$160B Market Cap. $5.48B Revenue. $10M EBITDA Per Head.

A lot of the things that we've been able to accomplish just don't make sense to people.

我们做成的很多事,对外人来说就是讲不通。

And in a world where things don't make sense, people think you're cheating.

在一个讲不通的世界里,人们的第一反应是——你在作弊。

The founder mentality has got to be chase winning.

创始人心态的核心,就是去追胜利。

In order for me to get paid anything, the stock had to clear that and then keep going up from there.

我要拿到一分钱报酬,股价必须先冲过那条门槛线,然后还得继续往上走。

Almost in every relationship of my life, I was never really present.

我这辈子几乎所有的关系里,我都没有真正"在场"过。

That fear of blowup is one of my big motivators.

"怕崩盘"这件事,是我最大的动力之一。

Now I have interviewed a thousand CEOs of the largest companies over the last 10 years.

过去十年我采访过上千位顶级公司的 CEO。

This guest Adam Foroughi is top five I've ever met.

这位嘉宾 Adam Foroughi,稳稳进我心中的前五。

Easily there is no company on the planet with numbers like AppLovin.

毫无悬念——地球上没有第二家公司能跑出 AppLovin 这种数字。

Of all the shows that I've done genuinely in the studio, this is the favorite one for me that I've ever done with a CEO.

所有我在 studio 里录过的节目里,这是我跟 CEO 做过的最满意的一集。

Adam, I'm so excited for this, dude. It's funny — I sent you the schedule beforehand, you're like "that's a lot of questions."

Adam,我太期待这一集了。挺好笑的——我提前把提纲发给你,你回我说"问题真不少"。

I stalked the [hell] out of you before this, just to be clear. So, thank you for agreeing to this onslaught of questions.

先说清楚:我录之前把你各种公开信息扒了个遍。所以,谢谢你愿意接受这一连串炮轰式提问。

Yeah, I like it. I try to go unscripted, so I can't say I reviewed them, but it was a lot of questions.

没事,我喜欢这种。我做事讲究不踩剧本,所以提纲我没认真看,但确实问题不少。

Well, don't worry. Reviewing them is always a way to have a manufactured conversation. So this is going to be completely unscripted.

别担心。提前对答案,出来的就是预制对话。这一集我们彻底不照本子走。

One thing that I'm always just trying to understand before we dive in is like the mentality of an entrepreneur.

每次开场我都想先搞懂一件事——创业者的心态。

There's two types of people. People that are motivated by losing or people that are motivated by winning.

人分两种:一种被"怕输"驱动,另一种被"想赢"驱动。

What are you — fearful of losing or inspired by the thrill of winning?

你属于哪种——怕输,还是被赢的快感点燃?

I think you almost — if you've had success, you almost have to be inspired by winning.

这么说吧——一旦你尝过赢的滋味,你几乎必然是被"想赢"驱动的。

If you're fearful of losing or you have a fear of failure, I feel like you're almost certain to be stuck.

如果你被"怕输"或者"怕失败"主导,那你几乎一定会卡住。

You're not going to take shots that are material, and you're going to protect downside more than go after upside.

你不会下注重的赌注,你会更花精力保住下行,而不是去争上行。

And I don't tend to believe that's really the founder mentality.

我不认为那是真正的 founder mentality。

If you took a risk once upon a time to start a business where there was nothing, you didn't even know what it was going to become, and you knew the odds were 99% likely that you were going to fail —

你当初冒着险从零起步、连这事会变成什么都不知道、心里明明白白知道有 99% 概率失败——

that in itself has to tell you the founder mentality's got to be chase winning.

光这一点就足以告诉你,创始人心态的本质就是去追"赢"。

And so over the years I've taken motivation through winning.

所以这些年我的动力来源一直是"想赢"。

And I think it's also important to note that founders don't tend to be motivated by money as well, if they're really successful.

还有一点值得说——真正成功的创始人,几乎都不是被钱驱动的。

That's something I like to ask in interview questions, and I found the best people are motivated by personal growth, development, being inspired, finding things intellectually stimulating, winning — but it never tends to be money.

这是我面试时爱问的问题。我观察下来,最厉害的人被驱动的东西是:个人成长、被启发、智力上的刺激、赢的感觉——从来不是钱。

Because money is a very tough thing to continuously be motivated by.

因为"钱"这件事,作为长期动力实在太难维系。

Eventually you will reach a point where money is no longer a motivator and then you need to find something else.

迟早你会到一个钱不再是动力的点,那时候你必须找到别的东西支撑你继续。

So, I've always pushed to win and I've always pushed to learn and grow, and those are the things that really got me going.

所以我一直在追赢,也一直在追学习和成长——这些才是真正让我跑起来的发动机。

Can I ask you — how does that change how you operate as a CEO?

我能问一下——这种心态,会怎么改变你做 CEO 的方式?

There was a baseline that I needed to feel like family was good, and I was fortunate enough to start a couple of businesses before they were successful.

我需要先确保家里这条底线是稳的——很幸运,我在做 AppLovin 之前就已经做过几家公司、做出过成绩。

So, I'd reached the baseline before I started this business.

所以做 AppLovin 之前,我那条"基本盘"已经够了。

And I said those businesses, I really aspired to get a single.

之前那几家公司,我的目标其实就是打个 single(一垒安打)。

I just wanted to get enough money where I didn't have to stress about money.

我只想赚到一笔够花的钱,让自己别再为钱焦虑。

Once I co-founded this business with my team and we started getting going, I never really needed anything from this monetarily.

真正跟团队一起把 AppLovin 做起来之后,我对它在"钱"上面就没什么期待了。

And so the interesting piece there is — as we were building up, we were growing really quickly.

有意思的地方就在这——我们当时一直在快速增长。

And in 2015, we got approached to sell the business for quite a lot of money — in the hundreds of millions of dollars, all cash.

2015 年有人找我们谈收购,价格不低——几亿美金,全现金。

Had I not had the singles before, it might have been something that was very enticing to just cash out the whole thing at that point in time.

如果我之前没打过那几个 single,那笔钱可能真的很诱人,当场套现走人也合理。

But because I knew my bank account was sound, I wasn't in it for money.

但我知道我账户里钱够花,我做这件事根本不是为了钱。

I was trying to build big — and really, I felt like this had to be the home run.

我是想把这件事做大——我心里很清楚,这一票必须是 home run(全垒打)。

I was able to think about the deal process there logically and understand that the business is growing really well. It's really sound.

所以我能冷静地看那笔收购的细节——我们业务增长很好,基本面非常扎实。

Why would we give it up on that upward trajectory?

我凭什么在它正在往上走的时候把它卖掉?

So, we were able to really play long — and I think in large part that's because I didn't start this at all considering the money I could make from it.

我们才能真正打长线——很大程度上是因为我创业一开始就没把"能赚多少钱"放在脑子里。

Speaking of not starting for the money — your total comp in 2023 was $83 million, making you the eighth highest paid CEO in America.

说到不是为钱创业——你 2023 年总薪酬 8300 万美金,是全美第八高薪的 CEO。

How do you think about that? What do people not see when they read headlines like that? What is the misconception?

你怎么看这件事?看到这种标题的人都漏掉了什么?他们误解的是什么?

To understand my comp in '23, you've got to really look backwards.

要理解我 2023 年的薪酬,你必须往回看。

When we went public — we went public in '21 — in the first year, the stock went up to about $40 billion market cap.

我们 2021 年上市。上市第一年,市值冲到了大约 400 亿美金。

In 2022, we fell about 92% to a little bit under $4 billion market cap.

2022 年我们跌了大约 92%,市值跌到不到 40 亿美金。

I, for the life of the company, had only taken equity that was my founder stock and based on the money that I originally put in the company.

公司这一路走来,我手里拿到的股权只有创始人股,以及最初我投进公司的那笔钱对应的那部分。

So I'd taken no compensation. I was taking basically the bare minimum to have benefits.

