Volver a PodcastsEvery
We Automated Everything With AI and Tripled Our Headcount
You prompt AI to do something, it blows your mind, you feel inadequate, you feel like,"Oh my god, this thing is going to take my job."
你用 AI 做了某件事,它让你大开眼界,你感到自我怀疑,觉得「天啊,这东西要抢我的工作了。」
And then it stops working and it looks back at you and says,"What should I do next?"
然后它停下来,回头看着你,问:「接下来该怎么办?」
The further away an agent gets from a human, the less valuable it is.
智能体离人越远,它的价值就越低。
If you just ride the models, you're going to be fine.
只要跟着模型走,你就没问题。
If you care about leading a really ambitious life, I truly think that this is going to make that more possible for more people.
如果你想过一种真正有抱负的生活,我真心觉得这会让更多人离这个目标更近。
Every is the only subscription you need to stay at the edge of AI.
想跟上 AI 前沿,只需订阅 Every。
If you care about being on top of the latest models and using latest tools, you have to subscribe to Every to separate out the signal from the noise.
如果你关心最新模型和工具,就必须订阅 Every,帮你从噪音中筛出真正有价值的信号。
Go to every.to/subscribe today.
今天就去 every.to/subscribe 订阅。
So, we are here uh because we're going to flip the script a little bit.
好,我们今天来做个反转。
I am going to be interviewing Dan Sick.
我来采访 Dan Sick。
uh about the piece that he published yesterday, May 21st.
聊聊他昨天发布的那篇文章,5 月 21 日。
Uh and we're going to try to understand why he wrote it, what's underneath his reasoning for it.
我们会尝试弄清楚他为什么写这篇文章,以及他背后的推理逻辑。
There's going to be some conflict.
会有一些交锋。
I'm going to I'm going to fight with him on it.
我要跟他掰一掰。
Let's fight.
开打吧。
and see, you know, bringing some of my opinions, which are more or less aligned, but uh trying to understand does this is this piece going to reflect the future in 10 years, in 5 years?
看看,带上我自己的一些看法,这些看法大致是认同的,但我想搞清楚,这篇文章描绘的未来,10 年后、5 年后会成真吗?
And who are you again?
你又是谁?
Um I'm Brandon.
我是 Brandon。
I'm our COO.
我们的 COO。
And that's it.
就这样。
So, the piece is called After
这篇文章叫「自动化之后」
And it it comes from this feeling that I have.
它来自我的一种感受。
And there's a video about this, and there's there's a piece, but just for people who who have not seen either of those things.
关于这个有个视频,也有这篇文章,但对于没看过其中任何一个的人,
It comes from this feeling that I have that at Every, we are as AI native, as agent native, as it as you as it gets, you know, if you swing a stick around in our Slack, you're you're as likely to hit a human as you are an
它来自我的一种感受:在 Every,我们是最 AI 原生、最智能体原生的公司,在我们的 Slack 里随手一指,碰到人类和碰到智能体的概率是一样的。
[snorts]
[哼笑]
Everyone's using Claude Code and codex and all these tools to do their job every day.
每个人每天都在用 Claude Code、Codex 这些工具完成工作。
Um and yet it feels like there's more human work to do than ever.
然而感觉上,需要人来做的事情比以往任何时候都多。
And in fact like since the GPT-3 days like we've grown from four people to like 30 people and we're hiring more now.
事实上,从 GPT-3 时代到现在,我们的团队从 4 个人增长到了 30 个人,而且还在继续招人。
And so it came from me looking at that and then looking at the environment and being like what's going on because the whole information environment if you look at Dario is out there saying like half of entry-level white-collar jobs may be wiped out.
于是我看着这一切,再看看外部环境,就会想:这到底是怎么回事?因为整个信息环境里,Dario 在外面说,初级白领工作有一半可能被淘汰。
Even even people like um like Ken Griffin from Citadel is like you can tell he just had this moment where someone showed him an AI doing like an advanced data or finance question and he was like holy [ __ ] like that's what I would pay PhDs to do for me and it just did it.
