软件正在吞噬劳动力

全球SaaS市场规模约为每年3000亿美元。仅美国的劳动力市场就高达13万亿美元。软件如今瞄准的目标，它所追逐的奖赏，正是劳动力市场。几乎每家软件公司本质上都是将一个文件柜转化成了数据库。但如今正在发生的是，整个流程正被端到端地彻底重构。

The worldwide SaaS market is about $300,000,000,000 per year. The labor market in The US alone is $13,000,000,000,000 What software is now going after, the prize that it's going after, is the labor market. Almost every software company has basically taken a filing cabinet and turned it into a database. But what's happening now is the whole thing is effectively done end to end.

软件吞噬了世界，现在它正瞄准劳动力。在a16z的LP峰会上，负责应用基金的主管合伙人Alex Rampell登台阐述为何真正的市场机遇不是3000亿美元的SaaS产业，而是13万亿美元的美国劳动力支出。从文件柜演变的数据库到真正执行工作的人工智能，Alex解析了基于结果的软件、新型定价模式，以及当智能代理自主销售、支持和收款时将发生的变革。

Software ate the world. Now it's coming for labor. At a16z's LP Summit, general partner Alex Rampell, who leads the Apps Fund, took to the stage to discuss why the real market opportunity isn't the $300,000,000,000 SaaS industry. It's a $13,000,000,000,000 US labor spend. From filing cabinets turned databases to AI that actually does the job, Alex breaks down outcome based software, new pricing models, and what happens when agents sell, support, and collect on their own.

让我们深入探讨。

Let's get into it.

我要谈谈软件如何吞噬劳动力。首先，马克十年前曾在《华尔街日报》发表过一篇关于软件吞噬世界的文章。既然劳动力属于世界的一部分，这自然是个延续话题。但如果今天只能传达一个核心观点，那就是劳动力市场——这几乎显而易见——远比软件市场庞大得多。全球SaaS市场规模约为每年3000亿美元。

I'm gonna talk about how software eats labor. First, Mark wrote an essay for The Wall Street Journal a long time ago, about ten years ago, about how software eats the world. I guess the labor force falls within the world, so this is a natural follow-up. But if there's only one takeaway that I can leave with you today, it's that the labor market, this is almost obvious, is so much bigger than the software market. So the worldwide SaaS market is about $300,000,000,000 per year.

全球市值约22万亿美元。仅美国的劳动力市场就达13万亿美元，规模悬殊。这并不意味着软件市场会突然变成每年13万亿美元的产业，但软件当前追逐的终极目标正是劳动力市场。我觉得没有什么比用世界上最著名的共产主义者画像开启风投会议更合适的了——这就是卡尔·马克思。

Worldwide market cap is about $2,200,000,000,000. The labor market in The US alone is 13,000,000,000,000, so much, much bigger. And it doesn't mean that the software market will suddenly be worth $13,000,000,000,000 a year, But what software is now going after, the prize that it's going after, is the labor market. And I figured there's nothing better than starting a venture capital conference with a picture of the world's most famous communist. That's Karl Marx right there.

如果你读过《资本论》（我大学时读过），其核心教义是资本与劳动力的对立，资本剥削劳动力。二者本是截然不同的概念。但令人振奋的是，这就像化学反应方程式——你们给我们的资本，我们投给企业。企业用资本做什么？购买或租赁GPU。

And if you've read Das Kapital, which I read back in college, if you remember, like, the central tenet is that there's capital and labor and capital exploits labor. They're two distinct things. But what's really exciting, this is almost like a chemical equation, is that the capital that you give us, we give to companies. What do the companies do with the capital? They buy GPUs or rent them.

他们雇佣工程师，购买咖啡。将咖啡和GPU交给工程师后，产出的软件就能替代劳动力。这就像新的E=MC²方程式。我们目睹无数企业正在实践这种模式，它们之所以能快速扩张，正是因为直接向客户或终端用户承诺：我们将代劳这项工作。

They hire engineers. They buy coffee. They give coffee and GPUs to the engineers, and then out pops software that does the job of labor. It's like the new E equals MC squared. And we're seeing this happen at so many companies, and these companies are scaling so quickly because they really are selling into clients or end users and saying, hey, we'll do this job for you.

我们不是给你软件。我们要为你完成一项工作。这就是我们的销售说辞。显然，这个概念本身并不新鲜。我的意思是，自动化已经存在很长时间了。

We're not giving you software. We're going to do a job for you. And that's the sales pitch. And obviously, like this concept is not new per se. I mean, automation has been around for a long time.

比如，缝纫女工，织布机取代了那些工作，但你仍然需要有人操作织布机。或者蒸汽船。帆船和蒸汽船显然使那种劳动形式效率更高，但你仍然需要有人操作蒸汽船并往里面铲煤，还要把帆升到桅杆上。当然，还有古腾堡发明的印刷机，那是一项重大创新。蒸汽动力印刷机让《洛杉矶时报》能印刷出数十万份报纸，而不是每次都需要人工手动摇动印刷机。

So, you know, seamstresses, the loom replaced those jobs, but you still had somebody that had to operate the loom. Or steam ships. Sailboats and steam ships obviously made that form of labor much more efficient, but you still need somebody to operate and shovel coal into the steam ship and put the sails up on the mast. And then, of course, the printing press invented by Gutenberg, that was a big innovation. And the steam powered press allowed the LA Times to print out hundreds of thousands of copies of that newspaper versus somebody manually cranking one each time.

最后，装配线是一项巨大的创新，它取代了手工组装每个小装置的人。现在你有了装配线。但是，你知道，很多人在装配线上工作。所以，这个概念并不新鲜。你总是可以用资本制造一台机器，然后在另一端获得更高效的劳动力。

And lastly, like, the assembly line was a massive innovation in terms of replacing people that hand assembled every gizmo. Now you have an assembly line. But, you know, a lot of people worked on the assembly line. So again, this concept is not new. Like, you could always take capital, make a machine, and then have more efficient labor on the other side.

