//本文来自硅谷A16Z投资博客
For centuries, the biggest goal of science was alchemy — to turn lead into gold. It turns out this is actually possible in nuclear physics and is called transmutation. Bombard mercury with neutrons in a nuclear reactor or particle accelerator…and presto! Gold.
几个世纪以来,科学的最大目标是炼金术——将铅变成黄金。事实证明,这在核物理学中实际上是可能的,称为嬗变。在核反应堆或粒子加速器中用中子轰击汞......然后 “瞧”金。
Today, thanks to AI, we are witnessing potentially the greatest transmutation in history. Software becomes labor. It’s the new E=MC2. Capital buys coffee, engineers, and GPUs. Out comes code that takes the role of labor. This will both grow existing software markets and create many new software markets where “per seat” pricing never allowed for a large outcome.
今天,多亏了 AI,我们正在见证历史上最伟大的嬗变。 软件变成了劳动力。这是新的 E=MC2。资本购买咖啡、工程师和 GPU。出现承担劳动角色的代码。这既会扩大现有的软件市场,也会创造许多新的软件市场,在这些市场中,“按席位”定价永远不会带来大的结果。

Historically, much of software digitized an offline form of storage, put it in a database, and then provided an accessible, permissionable front-end for the end-user who likely did not know SQL, with the speed benefits of a digitized, networked medium.
从历史上看,许多软件将离线存储形式数字化,将其放入数据库中,然后为可能不了解 SQL 的最终用户提供可访问、可许可的前端,并具有数字化、网络化介质的速度优势。
PeopleSoft and then Workday digitized the HR filing cabinet.
PeopleSoft 和 Workday 实现了 HR 文件柜的数字化。
Zendesk digitized old-fashioned support “tickets.”
Zendesk 将老式的支持“票证”数字化。
Quicken and Quickbooks digitized old-fashioned ledgers.
Quicken 和 Quickbooks 将老式分类账数字化。
Epic and Cerner digitized health records.
Epic 和 Cerner 将健康记录数字化。
Salesforce digitized the Rolodex and the chalkboard pipeline.
Salesforce 将 Rolodex 和 chebboard 管道数字化。
E-mail digitized, well, mail!
电子邮件数字化,嗯,邮件!
Instead of having Bill the HR clerk or Sally the accounting clerk literally “fetch” the files from the filing cabinet, William and Sarah in IT now exist to make sure that everyone in HR or accounting is properly provisioned and has access to the system. But accounting headcount and HR headcount is…the same.
IT 部门的 William 和 Sarah 现在不再让人力资源文员 Bill 或会计文员 Sally 从文件柜中“获取”文件,而是确保人力资源或会计部门的每个人都得到适当的配置并有权访问系统。但会计员工人数和人力资源人数是......一样。
But now — and this is the big change — the “users” of the digitized filing cabinet do not have to be humans.
但现在 — 这是一个巨大的变化 — 数字化文件柜的 “用户” 不必是人类。
Let’s look at Zendesk’s pricing:
让我们看看 Zendesk 的定价:
There’s a giant opportunity right now for all of the “digitized filing cabinet” companies — Intuit, Workday, Zendesk, etc. — to now charge for “quasi humans” to replace or augment the humans who accessed and acted upon the records in the past. This will likely necessitate large-scale changes to historical “per-seat” pricing models, since a business will need fewer (if any!) seats as the system…takes actions on its own, or even empowers humans to be 10x more productive.
现在,所有“数字化文件柜”公司(Intuit、Workday、Zendesk 等)都面临着巨大的机会,现在向“准人工”收费,以取代或增强过去访问记录并对其采取行动的人工。这可能需要对历史上的“按席位”定价模式进行大规模更改,因为企业将需要更少的(如果有的话)席位,因为系统......自行采取行动,甚至使人类的工作效率提高 10 倍。
Workday could charge for HR conversations. Intuit could charge for sending out collections letters for overdue receivables. Epic could charge for a post-appointment check-in. Salesforce could charge for, well, selling.