所以我从来没拿过薪酬。我只领最低限度——刚够走个福利流程那种。

At the bottom in '22, I made a decision — and that was, for the first time, to ask for compensation.

2022 年触底的时候,我做了一个决定——第一次主动开口要薪酬。

And the reason I did that is because I felt like — I'm public, turning this company around is a big task.

原因是——我们已经是上市公司,要把它扭转回来是个巨大的任务。

And I'd like to align myself with investors to say: I'm going to get paid, but I'm only going to get paid if the stock recovers.

我想跟投资人站在一条船上,告诉他们:我会拿钱,但前提是股价得真的恢复。

So the thresholds of compensation that the comp committee on the board granted me — at a minimum the stock was $9.

所以董事会薪酬委员会给我设的门槛是这样:底部的时候股价大概 9 美金。

We had to get to the first threshold — I think was about $38 to $40.

第一道门槛大概是 38 到 40 美金。

And in order for me to get paid anything, the stock had to clear that and then keep going up from there for me to get any sort of compensation.

我要拿到任何一分钱,股价必须先冲过那道门槛,然后还得继续往上走,我才能领到任何薪酬。

And then there were, I think, five or six levels from there.

在那之上还有大概五到六个递进的门槛。

So all the way up to a return to $80, which was our IPO price — and I had a term to go achieve it.

一路到 80 美金——也就是我们当初的 IPO 发行价——而且我必须在限定的时间内做到。

I feel like CEOs who are founders originally started a business taking really big risk.

我觉得身为创始人的 CEO,本来就是冒着巨大风险开始这件事的。

If the belief is that the CEO should then never get compensation ever again, it's completely flawed logic.

如果大家的逻辑是"那 CEO 之后就该永远不拿薪酬"——这个逻辑彻底是错的。

You want to give people who chase really big upside by creating really big things the potential to continuously have that really big upside.

你想让那些靠"做出大事"换大回报的人,继续保有"还能拿到大回报"的可能性。

Because it allows them to, in a way, just mentally stay motivated on what they're doing versus start drifting to other things.

因为这才能让他们在心智上继续盯在这件事上,而不是开始走神去搞别的。

Because any founder — even at the low point, I was worth quite a bit of money on paper at least with my equity.

毕竟任何一个创始人——哪怕最低点的时候,我账面上的股权也还是值不少钱。

I could have walked away and started something else.

我完全可以拍拍屁股走人,再去做点别的。

I've now had three successful businesses, so I believe I could have a fourth — but I really wanted to stay committed to the company and stay aligned with investors.

我现在已经有三家成功的公司了,我相信我能做出第四家——但我真的想留下来,跟投资人继续站在一条船上。

I think the other thing that people miss is that the CEO's job in a company, especially one that's gone from small to very, very large, is an incredibly lonely, very stressful role.

还有一件事人们没看到——CEO 这份工作,尤其是把公司从小做到非常大的那种 CEO,极其孤独,极其紧绷。

And if you talk to CEOs and founders, I know you do fairly often — these jobs are brutal.

如果你去跟 CEO 和创始人聊聊——你应该聊得不少——这种工作是地狱级的。

People these days are afraid to talk to me because I'm so busy. They perceive that, even though I'm not.

这两年人们都不敢跟我说话,觉得我太忙。但其实并没有——是他们的感知。

I'm the same person I was 10 years ago. They don't come up to me anymore.

我还是十年前那个我。但他们就是不再主动凑上来了。

Or people inside the company or outside the company — if the stock is doing well, believe you're smarter than you are.

公司内外的人——股票涨的时候,他们觉得你比实际更聪明。

And if the stock is doing poorly, believe you're going to be so stressed out you might jump off a building.

股票跌的时候,他们又觉得你压力大到要从楼上跳下去。

You don't understand the things that are going on in the CEO's mind when you really haven't done that role yourself.

你没真正当过 CEO,就不可能理解 CEO 脑子里在过的那些事。

And very few people in the world have built a business from small to very large and eventually taken it public.

这世上真正把一家公司从小做到非常大、最后还推上市的人,数得过来。

So I say that to say — it's a brutal job. It's lonely. It's stressful.

我说这些是想强调——这是个残酷的工作。孤独,紧绷。

You almost certainly are going to have distraction from personal life.

你几乎一定会被工作搅乱个人生活。

I don't know many founders who have had that kind of success that have fantastic personal lives.

我认识的取得过这种量级成功的创始人里,没几个个人生活过得特别好的。

Yet you end up distracted from your kids because you're always focused on work. It consumes you.

你最后会从孩子身上分神,因为你永远在想工作。它会把你吞掉。

And therefore to then say the CEO should take less pay is something that's quite unjust, because it's not understanding the role that the CEO has to absorb.

所以你再去说"CEO 应该拿少点钱",这个判断其实挺不公平——因为它根本没在理解 CEO 这份工作真正在承担的东西。

Have you ever questioned the sacrifice in any way?

你有没有质疑过自己付出的这些代价?

My brother has children, and I watch him have children and be an amazing parent and see my parents be grandparents.

我哥有孩子,我看着他带孩子,做一个非常棒的爸爸,也看着我父母当爷爷奶奶。

Dude, I'm just [expletive] grinding in the office till 11, 12 every night.

而我自己呢——每天晚上 11 点、12 点还在办公室里死磕。

A little bit of me questions the sacrifice sometimes, if I'm honest.

说实话,我有时候真的会问自己:这值吗?

I would say at the low point in '22, I decided to make some changes because I did question the sacrifice I was making.

2022 年最低点的时候,我开始做一些改变——因为那时候我确实开始质疑自己付出的代价。

There were two tolls that I saw that were being taken on my life.

我看到生活上有两笔账正在被扣走。

One was my health was decaying — and I felt like, if I'm this stressed out, I'm not sleeping, I'm drinking eight cups of coffee a day.

第一个是健康在崩——这种压力下我睡不着,一天喝八杯咖啡。

I'm losing my hair, losing my fitness, just losing the things that allow me to focus.

头发在掉、体能在掉,所有让我能保持专注的东西都在往下掉。

If I don't reverse that, I'm never going to be in a place to be mentally sound to run the business.

如果我不把它扭过来,我永远都不会在一个心智健全的状态去经营公司。

And I felt like as a CEO of a public company, I'm committing for the next 10, 20 years.

作为一家上市公司的 CEO,我承诺的是接下来 10 年、20 年。

I need to be here a long time. To do that, I needed my health.

我需要长期在场。要做到这一点,健康必须先回来。

So, I stopped what I was doing and reset that.

所以我停下手头的方式,把这部分重置了。

The other piece was — I felt like from my children, I'd drifted a little bit more distant because I wasn't paying attention.

第二笔账是——我跟孩子的距离正在变远,因为我没在用心陪他们。

And I think any founder or anyone who works at a tech company that's always on knows this experience, if they have kids.

任何一个创始人,或者在一家"永远在线"的科技公司工作的人,只要有孩子,都懂这种感受。

You hang out with your kids, but your mind is elsewhere.

你陪着孩子,但脑子在别的地方。

Either that, or you're on your phone. And so, you're never really present.

要不就是在看手机。反正你从来没真正在场过。

And what I realized is — almost in every relationship in my life, I was never really present.

我后来意识到——我这辈子几乎每段关系里,我都没真正在场过。

And so, what I tried to do to change that is at least take small moments to feel like I was present.

为了改变这一点,我开始至少给自己挤出一些"真的在场"的小片段。

Small moments might be 10 minutes at a time, because I'm not going to be able to sit down and have hours at a time.

所谓"小片段"可能就是一次 10 分钟——因为我没法连续坐下来好几个小时。

But if I gave myself 10 minutes at a time to hang out with one of my children or a couple of my children, I felt like — okay, now I'm actually committing to them to be 100% present.

但只要我能给自己 10 分钟去陪一个、或者两个孩子,我心里就觉得——好,这 10 分钟我真的 100% 在场。

And that was a change.

这就是一个变化。

And the third change I made at that point in time was that I started introducing hobbies to myself.