但现在发生的是，整个过程实际上是从头到尾完成的。我想带你回顾一下软件市场一直以来是什么样子，因为我认为理解过去是真正理解未来和可能性的最佳窗口。我的观点是，几乎每一家软件公司基本上都是把一个文件柜变成了一个数据库。这就是2.2万亿美元市值的来源。这就是每年3000亿美元软件收入的来源。

But what's happening now is the whole thing is effectively done end to end. And I want to take you through a journey of what the software market has always been, because I think understanding the past is the best window of actually understanding the future and what's possible. So my thesis is that almost every software company has basically taken a filing cabinet and turned it into a database. And this is where the $2,200,000,000,000 of market cap has come from. This is where the $300,000,000,000 in annual software revenue has come from.

我想指出的第一个例子是一家叫Sabre Systems的公司。如果你见过这个名字，它有两个a。这是一家位于德克萨斯州的公司，但这是美国航空和IBM的一个合作项目，当时IBM是世界上最强大的计算机公司，也是世界上最强大的科技公司，因为至少根据ChatGPT的说法，1959年预订一张机票是什么样子的？美国航空是怎么处理的？他们可能有，比如，很多像这样的文件柜，里面有很多这样的表格。

And the first example that I'd like to point to is a company called Sabre Systems. This is with two a's if you've seen it. It's a company in Texas, but this was a joint project of American Airlines and IBM back when IBM was the most powerful computer company in the world, technology company in the world, because at least according to ChatGPT, what would it have looked like to book an airline ticket in 1959? How did American Airlines handle this? They probably had, like, lots of filing cabinets with, like, sheets like this.

所以贝蒂·欧文斯打电话来说，我想坐在4A座位，和电话那头的人交谈。哦，等等。实际上，我想取消我的航班。他们必须擦掉它。不。

So Betty Owens calls up and says, I wanna sit in Seat 4 A, talks to somebody on the phone. Oh, wait. Actually, I wanna cancel my flight. They have to erase it. No.

实际上，我想坐在2C座位，再擦掉一次。文件柜基本上记录了一切。这效率很低。有很多人在操作文件柜。你无法从一个办公室共享信息到另一个办公室，因为一切都存放在一个文件柜里。

Actually, I wanna sit in Seat 2 C, erase that again. Filing cabinets basically kept track of everything. It was pretty inefficient. You had a lot of people that worked the filing cabinets. You couldn't share information from one office to another because everything was domiciled in one filing cabinet.

而Sabre通过将所有数据置于IBM大型机上，并在全球各地旅行社设置众多瘦客户端终端来访问该主机，从而改变了游戏规则。这就是旅游业如何被革命化的。Galileo为酒店业做了同样的事，Amadeus则为欧洲市场实现了这一变革。我是说，许多大公司由此崛起。

And Sabre changed the game by putting it all on an IBM mainframe and then having lots of kind of thin client terminals that were used by travel agents around the world that could access that mainframe. And this is how travel was revolutionized. Galileo did this for hotels. Amadeus did this for Europe. I mean, like, big companies came out of this.

当然，销售领域也经历了同样的过程。我记得昨天有人提到《拜金一族》这部精彩的电影。如果你还记得，片中有大量名片场景。大家都想拿到Glengarry的销售线索——那可是优质资源。

And, of course, the same process happened for sales. So I think somebody yesterday mentioned Glengarry Glen Ross, great movie. But if you remember, there are lots of business cards. Wanted to get the Glengarry leads. Those were the good leads.

那些线索原本都是纸质文件。对于像我这样的老辈人，可能还记得八十年代有家叫Axe Systems的公司。那是当时顶尖的CRM企业之一。1990年Goldmine紧随其后，1993年汤姆·西贝尔创立了Siebel Systems。但这些系统本质上都是将销售文件柜的内容，最初搬到了大型机上，直到1999年Salesforce出现，将其迁移至云端。

Those were pieces of paper. And for those of you that are old like me, you might remember a company called Axe Systems in the nineteen eighties. This was one of the big CRM companies of the day. Goldmine followed that in 1990, and then Tom Siebel started Siebel Systems in 1993. But all of these took the filing cabinets of sales, put them originally in kind of mainframes, and then Salesforce came in 1999, put that in the cloud.

但你看，同一位销售员——在五十年代虚构电影中需要翻阅实体档案的人——到了2010年就变成了查询Salesforce记录。流程相同，只是媒介不同。制造业和库存管理是另一个重大领域。假设你是个产品制造商...

But, you know, again, the same salesperson that accessed the physical file in a fictional movie set in the nineteen fifties would now access a Salesforce record in 2010. Same process, just different medium. Manufacturing and inventory. This is another big one. So imagine that you're a product manufacturer.

我有多少库存？销售情况如何？IBM再次引领潮流，但其他企业如1972年创立的SAP（至今仍在运营）、Bond、JD Edwards、Epicor、Sage、I2等，都是将传统记录数字化的巨头企业。

How many widgets do I own? What's my inventory? What's my sales? IBM again at the forefront of this, but other companies, some that are around today like SAP started in 1972, Bond, JD Edwards, Epicor, Sage, I two. These are all big companies that basically digitized old fashioned records.

