid and then you go back and think about it and you're like oh my god it's way stupider. And and this idea that like AI just gets rid of jobs it's is ancient as like you picked talked about the accountant. Like one of the things people thought was that computers would get rid of accountants. Yes. And and that was like IBM's pitch in like 1965 but what it actually did was like oh my god we could do so much more with accounting now that they're not like literally just adding numbers all day. Yes. And and I think when you look at like just the notion of like creating information, synthesizing, and all that. Like AI is is an accelerant for that for a person who knows what they're doing, and companies are suddenly going to want more of those people creating more of that information. Not to mention the fact that if AI is creating valuable information and there's more of it, then more people will need to consume it to do something. And the idea the essence of a company is acting on information. And and so this idea that information is just going to get produced easily and be in surplus and not used makes no sense

但是，照你这个说法——照你这个说法，80 年代有一本书叫《工作的终结》。对。不好意思，其实是 90 年代，它差不多在互联网爆发前六个月出版。整本书的核心论点是：技术革命彻底失败了，我们在生产力上毫无收获；但与此同时再也不会有新工作了，因为经济停滞——而写这本书的家伙自称是个未来学家。对。这整套说法，恰恰是这波 AI 浪潮里最有意思的一点：很多观点你第一次听到时觉得它蠢，然后你回过头去想一想，会发现“天哪，它其实蠢得更彻底”。这种“AI 就是要消灭工作”的想法，跟你刚才聊的会计师一样古老。当年人们就觉得计算机会消灭会计师。没错。那基本就是 IBM 在 1965 年左右的推销话术。但计算机实际带来的结果是：哇，既然会计师不用整天就只是加数字了，我们现在能在会计这件事上做多得多的事。没错。我觉得，当你看待“创造信息、综合信息”这类事情时，对于一个懂行的人来说，AI 是一个加速器；而企业会突然想要更多这样的人，去生产更多这样的信息。更别说，如果 AI 在创造有价值的信息、而且信息越来越多，那就需要更多人去消化这些信息、去做点什么。而一家公司的本质，就是依据信息采取行动。所以这种“信息将被轻易生产出来、然后供给过剩、却没人用”的想法，完全说不通。

[50:24] Aaron Levie

at all. Because as you know, like in the unstructured information world, the problem is that you can make it, but the consumption of it effectively is the gating to it. That and that's the gating factor now. Um we I I I think we had a conversation with one of our board members who's a chair of our audit committee and and so he's a CPA and and he was telling us, you know, kind of early in his career. I I can't even retell it cuz it felt so manual, but so I don't I don't even know how the world worked but I don't know how the world worked before all of the modern technology, but he was explaining the CPA's process and I was like, you know, it seemed like the most manual thing of all time, but but and and Stephen, I think this is right out of your book is like it was actually quite simple in in sort of the amount of things you could do because of of how undigitized and relatively manual the whole thing was. And computers actually only made it more complicated, more comprehensive, and thus created even more jobs because because of that complexity that that we introduced. And and you can just sort of see how easy this is to show up in so many areas of work is like we can just now we can afford to make things more complex. And so if if you make things more complex, then actually you eventually still run into now new constraints of who can understand that complexity. And and and and so like, you know, it's like to me it's like the funniest concept that the more code we write, the less we would need engineers. It'd be the opposite because because now your systems are even more complex than before, which means that you're going to be running into even more challenges of when you need to do a system upgrade or when there's downtime and you have to