那段时间我做的第三个改变,是开始给自己引入兴趣爱好。

So for example, I started learning how to surf in the last year or two years.

比如最近一两年我开始学冲浪。

You have to put the phone down. You have to be completely disconnected. You get mental ease.

你必须把手机放下。你必须彻底断网。然后大脑会松下来。

And in the absence of these changes, I feel like I would have felt like I'm giving away a big part of my own ability to be stable and happy.

如果没做这些改变,我会觉得我把自己保持稳定和快乐的能力,大半都拱手让出去了。

And by getting that back, I became a better CEO of the business.

把这些拿回来之后,我反而成了一个更好的 CEO。

I became someone who could be more thoughtful and who could be more long-term focused.

我变成了一个更深思熟虑、更能盯长线的人。

You said in 2022, you fell — you kind of dropped it in very casually. You fell, I mean, 92%.

你刚才轻描淡写地讲到 2022 年你们跌了——跌了 92%。

What does no one know about — respectfully — being in a trough that deep, that they should know?

没有亲历过那种深坑的人,你觉得他们最该知道但其实不知道的是什么?

To fall 92% in a year — you go down almost every single day of the year.

一年跌 92%——意味着这一年里几乎每一天股价都在跌。

Neil M once told me — what's the difference between being down 98% and 99%? Half.

Neil M 跟我说过一句话——跌 98% 和跌 99% 之间差多少?差一半。

Yeah, it's a lot. And when you realize you've fallen 92%, you've got to go up 10x to get back to where you started.

是,那一半很大。一旦你意识到自己跌了 92%,要回到起点,股价必须翻 10 倍。

So it's a bloodbath.

所以那真的就是一场血洗。

A couple things — one is, a lot of people think your psyche is tied to the stock. And beyond that, when you talk to some execs at public companies, they'll say they don't look at the stock price.

有几件事——第一,很多人觉得 CEO 的心理状态是跟股价绑死的。你跟某些上市公司的高管聊,他们会说"我不看股价"。

I can say I 100% look at the stock price.

我可以告诉你,我 100% 在看股价。

It is very hard to run a public company and say I'm just going to choose not to look at the stock price for a few days, a week, whatever — because you've got investors that care, you've got your team that cares, and it's a real-time ticker on what the world thinks of your business.

作为一家上市公司的 CEO,你说"我就几天、一个礼拜不看股价"——非常难做到。投资人在意,你团队在意,而且股价就是一个实时显示"世界怎么看你这家公司"的滚动条。

The challenges are — when everyone is telling you that the stock is going down every single day, investors are not buying your shares.

真正的挑战是——所有人天天告诉你股价又跌了,投资人不买你的股票。

It's very easy to go: am I doing something wrong?

这时候你很容易问自己:我是不是哪里做错了?

Is the business [screwed]? Is something here that I don't understand, that everyone else in the world is smarter than me on and understands?

这家公司是不是完蛋了?是不是有什么我没看明白、而全世界都比我先看明白的事?

And so it can make you lose confidence and second-guess yourself.

这种感受会让你失去信心,开始反复怀疑自己。

In the face of that, I think what's important is — if you believe in your business — maintaining conviction.

面对这一切,关键是——如果你相信这家公司,就守住那份 conviction(坚定)。

And so we did a couple things, and really that allowed us to turn the business around.

我们做了几件事,真正让业务转过了头。

As an advertising business, there's an advertising model that drives a lot of the success that we have on platform.

作为一家广告公司,我们平台上的成功大部分是被一个广告模型驱动的。

Everything we do is on a performance basis.

我们做的每一件事都是 performance-based(按效果付费)的。

So advertisers plug in, they aim to get a certain amount of revenue that's more than the ad dollars that they spend on the platform.

广告主接进来,目标是从平台上拿到的收入,要超过他们花在平台上的广告预算。

Now what delivers that equation for them is how potent our advertising model is.

能不能跑通这个等式,取决于我们的广告模型有多强。

And these models are recommendation system models — that's one of the earlier forms of machine learning that existed, and it's really gotten supercharged with what we see today in AI and the research advancements in LLMs.

这些模型本质是推荐系统模型——它是机器学习里出现得比较早的一类,现在被今天的 AI 和 LLM 研究突破狠狠加了一波。

In '22 at the very bottom, we said — we're on an older version of machine learning. We're going to completely throw out our technology, rebuild it, and go to what is really cutting edge and current in the field of recommendation systems.

2022 年最底部的时候,我们说——我们跑的是老一代机器学习。我们要把现有技术整个推翻、重建,直接切到推荐系统领域目前最前沿的版本。

To do that was a big internal change.

这件事在公司内部是一次大变化。

One, we had to slow down basically all research and development on the current system because we said we're going to throw it out. It's now outdated.

第一,我们必须几乎停掉所有针对老系统的研发,因为我们已经决定要扔掉它——它已经过时了。

It's not going to carry us forward where we've got to go.

它带不动我们要去的地方。

We had to turn over some people.

我们必须换掉一些人。

We had to turn over some people that had helped us get to that point — which, again, was a $30 billion IPO and up from there, and then cratered.

必须换掉一些把我们带到当时那个量级的人——别忘了,那个量级是 300 亿市值的 IPO,然后再被狠狠摔回来。

But a big business — we had to turn over some people who were committed to the old system and just say: the old system's done. We're rebuilding to something new.

公司规模虽然不小,我们还是得让一批已经跟老系统绑死的人离开,直接告诉大家:老系统结束了,我们要重新建一个新的。

And then we had to have conviction behind that bet, and rally everyone at the company that this is the right thing to do and we're going to go execute on it and win.

然后我们必须押注、扛住信念,把全公司召集起来说:这是对的事,我们就这么干,我们要赢。

How did you literally do that?

你具体是怎么做到的?

You have to voice confidence in your own bet.

你必须公开表达对自己这个判断的信心。

And so it's very hard to walk around confident when your stock's down that much.

当然,股价跌成那样,要做到"自信地走来走去"非常难。

People are calling you thinking you're suicidal.

那段时间有人打电话来,是觉得你可能想自杀的那种。

So you've got to drown out that noise.

所以你必须把这种噪音屏蔽掉。

People literally check that you're doing okay — yeah, you almost get the look like, you should probably go consult a therapist because it looks like you're going to kill yourself.

真的有人来确认"你还好吗"——那种眼神基本就是"你该去看个心理医生,你这状态像随时要做傻事"。

And I wasn't giving off those vibes — because at least I didn't think so.

但我自己其实没散发那种气场——至少我自己不这么觉得。

Because I've always had a belief that so long as we have conviction on a path and we've got a strategy that sounds right and we've got a motivated team behind it, we're good to go.

因为我一直信一个东西——只要方向上 conviction 够、策略听上去对、团队还有动力,这件事就能跑。

And so I was able to voice confidence internally.

所以我对内部能真正表达出信心。

In doing so, we were able to retain core team and the important people that we needed to go execute on this path forward.

这样我们才留住了核心团队、留住了执行这条路所必需的关键人才。

That's really the challenge you get into when the stock falls that much.

这就是股价跌到这种程度时,你真正面对的挑战。

It's really hard to understand — how can you retain people, when people are working and seeing the exact same thing we're talking about, and they're probably (and their families are probably) going: is this company a piece of [crap]? Why aren't investors buying the shares?

很难想象——人怎么留下来?因为员工跟你看到的是同一件事,他们(和他们的家人)很可能在想:这家公司是不是就是个垃圾?为什么没人买它的股票?

Well, it's easy to get tricked into believing it is, when it goes down 92%.

是啊——跌了 92% 的时候,你很容易自己骗自己,真以为公司就是垃圾。

That is a very material shift in terms of technology architecture, which leads to the layoffs.

技术架构上的这个大转向,很自然就引出了后面的裁员。

We're seeing a huge amount of layoffs today. Are those layoffs, do you think, due to AI efficiency, or do you think it's because of overhiring in COVID times?