我最喜欢的例子是图书馆卡片目录——这个理念的普及程度令人惊叹。图书馆存在已有数千年历史（想想亚历山大图书馆），杜威十进制分类法出现后，我小时候还得去查按字母排序的卡片目录。后来OCLC等公司将这些目录数字化，现在只需在图书馆电脑终端输入查询就能定位书籍。

My favorite one, just to show how pervasive this idea is, is there is actually a big business in library card catalogs. So libraries were have been around for a very, very long time. Library of Alexandria, right? Like long, long time. The Dewey Decimal System comes out, and when I was growing up, I'd go to like the card catalog and figure out, okay, they were all alphabetically sorted, and then eventually somebody created, started with this company called OCLC, built a reasonably sized business and sort of innovative Searcydynix of digitizing those card catalogs where now you enter into a terminal at the computer in the library, and then you figure out where your book is.

书籍是否在馆？法律案卷也是如此——八十年代我去律所时，大部分空间都被文件柜占据。PC Law等公司，以及LexisNexis和路透社的很大收入都来自帮律所数字化文件，否则这些文件会占满他们在第五大道的豪华办公室。我父母是会计师，记得小时候去他们办公室，五岁小孩根本无处玩耍——到处都是文件柜。后来Intuit推出QuickBooks数字化财务报表，还有七十年代的Peachtree、MYOB等众多公司，本质上都是在替代文件柜、文件柜。

Is it in stock? Legal case files, like every time I went to a law firm in the nineteen eighties, it's like most of the square footage was filing cabinets, and companies like PC Law, like a lot of LexisNexis and Reuters revenue comes from selling to law firms, digitizing things that would otherwise occupy so much square footage in the nice Fifth Avenue offices of law firms. My parents were accountants growing up, and I remember going to their office. And again, there was no room for a little five year old to run around because all filing cabinets Intuit comes along with QuickBooks, digitized financial statements, Peachtree, this is a company from the nineteen seventies, MYOB, like lots of companies. Again, filing cabinets, filing cabinets.

你们明白这里的套路了。软件史上我最喜欢的名字是文件柜。第一家电子健康记录公司叫Mumps，这可以说是软件史上最糟糕的名字。它像是输给了疟疾或麻疹之类的，但这是麻省总医院的项目，他们想取代文件柜，于是发明了这个叫MUMS的编程语言和数据库系统。而Epic和Cerner，Epic是全球最大的电子健康记录公司，成立于1979年。

You get the drill here. My favorite name in the history of software is filing cabinets. The first electronic health records company was a company called Mumps, which arguably is the worst name in the history of software. It, like, somehow lost out to malaria or measles or something, but this was Mass General Hospital, and they wanted to replace their filing cabinets, and they came up with this programming language and database system called MUMS. But Epic and Cerner, Epic is the biggest electronic health records company in the world, it was started in 1979.

再说一次，他们所做的只是将这些医院系统和医生办公室里堆积如山的文件数字化。最后说说人力资源和薪资。实际上，在Sabre之前，自动数据处理公司ADP早在1949年就成立了。如何记录考勤？那时有时间单、考勤卡。

So again, and what they did was just digitize these massive number of files that every hospital system and doctor would have in their office. And kind of lastly here, HR and payroll. So actually, even before Sabre, automated data processing, ADP, was started in 1949. How do you keep track of time and attendance? You had your time slip, your time cards.

如何计算代扣税款？所有这些公司应运而生，发展轨迹都是：我把文件放进大型机，然后像Workday这样的公司本质上就是PeopleSoft——由同一团队创立。他们将其搬上云端，但流程依旧。1940年查看你考勤单的人，和2015年的是同一批人，只是媒介从纸张变了。

How do I figure out tax withholding? All of these companies came out and again, the progression was I take the file, I put it in a mainframe, and then companies like Workday was effectively PeopleSoft. It was the same team that started PeopleSoft. They put it in the cloud, but the same process was there. The people that looked at your time and attendance slip in 1940 were the same people that looked at it in 2015, but now the medium was not paper.

不再是大型机，而是云端。我之所以提这些，是因为效率并未真正提升——文件柜由人查阅，数字记录同样由人查阅。对吧？

It was not mainframe. It was cloud. And the reason why I mentioned all of this is because nothing has actually gotten that much more efficient because the filing cabinets were read by humans. The digital records are read by humans. Right?

就像这位女士，无论她在做什么，帮客户解决支持需求之类的，过去是看纸张，现在是看电脑。理解这一点至关重要，因为整个软件商业模式必须改变。必须改变。我称之为星巴克式的SaaS高杯、大杯、超大杯模式。随便看哪家SaaS公司的首页，大概都长这样。

It's like this woman here, whatever she's doing, looking at helping a customer with customer support needs or something, once upon a time would have looked at a piece of paper, now is looking at a computer. And the reason why this is so important to understand is because the whole business model of software has to change. It has to change. This is what I call as homage to Starbucks here, the tall, grandi, venti model of SaaS. If you go to any SaaS company in the world, go to their landing page, it probably looks like this.

这家叫Zendesk的公司现已私有化，几年前被Premier和Hellman & Friedman收购。这家年收入20亿美元的公司在卖席位——最受欢迎的'超大杯'套餐是Suite Professional，每月115美元。

This is a company called Zendesk. It's now a private company. It was taken private by Premier and Hellman and Friedman a few years ago. And this is a $2,000,000,000 ARR company that sells seats. So their most popular, like the Venti package is you get the Suite Professional, dollars 115 a month.

但这两天我们讨论过，AI现在非常擅长处理客户咨询。如果你的每个客服效率提升9000倍，还需要多少席位？想象你有1000个席位，想象我的客服中心有1000名员工，每人年薪总成本7.5万美元。

But we've talked about in the last couple days how now AI does a really good job answering customer support queries. So how many seats do you need if every one of your agents is 9,000 times more productive? Imagine that you have a thousand seats. Imagine that I've got a thousand people working in my customer support call center. I pay each one fully loaded $75,000 a year.