一点都说不通。因为你知道，在非结构化信息的世界里，问题在于你能造出信息，但对它的消化吸收才是真正的瓶颈。这才是现在的限制因素。嗯，我们——我想起跟我们一位董事的一次对话，他是我们审计委员会的主席，是个注册会计师（CPA）。他跟我们讲他职业生涯早期的事，我都没法转述，因为听起来太手工了，我都不知道那个世界是怎么运转的——我是说，在所有现代技术出现之前那个世界怎么运转的。他在解释 CPA 的工作流程，我心想，这简直是有史以来最纯手工的活儿。但是——Stephen，我觉得这正好出自你书里的观点——正因为当时一切都没数字化、相对纯手工，所以你能做的事在数量上其实相当有限，流程反而很简单。而计算机实际上只是让它变得更复杂、更全面，从而创造出了更多的工作，正是因为我们引入了这种复杂性。你可以清楚地看到，这种现象在多少工作领域里都会冒出来：现在我们终于负担得起把事情做得更复杂。而一旦你把事情做得更复杂，最终你又会撞上新的约束——谁能理解这种复杂性。所以对我来说，最好笑的概念就是：我们写的代码越多，就越不需要工程师。恰恰相反，因为现在你的系统比以前更复杂了，这意味着你会撞上更多的挑战——当你需要做系统升级，或者出现宕机时，你得去……

[51:56] Martin Casado

figure out like what Well, how do I fix that problem or when there's a security incident? Uh uh And so, yeah, I mean, this is this is uh this is like we're just getting started with the jobs on this front. Right, it all Listen, we're we're a few years actually into this and you can actually look a bit at the data, too, right? Like, what are the companies that are hiring the fastest? Like the AI-native companies. They're hiring like crazy. Yeah. But not only that, like I remember there was this early prognostication, which is AI writing code will get rid of infrastructure. Like, it's going to commoditize infrastructure and like which which is this kind of very strange um prediction given the fact that there's more software than ever before been written, right? And right, sitting on the board of a bunch of infrastructure companies, some that have been flat for a while, they're all doing fantastic cuz there's so much software and there's so much more software out there now. And so, listen, if you look at the data on the ground from the companies, it's more software. The AI-native companies are hiring the the the the the most and so, it's very clear to me that we're in an expansion phase.

去搞清楚到底怎么——我该怎么修复那个问题，或者出现安全事件时怎么办。所以，是啊，我是说，在就业这件事上，我们才刚刚起步。没错，这一切……听着，我们其实已经进入这个阶段好几年了，你完全可以看看数据，对吧？哪些公司招人招得最快？正是那些 AI 原生公司。它们在疯狂招人。对。而且不只如此，我记得早期有这么一个预言：AI 写代码会消灭基础设施，会把基础设施商品化——这是个相当奇怪的预测，因为现在写出来的软件比以往任何时候都多，对吧？我在一堆基础设施公司的董事会里，其中有些公司业绩平了一阵子，但它们现在全都干得特别好，因为软件太多了，现在外面的软件比以前多太多了。所以，听着，如果你看一线公司的真实数据，就是软件更多。AI 原生公司招人招得最猛。所以在我看来非常清楚，我们正处在一个扩张期。

[52:53] Aaron Levie

And and and the maybe just my my only final point on this one at least is um is I think people we we have a little bit of a myopic view in Silicon Valley on on thinking that you know, engineering jobs are you go to work at at Google or name your your you know, tech company and startup and that that's an engineering job. And and look, we're so wired into that because of obviously the ecosystem that that we're all part of. Um and then you sort of forget well, like John Deere is trying to make automated tractors and Caterpillar is trying to have AI systems and Eli Lilly is trying to design even more pharmaceutical, you know, kind of you know, therapeutics. Um and and just you can go through 5,000 other companies. They're going to now have the next set of engineers that are going to use Claude Code and Codex and Cursor to be able to automate even more of of their businesses and be able to design and develop even more software for their workflows and their systems. And so, it just might you know be that you don't go and work on a social network and improve the social network algorithm. You go work at John Deere and you improve the the you know intelligent farming algorithm. And and that and and we just have to you know I mean this is sort of like like completely you know like Marc Andreessen predicted this you know 15 years ago like software's going to eat the world and what that means though is that everybody's going to have lots of software. And this gives everybody the ability to finally have lots of software, but you still need then an expert or a semi-expert to be actually going and prompting the the the you know the agent on what to do, reviewing its work, and managing the system that it builds. So, so all of the you know predictions on don't go into coding and