我们现在看到大规模裁员。你觉得这一波是因为 AI 提效,还是 COVID 期间过度招聘的宿醉?

I think it's the latter today, because the former is still yet to take full effect at most companies.

我觉得现在这一波主要是后者——因为 AI 提效在大多数公司还没真正起效。

But a couple years ago, we'd been growing really fast ever since we launched this model Axon 2 in April of 2023, and the stock recovered.

但往回看几年——我们 2023 年 4 月发布 Axon 2 之后一直在快速增长,股价也回来了。

It was mostly in '24 — we had a year where we probably grew near triple digits, but we ended up cutting the team's staff by 40-50% in most departments.

主要是 2024 年——那一年我们增长接近三位数百分点,但我们在大部分部门把团队规模砍了 40 到 50%。

The reason I did that is a belief that if the role was going to get automated, or that AI was not being adopted fast enough in those departments, it's time to let those people go and rebuild the organization as if we were building it knowing what technologies were available to us today.

我那么做是因为一个判断:如果一个岗位会被自动化、或者那个部门 AI 应用速度不够快——那就是该让这批人离开、并按"今天的技术存在,我会怎么搭这个组织"重新搭一遍的时候。

What roles did you assume at that time were going to get automated — if we deconstruct "going to" and then "not fast enough"?

能拆开讲讲吗——你当时判断哪些岗位"会被自动化"、哪些是"AI 应用不够快"?

First of all — over time, companies get bloated.

首先一件事——公司只要存在久了,就会变臃肿。

So I said: what are the process-oriented organizations? What got created even in our company, which runs really lean?

所以我问自己:哪些部门是"流程导向"的?哪怕我们这种本来就很精简的公司,也长出了什么?

We've got a really high revenue per employee and EBITDA per employee — but even at that time, we'd gotten bloated over a decade-plus.

我们人均营收和人均 EBITDA 都很高——但即便如此,十几年下来我们也变臃肿了。

So I looked at first: what are the process-enabling parts of the organization?

我先去看:组织里哪些部分是支撑"流程"运转的?

One was HR. HR as a function was necessary because you've got to be able to hire and fire people, but our team had gotten bloated and there was a lot of process the HR team was introducing in the organization.

第一个是 HR。HR 这个职能本身是必要的——你得能招人、得能裁人——但我们 HR 团队变臃肿了,而且它在公司里堆出了一大堆流程。

As a founder, I still remember the days we were 10, 20, 50, 100 people — and you didn't have that much process. You had one HR person per 100 to 200 people, and things felt faster.

作为创始人,我还记得我们 10 人、20 人、50 人、100 人那几个阶段——根本没那么多流程。一个 HR 顶 100-200 人,公司跑得明显更快。

So I went through and said: what are the processes I don't like at the company? Let me just eliminate those.

所以我一项一项过——哪些流程我不喜欢?直接砍掉。

Then we can go through and say: who are the gatekeepers of those processes? You can remove those people.

然后再问:那些流程的"守门人"是谁?这些人可以让他们离开。

And then you go to the areas that you're going to start seeing a lot more automation.

接下来再看——哪些领域接下来会迎来大量自动化。

So an example in our business is creative production.

一个例子是我们业务里的"创意生产"。

We felt like AI is going to get to the point where creatives are going to be automatically produced.

我们判断 AI 会发展到广告创意可以自动生成的程度。

You still need humans to innovate, but you can have less humans because a lot of the design work can be handed off.

你仍然需要人来做创新,但人可以更少——因为大量设计工作可以交给 AI。

In engineering, your best engineers can use these tools to really accelerate themselves, and your weaker engineers might not understand how to use these tools, or might only get a 2x instead of a 10x or 100x increase in output.

工程方面也一样——最强的工程师用这些工具能极大加速自己,弱的工程师可能不会用,或者只能提 2 倍效率,而不是 10 倍或 100 倍。

How do you think about the fear of moving before the market's ready?

你怎么看"在市场还没准备好之前就动手"这件事的风险?

There's a lot of promise, but you can fire before the creative tools are there to deliver what was needed.

前景很美好,但你可能在创意生成工具真正成熟之前,就先把人裁了。

Your earlier question about winning ties in here — I don't play in fear of failure or fear of losing.

你刚才问"为赢而做",这件事正好挂在一起——我不靠"怕失败、怕输"驱动自己。

And I also believe on my team across the board, our job for them has to be the right job for them at this moment — and "right" defined by the best place to have personal development and growth.

我对全公司团队也是这个标准——他们在我们这里的工作,必须是"此时此刻最适合他们的工作",而"最适合"的定义是:对个人成长和发展最有帮助。

And if we believe that every single person has a good role — if we think that's no longer true, we should part ways, good severance, and make sure they're free to go do something else.

如果有一天我们觉得某个人的岗位已经不再"对"了,我们就分开走,给体面的遣散包,让他自由去做别的。

Because I don't like to keep people in roles that are going towards a dead end.

因为我不愿意把人留在死胡同里。

Now, it was a bet and a belief that these technologies were going to get good enough to automate these roles away — but we didn't want to take the risk that we were going to keep people in dead-end roles.

这当然是一个押注,赌这些技术会成熟到足以把这些岗位自动化掉——但我们不愿意冒"把人留在死胡同里"的风险。

That just creates morale hit, that creates an organization that ends up optimizing to people who are just not happy.

那种状态只会拖垮士气,让整个组织围着"不开心的人"优化。

And we try to optimize to our best performers.

我们想围着最强的人优化。

Best performers — your A-players — want free rein to just go crush it.

最强的人——你们说的 A-player——要的是"放我去打,别拦着"。

But they don't want to be distracted by unhappiness.

但他们不愿意被身边那些不开心的同事分神。

They don't want to be distracted by people who are working at a role that is almost certainly going to get automated away.

他们也不愿意被那些"几乎注定会被自动化"的岗位上的人拖着。

And so by taking it and saying: build the culture as if we were building it today knowing what technologies are available to us — what would we look like?

所以我们的做法就是问:如果今天从零开始搭一个公司,以我们知道的技术现状,这个公司应该长什么样?

We just went to "what we would look like" — and then that forcing function made us have to get to an automated place faster.

我们直接照那个答案搭——这就形成了一个 forcing function,逼着我们更快走向自动化。

It would have been a lot slower had we had people that were trying to fight adoption of the technologies because they were fearful it was going to lead to their job loss.

如果我们留下一批因为怕丢工作而抵触新技术的人,这件事的速度会慢得多。

Do you think it is possible to have a company of your scale — 895 people with $4.3B revenue ahead — only A-players?

你们这种规模——895 人、年营收 43 亿——你觉得可能做到"全是 A-player"吗?

Is there a time when, by nature, you just have to have a native B-player?

还是说,公司到一定规模后,本来就必须有一批"原生 B-player"?

So our core business — we bought a couple of businesses. We have Adjust, which is an analytics company, and Wurl, which is a CTV business. Those two aren't integrated. They run their own business.

我们核心业务——我们收购过几家公司。Adjust 是个分析公司,Wurl 是个 CTV 业务。这两家没并进来,各跑各的。

If I just went to the core business, our core advertising products are about 400 people.

如果只看核心业务,我们核心广告产品就大约 400 人。

So call it some very high percentage of all the company's EBITDA comes from the core business.

公司绝大部分 EBITDA 来自这个核心业务。

If you then calculate the EBITDA per employee over 400, it's a really high number — I think it's reaching or over $10 million a head now.

用 400 人去算 EBITDA per employee,这个数字非常吓人——我记得现在已经达到甚至超过人均 1000 万美金。

So the question — can you have a team full of A-players? Not everyone can be an A-player in a team. You need some roles that are just there to be processed and keep the lights on.

所以你的问题——能不能整支团队全是 A-player?不行,团队里不可能人人都是 A-player。你必须有一些岗位就是处理流程、维持基本运转的。

We're a public company, so there's certain things that have to happen, just because they have to happen.