就连像 Ken Griffin 这样的人,来自 Citadel,你能看出他刚有了一个顿悟时刻:有人给他演示了 AI 解答一道高级数据或金融问题,他的反应是「我草,这就是我以前花大价钱雇博士做的事,它直接就做到了。」
And I feel like I'm watching a lot of people who maybe don't have a ton of experience with agents and don't have a ton of experience with the curve of improvement that we've been riding for the last like three three and a half years hit it for the first time and then come to all these conclusions about
我感觉,很多没怎么接触过智能体、也没跟着这几年改进曲线一路走过来的人,第一次看到这个东西,就开始下各种结论,说
oh my god like all work is going away we're not going to have jobs and I'm just like sitting here being like actually your intuitions when you first see a technology like this are usually very
「天啊,所有工作都要消失了,我们不会有工作了」,而我坐在这里心想,你刚接触一项新技术时的直觉,往往是非常
And we've seen a lot over and over again over the years that
这些年我们已经看到过很多次这种规律,
Every is a very good bellwether for where things are going because it's a it's a group of early adopters we have people in doing all sorts of work internally at Every and if something works here it there's a good bet that it's going to spread to other other places other other businesses that are that are adjacent to ours.
Every 是判断趋势走向的一个很好的风向标,因为这里聚集了一群早期采用者,我们内部有各种各样的人在做各类工作,如果某件事在这里行得通,基本上就会扩散到与我们相邻的其他地方和其他业务。
And so when I look around at Every I I see so much automation and I also see way more human work.
所以当我环顾 Every,我看到的是大量自动化,同时也看到了更多的人类工作。
So I was really um the the whole piece is saying here's the current state of work with agents.
这整篇文章想说的就是:这是当前智能体工作的现状。
And then pulling apart that paradox and sort of explaining why does why does more automation mean more work?
然后把这个悖论拆解开来,解释为什么自动化越多,工作反而越多。
Yeah, when I read the piece it was there wasn't like an explicit call to action in it, but I sort of felt this call to action of like there is actually a massive amount of hope right now in a world that is filled with a lot of doomers.
对,读完这篇文章,我没感受到什么明显的行动号召,但我有点隐隐感受到一种号召:在这个充斥着悲观主义者的世界里,现在其实有巨大的希望。
And um and this is why.
这就是原因。
Um
嗯
[snorts]
[哼笑]
but I am going to come out of the gate and ask you a devil's advocate question.
但我要一上来就问你一个反方问题。
Which is a couple hours before you publish this piece the CEO of ClickUp came out with this long tweet about why he fired 8,000 people and 3,000 people some
就在你发这篇文章的几个小时前,ClickUp 的 CEO 发了一条长推,说他为什么裁掉了 8,000 人、3,000 人,某
What?
什么?
I don't I don't think it was 8,000.
我不觉得是 8,000 人。
I was 20,000 people.
我说的是 20,000 人。
What I mean
我的意思是
[laughter]
[笑声]
I think it was like 3,000.
我记得大概是 3,000 人。
an entire economy.
整个经济体。
like 22% of of his workforce.
大概是他员工总数的 22%。
I don't think it was in the thousands, but yes, it was it was a
我不觉得是几千人,但确实是
it was a lot of his workforce.
确实是很大比例的员工。
Yeah.
对。
Yeah.
对。
So my question to you is um in a business like every we're growing super fast.
所以我想问你:像 Every 这样高速增长的公司,
Um what you wrote makes a lot of sense to me.
你写的东西对我来说很有道理。
And what you wrote theoretically makes a ton of sense in that AI is not autonomous right now.
你在文章里写的,从理论上讲非常说得通:AI 现在还不是自主的。
It has to be told what to do and then has to be checked.
它需要被告知该做什么,然后还要被检查。
We need to have that that sandwich that you described in in the piece.