所以每年人力成本是7500万美元。那我的软件成本呢？按115乘以12再乘以1000计算，软件年成本约为140万美元。可见人力成本远高于软件成本。

So $75,000,000 a year in cost for people. Well, what's my software cost? Well, it's a thousand times a 115 times 12. It's about $1,400,000 a year for the software. So the people cost is so much higher than the software cost.

对于像Zendesk这样的公司，可能面临两种发展路径。如果AI能解答所有问题，还需要多少客服席位？零。一个席位都不需要。AI处理一切事务，而Zendesk是按席位收费的。

And this could go one of two directions for a company like Zendesk. If it turns out that AI can answer all the questions, how many seats do you need? Zero. You don't need a single seat left. The AI answers everything, and then Zendesk is charging per seat.

这样他们的年收入就会从140万美元直接归零。那将非常糟糕。但另一方面，看看这个数学计算：目前每个客服每年处理2000个问题，使用Zendesk作为客服系统的企业，每个回答的人力成本约37美元，软件成本69美分，合计每回答成本38美元。

So they would go from 1,400,000.0 to $0 in revenue. That would be very bad. On the other hand, I mean, look at the math here. If each human is answering 2,000 questions a year, right now, your fully loaded cost as the company that uses Zendesk as your system of record for answering customer support, it's about $37 of human cost and 69¢ of software cost. The cost per answer is $38.

这是个粗略的公司案例。或许Zendesk可以改收年费500万美元——比如告诉客户：别再每年花7500万做客服了，改付我们500万就行。

And this is rough company example. Maybe Zendesk could charge $5,000,000 a year. Like, hey, don't spend $75,000,000 a year on support anymore. Spend $5,000,000 a year on support. Just pay it all to us.

别付我们140万，付500万就能省下7500万。Zendesk正站在悬崖边上：收入可能归零，也可能翻三倍。最终走向何方？我...我也不知道答案。

Don't pay us 1,400,000.0. Pay us 5,000,000, save 75. So Zendesk is really at the precipice of it's like the revenue could go to zero or the revenue could three x. And like, where is it gonna go? I I don't know the answer.

他们也不知道答案。我和他们CEO聊过，他们正在新西兰试点基于服务效果的定价模式。所以新西兰掌握着所有问题的答案，敬请期待后续发展。

They don't know the answer. I've been talking to their CEO. They're piloting outcome based pricing in New Zealand right now. So New Zealand holds the answer to all of our questions. So stay tuned.

再举个例说明劳动力/准劳动力市场的庞大规模：我们之前提到的13万亿美元工资总额，软件收入相比之下微不足道。单看护理行业——因为我们有家投资组合公司在这个领域——美国注册护士年收入约6500亿美元，450万从业者，这比全球软件市场规模还大。虽不意味着护理软件市场也会这么大，但说明这才是真正的竞争池。

And just to give again another example of just how much bigger these labor or quasi labor markets are, like the $13,000,000,000,000 of wages that we talked about, you know, software revenue is very small. If you just take one category, like just one particular profession and take the example of nurses just because we have a portfolio company in this space, nurses in The United States Of America earn about $650,000,000,000 a year. They're about four and a half million registered nurses. That's bigger than the entire worldwide software market. And it doesn't mean that the nursing software market will be this big, but it means like this is the pool that you're really playing against.

我想从文件柜开始讨论的原因在于，这将逐渐导向实际成果。软件将从单纯的文件存储功能，转变为能对文件柜进行智能操作。这意味着什么？以第一个例子来说，如果我有一个旅行文件柜，或许软件能帮我改签航班，或者我需要为儿子高中75名学生预订团体旅行。

And the reason why I wanted to start with the filing cabinets is like this is going to start moving to outcomes. Like, the software is going to go from being the filing cabinet to effectively operating on the filing cabinet. And what does that mean? Well, you know, take the first example. If I have the filing cabinet for travel, you know, maybe the software can rebook flights or I wanna book a trip for 75 kids at my son's high school.

我不再需要联系人工客服，只需与软件公司对话。直接联系美联航的AI系统，它们就能全程处理。销售方面——这其实很明显。

I need to talk to an agent. I don't need to talk to an agent anymore. I just talk to the software company. I talk directly to United Airlines, you know, AI, and they do the entire thing for me. Sales, I mean, this is kinda obvious.

Salesforce按席位按月收费，但它应该直接帮我销售产品。我不想为上千个席位付费，只想为实际客户付费。听着：

You know, Salesforce charges per seat per month. Salesforce should just sell for me. I don't wanna pay for a thousand seats. I wanna pay for customers. Like, hey.

去帮我获取客户，反正你本就是后台系统。或者调研我所有客户，给每人打30分钟电话，了解他们近况——是否会续约？制造业呢？

Go get me customers because you're the back end for that anyway. Or survey all of my customers, do a thirty minute phone call with every single one of them, see how they're doing. Are they Are they gonna renew with me? Yes or no. You know, what about manufacturing?

假设我生产零件，而关税政策正在调整。我希望了解关税风险，只需询问ERP系统：帮我研究这个。或者需要审计，或联系供应商确认能否按时交货。

Well, imagine that I make widgets and there are these things called tariffs that are happening. I'd love to figure out what my tariff exposure is. Let me just ask my ERP system. Research that for me. Or I want you to do an audit or call my suppliers and see if they're gonna still be able to ship me their things on time.

就像这个疯狂的图书馆目录系统。如果我的书逾期了，不该由图书管理员催还，而该由借阅软件公司提醒我：'朋友，请还书'或者'这本书很畅销，多进些货——比如Ben的书快卖完了'

Library card catalogs as crazy as this one is. You know, what if my book is overdue? The librarian shouldn't be calling me. The library card software company should be calling me saying, hey, buddy, return the book or order more books because this one's very popular. You know, Ben's book is selling out.