我们是上市公司,有些事就是必须做——单纯因为合规要求。

But like HR, for example — that I touched on earlier — we took a pretty large HR organization, one that I think had 70-80 people on it, and now might have 15.

但比如刚才说的 HR——我们当时 HR 大概有 70-80 人,现在大概只剩 15 人。

The people that we retained are your A-players in HR. They're the doers, the individual contributors. They just get stuff done. They don't get bogged down in process.

留下的这 15 人,是 HR 里的 A-player。他们是"干活的"——是 individual contributor。他们直接把事办了,不会陷在流程里。

And so every organization we said: how do we slim down to the best people?

所以每个部门我们都问:怎么瘦身到只剩最好的那批人?

And for us, the best people are defined by those that really want to come in and make a difference and learn and develop themselves — but not need process to get there.

"最好"的定义对我们来说是:真心想做出影响、想学、想成长——而且不需要靠流程才能做到这些。

No management layer, no slowdown. It's just people who just want to get [stuff] done.

没有管理层、没有拖速度的东西。就是那群只想把事办了的人。

And so that went through the entire organization, where we leaned up to just those kinds of people.

这套逻辑跑遍了整个公司,我们瘦下来后只剩这种人。

And then you start looking around the room, and you've got great people everywhere — then you enjoy working at that company.

然后你在公司转一圈,看见的全是厉害的人——你就会真心享受在这里上班。

A-players, what I've learned, can exist whether it is a back office role or an engineering role or a front-lines revenue-generating role — but A-players won't exist in bulk if you have a bunch of B's, C's, and D's around them.

我学到的是——A-player 可以存在于任何岗位,后台、工程、一线创收都行。但如果他们身边围着一堆 B、C、D,A-player 就成不了气候。

Can I ask — what role do you dislike most, but you have to keep?

能问一下吗——你最不喜欢、但又不得不留的岗位是哪个?

So long as I have all people who are doers, who are really high output, I don't dislike any role — because it fits our culture and every part of that comes together to build the business.

只要所有人都是"干活型"的高产出 IC,我没什么不喜欢的岗位——因为这种文化里,每个部分都贡献于业务本身。

But as an example — if you look at our exec team, we have CEO, CTO, CFO, and General Counsel. We don't have a CRO. We don't have a COO.

举个例子——我们的高管团队只有 CEO、CTO、CFO 和 General Counsel。没有 CRO,没有 COO。

Go down the list of other C-levels people might have — we don't have a CMO. We don't have a Chief People Officer.

别的公司常见的那些 C-level,我们都没有——没有 CMO,没有首席人力官。

None of these people. What's the point?

这些全都没有。我们没看到它们存在的必要。

I'm shorting the company. If you don't have a Chief Human Resources Officer, what are you doing?

那我现在就要去做空你们公司了。一家公司没 CHRO,你怎么活下去?

They manage the next person, then the next person, then the next person — to eventually the doer.

CHRO 管下面的人,下面的人再管下面的人,一层层往下传——最后才到那个真正干活的人。

The reason I state this is because we really built a culture of doers.

我之所以强调这个,是因为我们真的搭出了一个"干活者文化"。

It is very hard when you grow up — to go from a team you started, that was small, that was a team of doers, to eventually get large and go public.

公司一旦做大、最后还上了市——你想保留当初那个"小而能打的干活团队"的状态,极其难。

It's very hard to maintain that. We didn't, until we ran through the layoffs and started leaning up.

非常难维持。我们之前也没保住——是经过这次裁员、瘦身,才把它找回来。

And really, the catalyst for me was this guy who's now the CTO — Giovanni — came in and he started looking around the organization and kept saying: why do we have these people? Why do we have these processes?

真正的催化剂是现在我们 CTO Giovanni——他来了之后,到处转一圈,反复问:为什么我们有这些人?为什么我们有这些流程?

And it reminded me that the most important question to ask in business is "why."

他提醒了我——做生意最重要的问题就是"为什么"。

And so he inspired me to go: it's been 10 years, we're working with all these people, we have all these processes that we built over 10 years — why do we have these things?

他启发了我:做了十年了,我们身边这些人、这十年里搭出来的这些流程——为什么会有它们?

Why is it that I have this person who has this title who means nothing?

为什么我手下有这么个人,挂着一个根本不代表什么的头衔?

And I went through the whole organization. We just went back to the founding roots and tried to go back to that culture of doers.

于是我把整个组织过了一遍。我们回到创业最初的根,试图恢复那种"干活者文化"。

And the question "why" played a huge role in that, and we were able to get to a place where everything was leaned up to doers.

这一波"为什么"问出来,起了巨大作用——最后我们把所有部门都瘦身到只剩干活的人。

I'm a CEO listening, and I want to have a culture of doers and a culture of execution like you have. What are the biggest mistakes you see other CEOs make who want this culture but don't have it?

假设我是一个正在听的 CEO,我也想要你说的这种"干活者文化、执行文化"。你看到那些想要这种文化但没做到的 CEO,最大的错误是什么?

I think it's really hard if the train leaves the station and your team becomes bloated to go backwards.

一旦火车开出站、团队已经变臃肿了,想回头就非常难。

And the reason I say that — this is a challenge in software today — is it's not as simple as "go lay off 50-60% of people."

原因是——这也是今天软件行业普遍的难题——它不是简单地"裁掉 50-60% 就行"。

If the team is bloated and there's a mixture of A's, B's, and C's — your A's are probably already long gone, and what's left is like A-minuses to B-pluses and then down from there.

如果团队已经臃肿、A/B/C 混着——A-player 早就走了,留下的是 A-、B+,以及更往下的。

But it's people who like working in a process-oriented bigger company that are sticking around.

真正留下来的,是那些喜欢在"流程化的大公司"里工作的人。

If you go fire 50% of people and the culture and the team is mediocre, you're left with half mediocrity — and you're not going to get to where we hopefully are at, which is just a bunch of A-players who are doers.

这时候你裁 50%,剩下的还是一半平庸——你不可能跑到我们这种状态:一堆 A-player + 全是干活的。

The only way to fix a culture like that is to go and fire 99% of people and just rebuild it from the ground up. It's exceptionally hard to do.

修复这种文化的唯一办法,是把 99% 的人都裁掉,从地基重建。这件事极其难做。

Not a lot of people understand how to do that, because they don't know what they're looking for. And it's very, very hard to do that as a public company.

大多数人不知道怎么做——因为他们不知道自己要找的"对的样子"是什么。而且作为一家上市公司,这件事更难做。

People hear that this is the way to build things, and founders remember the glory days of 50 people in a room just building stuff and things moving incredibly quickly.

大家听说这种打法,创始人也会想起当年 50 个人挤一屋、什么事都飞快的好日子。

It's not particularly easy to take a company that's gotten large scale with a bunch of layers, big exec suite, and then take it back down.

但你要把一家已经做大、堆满层级、高管林立的公司,缩回去——非常不容易。

Will the layoffs that we are seeing not result in the desired improvements from the CEOs who are making them?

那我们现在看到的这一波大裁员,会不会达不到这些 CEO 想要的效果?

I think if they really know what they're doing and they understand how the company looked when it was highly efficient when it was founded, then it's plausible that it can get back to the roots.

如果他们真的知道自己在干嘛、还记得公司创立时高效的样子是什么,那是有可能回到根本的。

But if it's a company that's gotten bloated to the point of mediocrity, and it's just "let's fire half and try to automate roles," it's probably not going to get to the place that people think it should.

但如果这家公司已经臃肿到平庸的程度,光靠"裁一半、想办法自动化"——很可能去不到大家期待的那个状态。

We are seeing a deluge of SBC — stock-based compensation — at a level that we almost haven't ever seen before, I don't think, in corporate history.

我们正在看到一波 SBC(股权激励)海啸,量级大概是商业史上没见过的。

How do you feel and think about that?

你怎么看这件事?

So we've given roughly the same amount of stock every year in terms of absolute amount — and it's roughly $300 million.