我们需要你在文章里描述的那个三明治结构。
But in a business that is 8,000 people, 10,000 people that is mature and has built ways of managing like SOPs for managing their business does this manifesto and this thesis still hold true?
但对于一家 8,000 人、10,000 人、已经成熟、有一套用 SOP 管理业务方式的公司,这份宣言和这个论点还成立吗?
It's a
这是个
It's a really good question.
这是个非常好的问题。
Um
嗯
Are there a couple
有几个
There are a couple different questions here.
这里有几个不同的问题。
The first thing I want to do is like lay out the argument.
我想先把论点摆出来。
Why does Why does automation make more work?
为什么自动化会创造更多工作?
I'm sure many people listening to this also haven't read it.
我猜听众里很多人也还没读过这篇文章。
So take a second to explain that in detail.
花点时间详细解释一下。
I will do that.
好,我来说。
So basically, the idea is the way that AI works and the way it functions in the workplace is AI makes yesterday's expert competence cheap.
基本上,这个想法是:AI 在工作场所的运作方式是,AI 让昨天的专家能力变得廉价。
And by that I mean AI is trained on all of our outputs, all of the code and the writing and the design and and decision-making and everything that's ever been written.
我的意思是,AI 是基于我们所有的输出训练的,所有曾经写下来的代码、文章、设计、决策,一切。
And it makes that available to everyone for very cheap.
然后以极低的成本把这些能力提供给所有人。
So uh you can
所以,你现在可以
So anyone now with a prompt can use yesterday's competence to solve a programming problem, build an app, or write a uh write a piece like I did, write a report, or uh or, you know, make a YouTube thumbnail.
任何人现在只需一个提示词,就能用昨天的能力来解决编程问题、构建应用、写一篇像我写的那样的文章、写一份报告,或者做个 YouTube 缩略图。
And the interesting thing is that when you do that when when expert competence is available for cheap, it gets really widely adopted.
有趣的是,当这件事发生,当专家能力可以以低价获取,它会被广泛采用。
So everyone starts to do it.
于是每个人都开始这样做。
Everyone starts to like, you know, we see this internal
每个人都开始,你知道,我们看到这种内部
Everyone's making pull requests
每个人都在提 pull request
[ __ ]
[__]
this is crazy.
这太疯狂了。
Yeah.
对。
And and and like I'm making pull requests and ops people are making pull requests and you know, engineers are like writing essays and you know, there's all this line crossing basically for non-experts to do the thing that experts used to do.
而且,我在提 pull request,运营的人也在提 pull request,工程师们在写文章,基本上就是大量的越界,非专家开始做专家以前做的事。
And that feels very threatening to experts.
这对专家来说感觉很有威胁。
They're like,"Well, what's my job going to be now?"
他们会想:「那我的工作是什么?」
Mhm.
嗯。
And [snorts] what's interesting about that is because these tools are trained on outputs, are trained on yesterday's data the stuff that they do is uh with with default prompt with is it uh the stuff that they do with a default prompt all looks kind of similar and is all kind of right for the current situation, but it's like not actually totally right.
而有趣的是,因为这些工具是基于输出训练的,是基于昨天的数据训练的,它们用默认提示词做出来的东西,看起来都大同小异,对当前情况来说也大致正确,但实际上又不完全正确。
And so what happens is you sort of like flood the zone with tons of stuff that's like close, but not quite right.
于是发生的事情是:你用大量这种接近但不完全正确的东西把市场淹了。
And then you need to basically like and and well
然后你基本上需要,好
There's a
有很多
There's a lot of that at every too.
在 Every 也是这样。
There's a lot of
有很多
There's a lot of people doing what seems like great work and then you go under the hot under the hood and you're like, this isn't quite right.
有很多人做的事情看起来很好,但你深入进去一看,会发现,这不太对啊。
Maybe like the expert should do it.
可能应该让专家来做。
Yeah.
对。
Yeah.
对。
Yeah.
对。
Exactly.
没错。
Uh me for example.
比如我。