我在这一点上至少最后再补充一句：我觉得在硅谷我们有点目光短浅，总以为工程岗位就是你去 Google 上班、或者随便哪家科技公司、初创公司，那才叫工程师工作。看，我们之所以被这么束缚住，显然是因为我们都身处的这个生态。然后你就会忘了——比如 John Deere（约翰迪尔）正在做自动化拖拉机，Caterpillar（卡特彼勒）想要 AI 系统，Eli Lilly（礼来）想要设计出更多的药物、那种治疗方案。你还可以一路数过另外五千家公司。它们现在都将拥有下一代工程师，这些工程师会用 Claude Code、Codex 和 Cursor 去把自家业务的更多环节自动化，去为他们的 workflow 和系统设计、开发出更多软件。所以，可能只是说，你不再去某个社交网络工作、去改进社交网络算法，而是去 John Deere 上班、去改进那个智能农业算法。这——我是说，这基本上就像 Marc Andreessen（马克·安德森）15 年前预言的那样：软件将吞噬世界。但它的含义是，每个人都将拥有大量软件。而这给了每个人终于能拥有大量软件的能力，但你仍然需要一个专家、或者半个专家，去真正给那个 agent 下提示、告诉它该做什么，去审阅它的工作、去管理它搭建出来的系统。所以那些“别去搞编程、别去……”的预言。

[54:27] Martin Casado

don't go into software engineering, I think will be proven quite quite wrong. I think I mean look at that was super good Aaron and I think that that the the the base case of all of this is just that it there's too many people out there right now that don't like technology and have a static view of the world. So, when they look to whatever it is that they think AI is going to do and people hear automation, they just assume it's going to take things away. Like here's Well, we have a lot of people who like technology though that are also creating that uh that Right right. So, here's like this is article fighting the paper chase. Lower lower lower lower. Well, I'm looking at a feed. Even lower. I'm looking at a feed. What are they Oh, you're looking at a feed. Okay. Oh, sorry. Okay. Okay. Okay. Oh, I see. Oh, you're looking at my Mac camera and yeah oh, that's why. Oh, you're fancy. You're fancy.

“别去做软件工程”的预言，我认为会被证明是大错特错的。我是说，看——Aaron 说得太好了。我觉得这一切的基本盘就在于：现在外面有太多人不喜欢技术，而且用一种静态的眼光看世界。所以当他们去看 AI 将要做什么、当人们听到“自动化”这个词时，就直接假设它会把东西夺走。嗯，不过我们也有很多喜欢技术的人，他们也在创造那些……对对。所以，这里有篇文章叫《与纸海搏斗》。再低一点，再低，再低，再低。呃，我在看一个画面流。再低一点。我在看一个画面流。他们在……哦，你在看一个画面流。好吧。哦，抱歉。好好好。哦，我懂了。哦，你在看我的 Mac 摄像头，对吧，哦，难怪。哦，你真讲究，真讲究。

[55:14] Martin Casado

Right. So, this is like Time magazine every kid in high school read it 1981. And but the whole view of what computers would do would be they would automate the paper in a company. And so, the idea like the whole first generation of computing was literally taking paper forms and turning them into something on a screen, then printing them out, and then making it all easier. And you fast forward and it's all of these things that you just said Aaron like, you know, there was an era when lawyers didn't type. And so what happened was they just they they had people who were legal assistants, they called them paralegals, and they did all the typing. And then like some students at Harvard, they brought a computer into the classroom. And so this is I'm lowering it so you guys can see. Yeah, yeah, yeah, yeah. So they brought that's a original laptop in there in the early 1980s. And they brought they brought this computer into the classroom and then they got thrown out for using it and but they were literally they used to used to do law school and you'd write the essays in longhand in a book and then the professor would have to read them. And now of course you just type them and you have access to the database of all the citations, but that's exactly like nobody deals with a lawyer who isn't in track changes with their with your contract. Right. And and I last I checked there are way more lawyers today than there were 30 years ago. And they all are every human lawyer you talk to is a computerized lawyer. Their citations come from from the internet, their their information in the brief comes and they type the brief.