我们每年发出去的股权,绝对量上一直差不多——大概 3 亿美金。

If you think about our market cap — I think our market cap is about $150 billion — our burn on stock-based comp is very low.

我们市值大约 1500 亿美金,3 亿的 SBC,稀释比例非常低。

And so you can judge us on cash flow minus SBC, which I generally think is the right way to judge companies.

所以你可以用"现金流减去 SBC"来评估我们——我认为这才是评估公司的正确方式。

What's happened in tech, though, is that there's been an expectation that stock-based comp will be high at companies.

但科技行业的现状是——大家默认公司就该发很多 SBC。

As stock prices have gone down, especially in software companies of late, you have a downward spiral that's formed.

而股价一旦下跌——尤其最近这些软件公司——就会形成一个死亡螺旋。

All of a sudden a company is burning 3% of their cap table every single year to pay out equity to the team, falls 66%, and now you're at 10%.

一家公司本来每年发出 3% 的股权激励,股价跌 66% 后,稀释比例瞬间变 10%。

And you're at a level of dilution that it's incredibly hard to come out from underneath.

到了这个稀释级别,几乎不可能再爬出来。

And so it makes it hard to bet on those companies when they're burning that much equity.

这种烧股权的公司,你很难真的押注。

What I found, and what we implemented in '22 when we fell a lot, is that certain people have enough compensation to not take risk on the stock if the stock's going to be volatile.

我们 2022 年大跌时改了一件事——有一部分员工的固定薪酬已经够多了,股价波动他们其实承担得起风险;另一部分人不能。

We used to believe that every single person should have equity granted by the company.

我们以前的逻辑是——每个员工都该拿到公司发的股权。

Instead, we went to a place where we said: the top 10 to 15% of the company will get equity, and the rest won't. They'll have the right to buy equity, and there's ESPP programs that let employees buy equity at a discount if they so choose. Otherwise, they'll just be paid on cash.

后来我们改成:公司前 10-15% 的员工拿股权,其他人不拿。其他人有"自愿购买股权"的权利,我们有 ESPP 让员工以折扣价买股票。不想买的,就拿纯现金薪酬。

I remember when I first started my career, I couldn't have taken risk. I was basically going paycheck to paycheck, right?

我刚工作那会儿,我没法承担股价风险——我是月光族,基本上靠每月那张薪水单过日子。

If you got 25% of your pay in stock and it went up, great. You feel great. But if it falls 92%, you're like — damn, I can't pay my rent.

你拿到 25% 的股票工资,股票涨了——爽。但跌了 92%,你的反应就是:糟糕,房租付不起了。

That's a real problem. So we took it to a point where people who had the luxury of being able to take upside got upside. Everyone else got cash comp. They had the decision themselves, and we controlled this burn.

这是个真问题。所以我们改成:负担得起赌上行的人,才拿股权上行。其他人拿现金。员工自己决定,而我们也控制住了稀释的烧速。

And so we got into the position where it just wasn't burdensome to our business.

这样一来,SBC 就再也不是我们业务的负担了。

And I think companies tend to give away their stock too cheaply and too broadly, not understanding who actually can drive the value of the equity — and also believing that the investors are going to be accepting of really high burn rates.

大多数公司发股权时,既廉价又广撒——他们没想清楚到底谁能真正撬动这股权的价值,也错以为投资人会接受很高的稀释速度。

Why do you believe cash flow minus SBC is the right way to value companies?

为什么你觉得"现金流减 SBC"是评估公司的正确方法?

I think cash is king. It's simple.

现金为王。逻辑简单。

I look at accounting practices, I look at EBITDA numbers, and what's clean EBITDA versus not. End of the day, net income, cash — these are clean things.

我看会计处理、看 EBITDA、看"干净 EBITDA"和"不干净的 EBITDA"。但归根到底,净利润、现金——这些是最干净的指标。

If a company generates a billion dollars of cash but gives out a billion dollars of equity and says "I'm just going to buy my equity at a billion dollars" — they're not generating any cash.

如果一家公司一年挣 10 亿现金,但同时发出 10 亿股权,然后说"我用 10 亿现金回购掉那些股权"——它其实根本没产生任何现金。

So what's the real value of that business? Either you're diluting and they're paying all of the cash they generate to buy the equity back to offset the dilution, or they're building up a cash balance that just offsets the dilution.

那这家公司真正的价值是什么?要么是稀释 + 用挣到的现金回购对冲;要么是攒着现金留着以后对冲——本质上都没创造价值。

So what's the point of believing that the cash flow is real in that case?

那种情况下,你为什么还要假装"这家公司有真正的现金流"?

So for me, it's just distill businesses down to the simplest metric, which is cash flow minus SBC.

对我来说,把生意蒸馏到最简单的指标——就是"现金流 - SBC"。

We mentioned the creative changes that are happening with AI and how that impacts output. Engineering is one that you mentioned several times. How have you seen engineering productivity change with AI in the last year or two?

我们聊了 AI 给创意带来的变化。你也几次提到工程——过去一两年,你看到工程的生产力被 AI 改变成什么样了?

Databricks, I think, released yesterday that 50% of their code is generated by AI.

Databricks 昨天好像公布了:他们 50% 的代码是 AI 写的。

Ours is a higher percentage than that — 80-90% probably — but that discounts quality over quantity.

我们比这个高,大概 80-90%——但这个数字本身把"质量 vs 数量"这件事压扁了。

If you just shoot off for a percentage of tokens consumed, you could get to a place where you're just creating slop.

如果你只追"token 消耗占比",你很容易跑到"全是 slop(垃圾输出)"的境地。

If you're incentivizing slop, you're not going to get very far as a business. You're going to have massive fees to go pay the large language model businesses, but you're not going to get further as a business.

激励 slop 的公司走不远——你会给 LLM 厂商付一大笔钱,公司本身却没真正前进。

What's important is — are your engineers good enough to use these technologies to accelerate what creates value for the company? And can you measure that?

真正重要的问题是——你的工程师够不够好,能用这些技术加速"真正给公司创造价值的事"?而且你能不能度量出来?

It's great to deploy an army of agents to go do my work for me — but if it's unclear what the deliverable is, and it's unclear that deliverable is aligned with actual growth in the business, then it's just waste.

部署一支 agent 大军帮你干活——听上去爽。但如果交付物不明确、又不确定这交付物跟业务增长挂钩,那就是浪费。

And so it's easy to say "look, percentage of code is high" — because truly, if you set off the agents to start writing code, they're going to contribute more code than humans.

所以说"AI 写的代码占比很高"——这话说出来容易。你只要把 agent 放出去写,它写的代码量一定超过人类。

But is there value created?

但有没有产生价值?

And so everything we do with a lean team is try to go to value creation.

我们这种精简团队的逻辑,就是一切围绕"价值创造"。

If you optimize to that, you get the most out of the agents without looking at the superficial metrics — more so trying to distill it to: was your investment in tokens covered by the amount of revenue that you created from the code?

围绕价值优化,你才能真正榨出 agent 的价值,而不是盯着表面指标看。更准的问法是:你烧的那些 token,有没有被它写出来的代码所带来的营收覆盖掉?

What does it mean to move to value creation? Like, how do you do that?

"围绕价值创造优化"具体怎么做?

You have to understand the KPIs of the business that drive the business.

你必须真正搞清楚:这家公司里,哪几个 KPI 是真正驱动业务的。

Our organization was built pretty nicely for this — for the era that we're in.

我们的组织,对这个时代来讲,搭得正好。

One, we don't have a product organization. Our engineers are meant to be product managers.

第一,我们没有"产品部门"。我们的工程师本身就是产品经理。

And if you think about what's happening with AI-native engineers today — they have to be really imaginative. They have to be product people.

今天的 AI-native 工程师必须很有想象力。他们必须是"产品人"。

They don't have to know how to write code, but they have to be able to audit code — because frankly, you can't just go type out what you need in a complex system and get a deliverable and it's done.