你就该多订购几本。法律案卷领域也如此，许多新兴软件公司不再只是考勤记录系统，而要能起草合同、完成实际工作，甚至开始计费服务。

You order more copies of that. Legal case files, like a lot of software companies that are popping up in this space, it's no longer the system of record for, time and attendance. Like, draft me a contract. Like, do that work for me. You could start billing out.

再次强调，尚不清楚这种定价模式将如何运作，但请开始为此计费。会计与簿记，你知道的，应收账款账龄汇总。这个概念由来已久。我有很多公司欠我钱，很多客户也欠我钱。在1940年，我会看着打印出来的清单说，我要给这些客户打电话，希望能收回款项，或者派个拿着撬棍的家伙去，总得采取些措施。

Again, unclear how the pricing model of this is going to work, but start billing for that. Accounting and bookkeeping, you know, AR aging summary. Like this concept has been around for a long time. I have a lot of companies that owe me money, a lot of clients that owe me money. In 1940, I'd look at the printout and say, I'm gonna call these customers and hopefully collect for them or send the guy with the crowbar or do something.

到了二月，我会查看QuickBooks的数据并做同样的事。现在软件公司可以像QuickBooks那样开始给客户的客户打电话说，嘿，你欠我们钱。请还钱，我现在就能通过电话接受付款。健康记录。如果你看到我一瘸一拐地离开舞台，很遗憾，我大约三个月前做了跟腱修复手术。

In February, I'd look at the QuickBooks thing and do the same thing. Now the software company can go call like QuickBooks is gonna start calling customers of of their clients and say, hey, you you owe me money. Please pay me back, and I I can accept a payment on the phone right now. Health records. So if you saw me limp off stage, I sadly had Achilles repair surgery about three months ago.

我不推荐这个手术。术后第二天，斯坦福医院打电话问我，亚历克斯，从1到10分，你的疼痛等级是多少？我说11分。他们说真幽默。我说，不，真的是11分。

I don't recommend it. The day after my surgery, I got a call from Stanford Hospital saying, you know, Alex, on a scale of one to 10, what's your pain level? And I said 11. And they said very funny. And I said, no, it really is 11.

AI护士无法实施心肺复苏，对吧？AI护士也无法处理枪伤受害者。但AI护士完全可以给像我这样四十多岁的中年病人打电话问，你感觉如何？有什么我们能帮忙的吗？

And an AI nurse can't do CPR. Right? An AI nurse can't attend to a gunshot wound victim. But an AI nurse can totally call a mid forties patient like me and say, how are you doing? Is there anything we can help with?

比如，你发烧了吗？哦，你发烧了。你现在应该马上去医院之类的。重申一遍，这些都是可实现的成果。如果你有我的医疗记录，这些就是可以执行的操作，并为此收费，每通外呼电话收费20美元。

You know, do you have a fever? Oh, you do. You should go to the hospital right now and stuff like that. And again, these are the outcomes. These are the operations that can be performed on if you have my medical record, that's what you do with it and charge for that charge $20 for that outbound call.

人力资源与薪酬。我如何进行背景调查？如何确认你简历上写的确实在这三家公司工作过？你知道吗？Workday应该直接联系这三家公司询问，亚历克斯真的在那里工作过吗？

HR and payroll. How do I do a reference check? How do I make sure that you on your resume, you said you worked at these three companies? You know what? Workday should call those three companies and say, did Alex really work there?

解释福利政策。协助注册。如果Workday开始提供这些服务，他们的收入可能会翻三倍，因为他们已经是记录系统了。就像Airbnb的故事，众所周知，Airbnb最初是通过某种程度上抓取Craigslist起家的。Craigslist是90年代中期那个糟糕的网站，自那时起就再未改变过，上面有大量公寓出租信息。

Explain benefits. Help with enrollment. Workday could probably triple their revenue if they start doing this because they're already the system of record. So if you know the story of Airbnb, Airbnb famously started by somewhat scraping Craigslist. Craigslist is this horrible site from the mid nineties that has not changed since the mid nineties, and they have lots of listings for apartments.

你去看看公寓，有一半时间都是骗局，一半时间已经被预订了，或者你知道，它每天都在重新上架，他们基本上就是把这些房源放在一个更好的界面里，然后称之为爱彼迎。他们就是这样起步的。这是我们看到的最令人兴奋的事情之一，这是克雷格列表上的真实广告。因为我那令人沮丧的跟腱受伤，我有了大把时间——我以前每天跑步、骑车，现在做不到了。所以我就在克雷格列表上闲逛找工作。

And you go look at an apartment and like half the time it's a scam, half the time it's already been booked, or you know, it's been relisted every single day, and they basically put them in a better interface and called it Airbnb. And that's how they got started. And this is one of the most exciting things that we're seeing, which is here's a real ad on Craigslist. Because since I have so much copious time due to my sad Achilles injury, I I used to run every day, bike every day, can't do that anymore. So I hang out on Craigslist looking for jobs.

当然不是为我自己找。这是广场巷验光所的招聘广告，他们要招一名前台接待员，这个职位空缺已经挂了六个月。整整六个月都招不到人。现在加州规定必须标明薪资待遇。

I'm not for myself, of course. And here's a job for Plaza Lane Optometry. They're hiring a front desk receptionist, and they've had this job opening for six months. It's been hanging out there for six months. The job like, you have to now say in California how much your chart how much you're gonna pay.

年薪4.5万美元。按照供需关系，如果他们开价10万美元年薪，这个职位早就招到人了。但他们需要把年薪控制在4.5万美元才能符合成本模型。看看职位要求——如果你能看清的话——第一条是开关店铺、锁门。人工智能可干不了这个。

It's $45,000 a year. So supply and demand, if they probably said we're gonna pay a $100,000 a year, they would have filled this job. They're paying they they need to pay $45,000 a year to make this work within their cost model or something. If you look at the job responsibilities, if you can read this, like the first one is like open and close the shop, lock the door. AI can't do that.