对。所以，这是 1981 年的《时代》杂志，每个高中生都读过它。当时对计算机将要做什么的整体看法是：它们会把公司里的纸质工作自动化。所以，第一代计算机的整个理念，字面意义上就是把纸质表格搬到屏幕上变成某种东西，然后打印出来，再让这一切变得更轻松。然后你快进到现在，就成了你刚才说的那一切，Aaron——你知道，曾经有个年代律师是不打字的。当时的做法是，他们雇一批法律助理，叫他们“律师助理（paralegal）”，由这些人负责所有打字工作。后来哈佛有几个学生把一台电脑带进了课堂。所以——我把它放低一点，让你们都能看到。对对对对。他们带进去的那是 80 年代初一台最原始的笔记本电脑。他们把这台电脑带进课堂，结果因为用它而被赶了出去——但他们当时真的是，过去读法学院你得用手在本子上写文章，然后教授得读这些手写稿。而现在你当然是直接打字，还能访问所有判例引用的数据库。但这就像——现在没人会去找一个不用“修订模式（track changes）”跟你处理合同的律师。对。而且据我所知，今天的律师比 30 年前多得多。他们每一个人——你接触到的每一个真人律师都是“计算机化的律师”：他们的引用来自互联网，辩护状里的信息靠它来获取，辩护状也是他们打字写出来的。

[56:41] Aaron Levie

think I I think we we you know, kind of going back to the my myopic approach, I think we maybe over like I mean as a big lover of technology, I I wish this was true, but I think we just over assume that like everybody's job is is just they're just inside of Microsoft Word and they're just typing a a word document. It's like like I mean, most of the time with lawyers I'm like, you know, strategizing something or or they're working through a a complex analysis of a situation um and it's not like I could go to an AI for for advice, but and but that would probably only increase the chance that I go and then call a lawyer to say, "Hey, what do you think about this this situation that that you know, that I'm that that we're dealing with?" Um and so a lot of these jobs just have a lot of context that aren't sitting just you know, literally on the on the computer doing all the work. They they do have to kind of touch grass um as a part of the job and and so then AI yeah AI will help automate the creation and production of the content and the review and of the information but then it still has to be incorporated into the real world of of real value production. I I feel like we're live and we're supposed to end at 4:00. So what I'm going to I'm going to do is just say we're live and it's 4:00 and I guess that means we just stop and some lights fade or something. None of us have done this before. We don't know what's supposed to happen. But Someone is waving at me and smiling saying yes, I think you're right. The smile The smile means stop talking. Okay. All right. Well, it was great to see everybody. Bye, everyone.

我觉得，回到我说的那个“目光短浅”的角度——我觉得我们也许过度地，我是说，作为一个特别热爱技术的人，我真希望这是真的，但我觉得我们只是过度假设了：好像每个人的工作就是待在 Microsoft Word 里、就是在敲一份 Word 文档。但其实——我是说，大多数时候我跟律师在一起，都是在谋划某件事，或者他们在梳理对某个情形的复杂分析，而不是说我可以去找 AI 要建议——但那其实大概只会增加我去打电话给律师的概率，去说“嘿，你怎么看我们正在处理的这个情况？”。所以很多这类工作都带有大量上下文，这些上下文并不是干干净净地就摆在电脑上等着干活的。它们确实需要去“接接地气”，这是工作的一部分。然后 AI——对，AI 会帮忙把内容和信息的创造、生产以及审阅自动化，但接下来这些还得被整合进真实世界、整合进真正价值的生产里。我感觉我们是直播状态，而且我们本该在 4:00 结束。所以我打算做的就是直接说：我们在直播，现在是 4:00，我猜这意味着我们就这么停下，然后灯光淡出或者怎样。我们谁都没干过这个，不知道接下来该发生什么。不过——有人在朝我挥手、微笑着说“是的，我觉得你说得对”。那个微笑——那个微笑的意思是“别说了”。好。好的。那么，很高兴见到大家。各位，再见。