他们可以不会自己写代码,但必须会审代码——因为说实话,在复杂系统里,你不可能一句话敲进去就拿到一个完整可交付的东西。

They still need to be able to review the code and make sure what they're checking in is safe and high quality.

他们还得能审代码、确保提交进去的东西安全、质量过关。

But first and foremost, they need to know what the business needs and they need to know how to measure it.

但首要前提是:他们必须知道业务需要什么、必须知道怎么度量它。

And so our business — with a lean team, and one where when you push a model improvement, it is with certainty that it's easy to see it reflected in accuracy numbers in the model and also revenue growth in the business.

我们这个业务,因为团队精简,而且只要推一个模型改进——你会在模型准确度数字和公司营收增长上,都立刻看到。

The team knows what the KPIs are that they're optimizing to.

团队清楚地知道自己在优化什么 KPI。

And because they know that, they can then align with what an agent or army of agents is going to do on their behalf, and try to get to that point of getting the most value out from the investment that we're making.

正因为他们清楚,他们才能把 agent(或一支 agent 大军)对齐到自己的目标上,把这笔投入榨出最大价值。

I think it's very hard in a lot of businesses to understand exactly what are the KPIs that we're optimizing to.

很多公司其实根本说不清,自己究竟在优化什么 KPI。

So they just go: let's just write a bunch of things and see what sticks. Then you're walking on a slippery slope.

于是他们就"先写一堆东西看哪些有用"——这就走上一个滑坡。

You may have so much cost ballooning from token usage that you don't actually get the type of revenue growth you need to cover it.

最后你可能 token 成本爆表,但根本没拿到能覆盖这成本的营收增长。

Can you talk to me about when you optimized for a KPI that turned out to be wrong, and what you learned from that?

能不能讲一次你们围着一个错的 KPI 优化、最后发现错了的经历?学到了什么?

Our business is pretty simple. I don't know that we ever optimized to something that turned out to be wrong, because we've always optimized to the same thing.

我们业务挺简单。我不太记得有哪次我们围着错的目标优化过——因为我们一直就是优化同一件事。

There are two things that drive our business.

有两件事在驱动我们的生意。

If the model is more accurately predictive, it's going to drive more revenue for the customer — the advertiser — than their media cost spent. And everything is measurable in our system.

第一,如果模型预测更准,广告主从中获得的收入就会超过他们投的广告费用——而且我们系统里这一切都能精确度量。

If that function holds true, revenue should grow as well alongside it.

这条函数成立,公司营收就会同步增长。

And so because everything is real-time tracked, and because we've always had a very consistent business model — where we don't sell the belief that something worked, we sell the actual fact that something worked, and we can measure everything —

所有动作都是实时追踪的。而且我们的商业模式一直一致——我们不卖"我们觉得这件事有效",我们卖"这件事真的有效",并且每一步都能度量。

we ended up in a lucky spot where the business was built really well to be able to utilize the types of technologies that we're seeing out there.

所以我们运气不错——业务本身的结构,恰好特别适合接住外面出现的这些新技术。

You said multiple times about it's very easy to have massive spend on the LLMs and just blunt AI slop being created. How did you think about the decision whether to invest in your own model — as Harvey did, as Cursor did, TBD on how that goes — versus use existing frontier models?

你提了很多次:很容易在 LLM 上烧大钱、最后只产出一堆 AI slop。你怎么决定的——是自己训自己的模型(像 Harvey、Cursor 现在做的,效果还不好说),还是直接用现成的 frontier 模型?

Look, we're not an interface on top of large language models. There's usage of large language models in the company for productivity. There's some usage of large language models in our core business as well.

我们不是"建在 LLM 之上的一个 interface"。公司内部确实用 LLM 来提效;核心业务里也用到一些 LLM。

But a recommendation system model is something that drives engagement — what you see on content on a social network. It's something that drives most advertising products in the world today: Facebook's ad system, TikTok's ad system, ours.

但"推荐系统模型"是另一类——它决定你在社交网络上看到什么内容,决定全世界大部分广告系统的运转:Facebook、TikTok 的广告系统,我们的广告系统,都是。

This is a space of machine learning that really hit its stride about a decade ago, and I would say really accelerated with some of the research that we've seen come out of the large language model space lately.

推荐系统这块,十几年前就开始进入成熟期了——而且最近 LLM 领域的研究又给它狠狠加速了一波。

But it's a space where you can't just go defer to the large language model and say "based on what you know about this user and the data I have available, what's the next ad to see?" — that wouldn't work as well as a custom model built for this purpose.

但你不能直接把广告决策推给 LLM,问它"按你了解的用户 + 我手里这点数据,下一条广告投什么?"——那个效果远不如专门训的模型。

In a world where you get to a place where you're in a category where you're utilizing the large language model or you're building an interface on top, you better build a moat really, really fast — given how exceptionally talented companies like Anthropic are about releasing product on top of their own models.

如果你做的事就是"用 LLM"或"在它之上做 interface"——你最好以非常非常快的速度建起一条护城河。因为像 Anthropic 这样的公司,自己往自家模型上叠产品的能力,极其强。

Do you think the majority of companies we see created today will be commoditized, eaten by Anthropic and OpenAI and frontier models?

你觉得今天创立的大部分公司,会被 Anthropic、OpenAI 和 frontier 模型吃掉、商品化掉吗?

I would be very nervous if I was building a business as an interface on top of those.

如果我是在 frontier 模型之上做 interface 的公司,我会非常紧张。

What does your team use internally engineering-wise? Cursor or Claude Code?

你们内部工程师主要用什么——Cursor 还是 Claude Code?

Most people are on Claude Code. Codex is utilized as well, and Cursor less so these days.

大部分人用 Claude Code。Codex 也用。Cursor 最近用得少了。

You have 895 people today. How many people will AppLovin be?

你们现在 895 人。未来 AppLovin 会有多少人?

It's tough to say. Again, it's 400 people in the core business. We run lean.

不好说。核心业务是 400 人。我们一直跑得精简。

Is it going to be 800 on the core business? I highly doubt it. Is it going to be 50 on the core business? I'd love it, but I highly doubt it too.

核心业务会变成 800 人?我严重怀疑。会变成 50 人?我很想,但也不太可能。

So, I think we're sort of in a range of a good level for what we need for what we're doing today.

所以我觉得现在这个规模,对我们今天做的事来说,大体在合适的区间。

Now, if some of the things that we'll take bets on over time work, we'll need more people around other businesses. But if we're just executing on our core business, it's very likely we don't need to go hire a whole lot more.

将来如果我们押注的新业务跑通,围绕新业务会需要更多人。但只要还在跑核心业务,我们大概率不需要再大幅扩招。

You said about a kind of execution team of doers — it sounds great, but it's very hard to do, and you need great (god, I sound like a real corporate) alignment. But you don't do one-on-one meetings. How do you do the culture of execution without one-on-one meetings and without the traditional corporate scaffolding?

你说的那种"执行型干活者团队"——听起来很美,但很难做到,需要很强的所谓"对齐"(说出这词我自己都觉得在念企业话术)。可是你又不做 1-on-1——那这种执行文化是怎么在没有 1-on-1、没有传统企业脚手架的情况下跑起来的?

One of my beliefs is that really good people figure out a way. They don't need a whole lot of mentorship.

我的一个底层判断是——真正厉害的人会自己想办法。他们不需要太多 mentorship。

So if people on my team — if they directly report to me — I never do one-on-ones. I don't do reviews.

所以我团队里所有直接向我汇报的人——我从来不做 1-on-1,我也不做绩效评审。

If I don't like something they're doing, they know about it in real time via chat. If I like what they're doing, they don't need to know — they know that I respect them and they're good to go.

他们做的事我不喜欢,实时通过聊天告诉他们。我喜欢的,他们不需要知道——他们清楚我尊重他们,他们就能继续干。

Good people don't need that type of handholding usually.

真正厉害的人,通常不需要那种"手把手带"。

And what ends up happening is people who need a lot of development — those people aren't the people I want on this team of A-players.