Workday 可能会对 HR 对话收费。Intuit 可能会对逾期应收账款的催收函寄出收费。Epic 可能会对预约后办理登机手续收费。Salesforce 可以收取销售费用。
To be clear, this does not mean the end of white-collar work – if anything, AI will likely create new “AI jobs” that just weren’t possible with human costs or intermittent demand. There’s a popular saying in economics: “The cure for high prices, is high prices.” Meaning: as the price of an item goes up, more manufacturers decide to produce it (or existing manufacturers produce more), therefore increasing the supply, therefore decreasing the price. In the case of skilled labor, though, there’s simply too much latency. If nurses were paid 5x more tomorrow, it might not appreciably change tomorrow’s supply at all (maybe some retired nurses would un-retire!), but it would likely cause more to enter nursing school and graduate in 3 years. 3 years. What if a surge of nurses are needed…tomorrow?
需要明确的是,这并不意味着白领工作的结束——如果有的话,人工智能可能会创造新的“人工智能工作”,而这在人力成本或间歇性需求下是不可能的。经济学中有一句流行的话:“解决高价格的良方是高价格。意思是:随着商品价格的上涨,更多的制造商决定生产它(或现有的制造商生产更多),从而增加供应,从而降低价格。但是,对于熟练劳动力来说,延迟实在是太大了。如果护士明天的工资高出 5 倍,那么明天的供应量可能不会明显改变(也许一些退休护士会不退休!),但这可能会导致更多人进入护士学校并在 3 年内毕业。3 年。如果需要大量护士怎么办......明天?
For other professions, the training is “on the job” and it’s simply impossible to align the training time with the ebbs and flows of the business. Mortgage rates went down in 2021, creating a surge of demand for mortgage brokers…which ended suddenly as rates went up. Delta Airlines likely needed tens of thousands of trained support reps in the wake of the Crowdstrike disaster. The fifth fastest-growing job in the U.S. over the past 20 years is “compliance officer” where, again, training (and not just wages!) create shortages. AI will always show up to work, can be trained instantly, and will happily fill the “market void” caused when there is only intermittent or episodic demand for highly skilled work. Why go to school or learn something for years if your skill is only needed twice a year? AI does not have this problem.
对于其他职业,培训是“在工作中”,根本不可能使培训时间与业务的起伏保持一致。抵押贷款利率在 2021 年下降,导致对抵押贷款经纪人的需求激增......随着利率上升,这种情况突然结束。在 Crowdstrike 灾难发生后,达美航空可能需要数万名训练有素的支持代表。美国增长速度第五大的工作岗位 在过去的 20 年里,“合规官”再次导致培训(而不仅仅是工资)造成短缺。AI 将始终出现在工作中,可以立即进行训练,并且会很乐意填补对高技能工作只有间歇性或间歇性需求时造成的“市场空白”。如果你的技能一年只需要两次,为什么还要上学或学习很多年呢?AI 没有这个问题。
From a venture capital lens, where almost every company has a compelling “Why Now” (if it was such a good idea, why didn’t it work before?), I’d like to think there have been three eras of cloud software:
从风险投资的角度来看,几乎每家公司都有一个引人注目的“为什么现在”(如果这是一个好主意,为什么它以前没有奏效?),我想认为云软件已经有三个时代:
- The Original Cloud Era (~1999-2007), when markets were big enough for pure software, largely displacing on-premise software. This is when players like Salesforce, NetSuite, Veeva, Hubspot, etc. emerged.
最初的云时代 (~1999-2007),当时市场足够大,可以容纳纯软件,在很大程度上取代了本地软件。这就是 Salesforce、NetSuite、Veeva、Hubspot 等参与者出现的时候。 - The Financial Services-Enabled Cloud Era (~2010-Present), when markets became big enough with embedded payments, lending, and other “fintech” add-ons. For example, the software market for restaurants was small, but the “software + payments” market for restaurants is big — hence Toast (founded in 2012). The software market for HVAC contractors was small, but the “software + payments” market is big — hence ServiceTitan (founded in 2012), etc.