但其他很多职责AI都能完全胜任。比如和保险公司扯皮、提前一天打电话提醒患者避免爽约。如果有人没来就诊，对验光师来说是重大损失。如果你评估验光市场是否适合软件开发，肯定会断然否定——因为广场验光所每年软件预算可能只有500美元。

But a lot of the other job responsibilities AI can completely do. It's, you know, argue with insurance company, call patient the day before their appointment to prevent no shows. If somebody doesn't show up, that's a big that's a big opportunity cause for the optometrist. And if you were to look at the optometry market and say, is that a good software market? You would say categorically no, Because plasma optometry probably spends $500 a year on software.

他们可能就买了个微软Office许可，用Squarespace或Wix建了个网站，仅此而已。所以年软件支出大概500美元。在这个新时代，我们看到很多公司这样做：他们浏览克雷格列表。

They probably have a Microsoft Office license. They probably have a website with Squarespace or Wix, and that's it. So they probably spend $500 a year. So in this new world, and again, we're seeing a lot of companies doing this. They peruse Craigslist.

找到招聘信息后就说：'广场巷验光所您好，我想应聘这个职位'。验光师回复：'太好了，说说你的资质？之前在哪里工作？'软件公司会说：'我知道这听起来很疯狂，但我们能完成除开关门之外的另外八项工作。'

They look for a job listing, and they're like, hey, Plasalain Optometry, I want to apply for the job. And you know, the optometrist is like, okay, that sounds great. Tell me about your qualifications. Where did you work before? It's like, I know this sounds crazy, but I'm a software company, and I can't I can't close the doors and do this, but I can do these other eight things.

'能给您演示吗？'起初验光师会拒绝，但最终会说'好吧试试看'。而且每年只要2万美元，远低于你们招不到人的4.5万人工成本。这事正在真实发生，简直不可思议。

Can I give you a demo? And at first, the optometrist is like, no, I wanna oh, okay. I'll I'll try it out. And by the way, it's $20,000 a year, which is much less than the $45,000 a year that you were gonna pay for a human that you can't even hire. So this is happening right now, and it's incredible.

而且，再次强调，像软件支出，对这些人来说是小头。有效劳动力支出则要高得多。这正是推动许多冷门行业市场大幅增长的原因。所以与其我空谈，不如给你们一个实例。这是我们投资组合中的一家公司——快乐机器人，服务于货运和卡车运输领域。

And, again, like, the software spend, small for these people. The effective labor spend, much, much higher. And and this is what's massively growing the market for a lot of these obscure industries. So I just thought rather than me talk, I wanna give you an example. This is one of our portfolio companies, Happy Robot, which serves the the the freight and and trucking space.

那么请注意听。这是一通与潜在客户及AI进行的电话谈判实录。

So so listen. This is a negotiation on the phone conducted with a prospective client and AI.

好的。伊利诺伊州乔利埃特到

Yep. Joliet, Illinois to

明白。这是今天从伊利诺伊州乔利埃特出发，上午6点至下午2点间装货。周一上午6点至下午4点间送达。这单我出价700美元。您要预订这趟货运吗？

Alright. So this is Joliet, Illinois to today between 6AM and 2PM. Delivers Monday between 6AM and 4PM. I have 700 on this one. Would you like to book this load?

哎呀...这单我得要800才行。

Gee. I'm gonna need 800 for this one.

我可以确认下。目前我们给不到800。您有可能稍微接近货运平台的标准报价吗？

I can check on that. So we can't do eight at this point. Any chance you could come a bit closer to the load board rate?

775美元是我的底线，真的不能再低了。

I could I could do with $7.75. It's really my lowest.

我明白了。所以价格不可能比7.75美元更优惠了吗？

I understand. So no chance you could do any better than $7.75?

7.5美元怎么样？

How about $7.50?

让我看看。好吧，我这边可以给你735这个数。

Let me see. Okay. I was able to get you 735 here.

735？好吧。

735? Okay.

那么这笔交易就结束了。谁是机器人，谁是人类？这就像新版图灵测试。待会儿你们可以用自己的理论来考我。再来看另一个例子。

So that that's a closed transact. Who who's the robot, and who's the human? This is like the new Turing test. I'll let you guys quiz me afterwards on that with with your own theories. Here's another example.

这家名为Salient的公司从事催收业务。如果你是贷款方，比如汽车贷款机构，你需要收回贷款还款，而Salient为国内许多汽车贷款机构提供服务。这是他们另一个通话案例。

This is a company called Salient that does collections. So if you have if you're a lender, if you're an auto lender, you you have to collect loan you have to collect repayments on your loans, and Salient serves a lot of auto lenders in this country. And here's here's another call from that.

目前您的账户已逾期51天，欠款金额为825.35美元。请问今天能安排还款吗？

Currently, your account is fifty one days past due for $825.35. Would you be able to make a payment today?

Salient的厉害之处在于它能说几十种语言。比如塔加洛语、越南语、中文普通话。这确实是个关键点——它不仅仅关乎成本问题。

So the cool thing there is that Salient speaks dozens of languages. So it speaks Tagalog. It speaks Vietnamese, Mandarin. And again, like, this is a really important point. It's not just about cost.

关于成本，人们普遍存在误解。并不是说'因为人力太贵而AI更便宜，所以AI会抢走所有工作'。AI能解决很多问题，比如间歇性需求就是个大难题。想象我是黑色星期五的零售商。

Cost like, this is where people are getting this wrong. It's not like, oh, AI is gonna take all the jobs because humans are too expensive and AI is cheaper. AI does a bunch of things. Like, intermittent demand is a big one. Imagine that I'm Black Friday retail.