金融服务支持的云时代(~2010 年至今),当时市场随着嵌入式支付、贷款和其他“金融科技”附加组件而变得足够大。例如,餐厅的软件市场很小,但餐厅的“软件 + 支付”市场很大——因此 Toast(成立于 2012 年)。暖通空调承包商的软件市场很小,但“软件 + 支付”市场很大——因此 ServiceTitan(成立于 2012 年)等。 - The “AI-Enabled Outcomes” Cloud Era (Present – ), when the software market wassmall because labor is the primary cost and the software need — the digital filing cabinet — was relatively small or generic. The software provider for many professions is just Excel and Word, versus specialized tools.
“AI 支持的结果”云时代(现在 – ),当时软件市场很小,因为劳动力是主要成本,而软件需求(数字文件柜)相对较小或通用。许多专业的软件提供商只是 Excel 和 Word,而不是专门的工具。
This isn’t to say that the only AI successes will be “net-new” categories. But as is often the case, in the battle of distribution vs. innovation, distribution is the default winner — and existing software companies have distribution. Particularly now, in 2024, when AI is the top priority for virtually every software company and end customer. In 2007, most CEOs thought the iPhone was stupid (no keyboard!) and Blackberry was better. In 1996, most retail CEOs thought that the internet was a toy, a fad, and surely people wouldn’t buy things from a web browser. That allowed new companies to fill this void. In 2024, it is almost impossible to find a CEO who thinks AI is a bad idea.
这并不是说唯一的 AI 成功将是 “全新 ”类别。但通常情况下,在分销与创新的较量中,分销是默认的赢家 — 现有的软件公司拥有分销。尤其是现在,在 2024 年,AI 几乎是每家软件公司和最终客户的首要任务。在 2007 年,大多数 CEO 认为 iPhone 很愚蠢(没有键盘!),而黑莓更好。1996 年,大多数零售业 CEO 认为互联网是一种玩具,一种时尚,人们肯定不会从网络浏览器购买东西。这使得新公司能够填补这一空白。到 2024 年,几乎不可能找到一个认为 AI 是个坏主意的 CEO。
There will effectively be three kinds of origins and outcomes for AI software companies:
AI 软件公司实际上将有三种来源和结果:
- AI tools that run on top of existing software (think: automatic meeting notes for Zoom meetings)
在现有软件上运行的 AI 工具(想想:Zoom 会议的自动会议记录) - AI tools that run on top of existing software that have a shot of displacing that existing software (think: meeting notes for Zoom Meetings…where said company then builds video conferencing and pitches you to ditch Zoom)
在现有软件上运行的 AI 工具有可能取代现有软件(想想:Zoom Meetings 的会议记录......然后,该公司建立视频会议并向您推销放弃 Zoom) - AI tools that turn into labor — a net-new category, completely untouched by software until this point (think: the software conducts the meeting for you!)
变成劳动力的 AI 工具——一个全新的类别,在此之前完全不受软件的影响(想想看:软件会为你主持会议!
Platform shifts have always enabled the first two (internet versions of X, mobile versions of X, cloud versions of X). But what is most exciting about the AI revolution is that the enterprise software market — which looks big at $300B/year in spend — is infinitesimal compared to the white-collar labor market, at many, many trillions of dollars a year. It’s why many of the fastest-growing companies we are seeing have been the “known unknowns” of taking existing expensive services, and delivering low-price point products (created by AI) to the masses.
平台转移始终支持前两个版本(X 的互联网版本、X 的移动版本、X 的云版本)。但人工智能革命最令人兴奋的是,企业软件市场——每年的支出为 300B 美元——与每年数万亿美元的白领劳动力市场相比是微不足道的。这就是为什么我们看到的许多增长最快的公司都是“已知的未知数”,他们利用现有的昂贵服务,并向大众提供低价位产品(由 AI 创建)。

And we are in the very first inning of software eating – and augmenting – labor.
我们正处于软件吞噬和增加劳动力的第一阶段。




被折叠的 条评论
为什么被折叠?