作为零售商，大家都知道黑色星期五期间销量激增。传统做法是临时雇佣大量收银员（实体店）或客服人员（网店）来处理咨询。但元旦过后呢？难道要全部解雇，等到十一月再重新招聘？实际上九月份就得招人培训了。

I'm a retailer, and people know that, you know, sales are much higher around Black Friday. So I would have to hire tons of cashiers or if I'm an online retailer, hire lots of customer support people to go answer queries and whatnot. And then what do I do on January 1? Do I fire them all and then rehire them in November? But actually, I should probably hire them in September because I have to train them.

这很棘手。许多行业都存在间歇性需求问题——比如美联航遇到芝加哥恶劣天气时（虽然这种情况很少见），不可能一夜之间招聘培训上万名员工。而这正是AI的强项。另一个例子是那些令人沮丧的工作。

Kind of tricky. And there are a lot of other countries that have intermittent demand issues. I mean, United Airlines, if they have bad weather over Chicago, which of course never happens, they can't just hire and train 10,000 people overnight. But again, AI very, very good at that. The other example is there there are a lot of demoralizing jobs out there.

什么叫令人沮丧的工作？催收就是个典型。你得打电话说'您逾期了'，根据我听过的一些通话记录，约半数情况会遭到对方辱骂。就像'嘿你欠我钱'，对方回'去你妈的别再来电'。

What is a demoralizing job? Well, collections is kind of a demoralizing job because you go call people and say, hey, you're overdue. About half the time, and I've listened to some of these calls, there are lots of expletives on the on the other side. Right? So it's like, hey, you owe me money.

反复循环：'提醒还款'-'滚远点别烦我'。人类会对此感到疲惫，但AI不会受影响。所以这类打击士气的工作非常适合AI。

It's like, f you never call me again. It's like, call back again. Hey, you owe me money. F you never call me again. A human would get kind of tired of that.

这不是份愉快的工作。AI不会感到困扰。再说回监管确定性的问题...

It's not a fun job. AI doesn't get bothered. So again, demoralizing job is very, very good. Regulatory certainty. So go back to this.

要知道，我曾联合创办了一家名为The Firm的公司。我们每个季度都会接受关于UDAP的培训，哦抱歉，是不公平、欺骗性和滥用性行为的法规。有许多法律规定了你能和不能对客户说的话。想象一下，你打电话给客户说‘嘿，你欠我钱’，而客户回敬一句‘去你的’。

You know, I I co founded a company called The Firm. We got trained every quarter on UDAP, unfair, sorry, unfair, deceptive and abusive practices. So there are lots of laws that mandate what you can and can't say to a customer. So imagine that you're calling a customer and say, hey, you owe me money. The customer says, F you.

电话那头的人可能刚经历了糟糕的一天，你没法责怪他们。但如果回骂一句‘去你的，客户’，那接电话的人可就惹上麻烦了，对吧？当你能够有效编程让机器人全程处理通话时，这种情况的确定性会高得多。

And then you have somebody on the other end who's had a kind of a bad day and you can't blame them. It's like F you back customer. They're going to get in trouble for that. Right? And you have more certainty when you can program effectively a robot to conduct the entire call end to end much more so than people.

我最喜欢的例子是某个学了太多语言的人——我会说俄语、日语和一点西班牙语，花了太多时间学习这些。语言就是语言。现实情况是，如果我只说波斯语，斯坦福医院有能打电话询问我疼痛等级的波斯语护士吗？如果我只说蒙古语呢？

And my my favorite example here is somebody who studied, you know, too many I I I speak Russian, Japanese, and a smattering of Spanish and and spent way too much time learning that stuff. Language language is languages. I mean, it's just fact that a nurse, an AI nurse like, what if I only spoke Farsi? Does Stanford have any Farsi speaking nurses that can call me about my pain level? What if I only spoke Mongolian?

现在他们有了。AI护士、AI催收员、AI谈判专家——所有这些都能立即以数十种语言完成。你根本不会为此雇佣人类。你没法在爱荷华州随时找到一个会说塞尔维亚语的人来做这种间歇性又令人沮丧的工作。AI在这方面表现优异。

Like, now they do. An AI nurse, an AI collections agent, an AI negotiator, like all of these things can be done in dozens of different languages instantly. And you just wouldn't hire a human for that. You can't get somebody in Iowa that speaks Serbian on demand intermittently with a demoralizing job and so on and so forth. You know, AI does great.

AI扩展市场的能力也很重要。这就是为什么我要从档案柜的故事讲起——因为合规领域原本没有软件公司。根据美国劳工统计局数据，美国增长最快的职业是美甲师（我干不了这个），第二快的是合规专员。

And it's also very important that AI expands the market. This is why I wanted to start with the story of filing cabinets because there was no software company for compliance. Because this is actually a fact from the Bureau of Labor Statistics, The fastest growing job in America is manicurist, pedicurist. I can't do that. The second fastest is compliance officer.

对于合规专员，你不需要软件。如果你是花旗银行，你需要的是更多人。没有公司能有效开发这类软件，因为软件市场太小而人力市场很大。现在你可以走进花旗说：‘我能提供端到端的合规软件解决方案’。

So compliance officer, you don't need software for compliance officers. If you're a city bank, you need more people. And no company has popped up effectively doing software because it's like the software market's kind of small. The people market's very large. Now you can roll into Citi and say, hey, I will do compliance for you end to end as a software solution.

每年收他们1000万美元，我就是你们的合规追踪软件。因为以前他们只会购买更多微软Office授权或增加催收人员——催收领域根本没有软件公司，只有催收机构和催收员。现在你实际上可以凭借语音技术切入，虽然这会快速商品化（我们很清楚这点），但你能转型为具有真实软件收入、软件利润和软件留存率的软件公司。

Pay me $10,000,000 a year, and I'm your software product that kinda tracks everything. Because before they would just they would just get more Microsoft Office licenses or collections. Like, there is no software company for collections. There are collections people at collections firms. Now you can actually enter potentially with the wedge of voice, and this is gonna get commoditized quickly, and we realize this, but you can backfill into a software company with real software revenue, real software margins, real software retention.

你们将开始看到的另一件事是，现在有许多非AI公司因为AI而运作。由于我提到的那次非常不幸的受伤，正如我所说，我是个狂热的自行车爱好者。我的车库堆满了闲置的自行车，希望未来能再次使用它们。为什么没有自行车版的Airbnb呢？

The other thing that you're gonna start seeing us do is there are a lot of non AI companies that now work because of AI. So because of my said very sad injury, I'm I'm a big cyclist, as I mentioned. My garage looks like this with a lot of bikes that aren't being used. I hope to use them again in the future. Why isn't there an Airbnb for bicycles?

为什么没人开发这个？我来告诉你原因——这是个糟糕透顶的主意，所以没人做。为什么说它糟糕？

Why hasn't somebody built that? Well, I'll tell you why. It's a very bad idea. That's why nobody's done it. And why is it a bad idea?

因为核心原则（AI？不，可能是二月份的事，这不重要）是：如果客户获取成本加上商品销售成本超过客户终身价值，就别继续。这根本不算生意。但有了AI后，想象我想做自行车版Airbnb——

Because the central tenant AI, not AI, it could be in February. It doesn't really matter. If your customer acquisition costs plus your cost of goods sold is greater than your lifetime value, do not proceed. That's not a business. But now with AI, you know, imagine that I wanted to start Airbnb for bicycles.

我该怎么找到车库里有闲置自行车的人？难道要雇一群帕洛阿尔托的昂贵斯坦福学生，提供20种椰子水口味，迎合千禧代的癖好才能开展销售？不。我会让AI销售代表打电话，每个AI销售年成本仅几百美元，而非十万美元。

Well, how am I going to go get the people that have spare bicycles in their garage? Am I going to hire a bunch of like, you know, expensive Stanford kids in Palo Alto and have to give them 20 different flavors of coconut water and, you know, cater to their millennial needs in order for them to work in the sales operation of Palo Alto? No. I'm gonna have an AI sales rep call everybody. And the cost per AI sales rep per year is a few $100, not a $100,000.

连椰子水都不用准备。出现紧急情况？现在有AI客服专线能处理一切——报警或其他需求。最后，如何审核用户？

No coconut water necessary. Well, what if there's an emergency? Well, you know, now there's a +1 800 number that you call that's an AI rep that can do everything. Call the police, call whatever. And, you know, lastly, like, how do I screen this person?

背景调查？自行车状况检查？是否被盗？这些我们不会投资的愚蠢业务环节，AI全能搞定。

Do a background check? Is the bicycle good? Is it stalled? Like, whatever you would need to do this this silly business that we're not gonna fund. AI can do this as well.

于是整类曾被客户获取成本或商品成本扼杀的业务变得可行。当然，你还能用AI快速生成代码进入这些领域。AI基础设施正在压低CAC和COGS，很快所有公司都会使用这些工具，市场规模将呈指数级扩张。

So you actually have a whole category of of business, which was like, it would have worked except for this pesky customer acquisition cost or pesky pesky cost of goods sold. And of course, AI, you can vibe code your way into one of these businesses anyway. So, you know, massively expands the size of the market. The non AI market, given the AI infrastructure just allowing, you know, CAC to come down, COGS to come down for now. I mean, every company is going to start using these tools.

所以，你知道，这会变成尤吉·贝拉名言的翻版——‘这里太拥挤了，反而没人来了’。但现在很多公司正在尝试一些几年前根本行不通的老新点子。这是一个全球性的机遇。我是说，比如美国劳动力市场，规模庞大，每年高达13万亿美元。

So, you know, it will become a version of the Yogi Berra quote, It's so crowded, nobody goes here anymore. But now a lot of companies kind of prosecuting sometimes older new ideas that just wouldn't have worked five years ago. And this is a global opportunity. Mean, like the The US labor market, big. It's $13,000,000,000,000 a year.

全球劳动力市场要大得多。而我们代表您的资本在这里的工作，我们代表您的资本的劳动，就是找到那些能让软件行业相形见绌的最佳企业。非常感谢大家。

The worldwide labor market's so much bigger. And our job on on behalf of your capital here, our our labor on behalf of your capital is to find the best companies that will make software look small. So thank you very much.

感谢收听本期a16z播客。如果您喜欢本期内容，请务必点赞、评论、订阅、给我们评分或写评论，并与亲朋好友分享。更多节目请访问YouTube、Apple Podcasts和Spotify。在X平台关注我们@a16z，订阅我们的Substack专栏a16z.substack.com。再次感谢收听，下期节目再见。

Thanks for listening to this episode of the a 16 z podcast. If you like this episode, be sure to like, comment, subscribe, leave us a rating or a review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on x at a sixteen z, and subscribe to our Substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode.

温馨提示：本内容仅作信息参考，不应视为法律、商业、税务或投资建议，也不用于评估任何投资或证券，且不针对任何a16z基金的现有或潜在投资者。请注意a16z及其关联机构可能持有本期播客讨论企业的投资。更多详情及投资披露链接，请访问a16z.com/disclosures。

As a reminder, the content here is for informational purposes only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any a sixteen z fund. Please note that a sixteen z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a 16z.com forward slash disclosures.