结果就是——那些需要被大量培养的人,并不是我希望留在这支 A-player 团队里的人。

And so we tend to shy away from a lot of traditional management techniques.

所以我们刻意远离大量传统管理手法。

Another example of this is something like learning and development. A lot of companies try to structure all the onboarding and learning and development processes — "you're new at my company, here's how you should learn the business."

再举一个例子——L&D(学习与发展)。很多公司把入职、学习、发展全套流程化:"你是新人,这是你该怎么学公司的"。

I remember in school I hated classes that were structured. I didn't learn anything. You couldn't retain it. I wanted to learn as I went.

我读书时最讨厌那种结构化课程——什么都没学到、记不住。我喜欢边做边学。

My first couple jobs out of school, I came in and I was just curious and I figured stuff out.

我毕业后的头几份工作,我就是带着好奇心进去,自己摸索出来的。

I've seen a pattern that our best people come in, they ask questions, they figure things out.

我发现一个规律——我们最强的人,进来后就是问问题、自己搞清楚。

And so we don't really have formal learning and development, and it's completely disconnected from what you would expect at a company.

所以我们没有正式的 L&D 体系——跟一般公司你想象的完全不一样。

We don't want to structure people. We want to get really curious minds who come in, who are loud enough to get what they need to get, and who can learn.

我们不想"结构化"人。我们想要那种自带好奇心、声音够大能拿到自己需要的东西、能自学的人。

Now I'm going to tie it to the AI-native world today. The benefit of not doing things in these one-on-one silos and very structured is — you can document everything in Slack or transcribed video calls.

把这件事接到今天的 AI-native 世界——不搞 1-on-1 这种封闭式管理的最大好处是,所有东西都沉淀在 Slack 或转录过的视频会议里。

If you do that, any new person can come in and go: "Hey, Claude, summarize for me what Adam cares about over the last quarter and write me a book of everything that matters to him. And take the person who's running the best sales calls and summarize what they do, and tell me what I should know on this job."

这样的话,任何新人进来都能直接问:"嘿 Claude,把 Adam 这个季度在意的事总结给我,给我写一本'他在乎什么';再把销售里跑得最好的那个人的电话总结给我,告诉我这份工作我该掌握什么。"

You start asking these types of questions, you start getting really good output — because all of the information is available to Claude — and you end up getting a person who can actually develop themselves through curiosity and output out of the models.

你开始问这种问题,就能拿到非常好的输出——因为所有信息都对 Claude 开放——最终你养出一个真正能靠好奇心 + 模型输出自己成长的人。

And that is a much more capable future employee than someone who was just told "here's what you need to know."

这种员工,远比那种"我告诉你应该学什么"训练出来的员工更有未来。

You said about transcribed video calls having a lot of quality data that can be used, summarized — that's great. I believe in in-person strongly. How do you think about in-person versus remote and the value?

你说视频会议转录数据可以拿来用、做总结——这点很好。但我很相信面对面。你怎么看面对面 vs 远程的价值?

End of the day, we're a sales business talking to advertisers. So I do believe there's a lot of value to building relationships in person.

归根到底,我们是个跟广告主打交道的销售生意。所以面对面建关系,确实有大量价值。

I think you do have a loss in ability to feed that information into the model and show other people what you're doing in those in-person moments.

面对面的代价是——这部分信息很难塞回模型、也很难让别人看到你在那一刻在做什么。

So what we tend to do is believe the vast majority of communication needs to be written or through a video call. And when you need to build a relationship with key clients, you go in person and you take them out.

所以我们的默认是——绝大多数沟通必须以书面或视频会议形式留下;真要跟关键客户建关系的时候,飞过去当面带他们出去。

If you take them out in a social gathering, you can send notes into a chat around that client and have that as your history on the in-person meeting.

出去社交完,你可以把笔记发到那个客户的群里——这就成了你这次线下会议的留痕。

But you can't replace in person.

但面对面不可替代。

As human beings, as we go to this world where bots are going to do more for us, in person's even more important.

作为人——在 bot 越来越多替我们干活的世界里——面对面只会更重要。

I'm similar to you in terms of a focus on execution, and I get told that not everything has to be productive — sometimes being deliberately unproductive is almost productivity. À la team drinks. I don't want to do team drinks on a Friday at 5:30.

我跟你一样,执行导向。但总有人跟我说"不是所有事都得'有产出',刻意'不产出'本身也算产出"。比如团队聚餐喝酒。我才不想周五 5:30 跟团队喝什么酒。

Can we not bond over a whiteboard in a project we're working on? We all love what we're doing. Like, can we not do that? Why do we have to go and sit and drink in a pub? But I'm told that that's productive culture.

我们一起对着白板做项目不就够建感情了吗?我们都爱手上的活儿,为什么非要坐到酒吧里喝酒?但所有人都告诉我"那才叫高效文化"。

What I found is — in the most productive moments with your best people, you get in heated debates, like yelling matches.

我观察到——跟最强的人合作时,最有产出的时刻,经常是激烈的争论、甚至是吼起来。

If you get really heated with someone and you go right back to "let's just crank" — and there's no moments where you go out to dinner, you have drinks, you get to bond — you sometimes lose the human side of things, and sometimes resentment can build, and then things can become unproductive.

如果跟人吵完之后马上就回到"接着干"——中间没有一起吃个饭、喝个酒、建感情的间隙——你会丢掉关系里属于人的那一面,慢慢积出怨气,反而开始降速。

When you remember that you're just a bunch of smart people in a room trying to figure stuff out, and you really remember that at something like a dinner or drinks, you end up creating productivity out of those moments.

当你在饭桌或酒局上重新意识到"我们就是一群想把事搞明白的聪明人"——那一刻反而会变成有产出的时刻。

The other thing I found is when we go out and we drink and we start shooting the [shit], really good ideas can come of that too.

还有一个发现是——一起出去喝酒乱聊的时候,真的会冒出非常好的点子。

It's not that we're going out with a bunch of co-workers and talking about baseball. We're going out with a bunch of co-workers and getting drunk together and talking about work opportunities.

不是说我们出去聊棒球。我们是出去喝醉,然后聊工作上的机会。

Sometimes your best ideas come out of those moments.

最好的点子,有时候就是在那种时刻冒出来的。

You also don't attend conferences. Why don't you attend conferences? How do you think about that?

你也不参加会议(行业大会)。为什么?

That's not true anymore. I do go to conferences now.

这话现在不成立了。我现在参加会议。

When we fell in 2022 — one of the things we did on the investor relations side, you fall 92%, no one's buying your stock — we said we're going to buy our own shares and we're going to shut down investor relations. Cuz what's the point?

2022 年大跌时,在投资者关系这一块——你都跌 92% 了,没人买你的股票——我们说:我们自己回购,投资者关系暂停。因为意义何在?

Why do I need to go to a conference to explain to everyone who's selling my shares to buy my shares?

我为啥要去会议上,跟那些天天在卖我股票的人解释为什么他们应该买回去?

You're not going to convince someone to buy your shares when they're convinced every day you're going down.

在他们坚信你每天都在跌的时候,你说什么都不会让他们买。

So I just said — a better use of my time is focusing internal and focusing on the long term. And it's a bad use of my time to go to conferences.

所以我那时说——我的时间花在内部、花在长期上更好。去开会就是浪费时间。

As a public company CEO, you are supposed to go to conferences. You're supposed to meet with investors. So for a period of a couple years there, '22 and '23, we basically just shut all that down.

作为一家上市公司 CEO,你"应该"去会议、应该见投资人。但我们 2022、2023 那两年,基本上把这些全停了。

Eventually, when the stock started gaining traction and the market cap was really recovering, I realized those were key parts of the role.

等股价开始反弹、市值真正回来,我才意识到——这些其实是 CEO 工作的关键组成部分。

And I like to challenge myself and do things well, even if they're uncomfortable to me.

我喜欢挑战自己——哪怕事情让我不舒服,也要做好。

Here we're sitting and we're having a one-on-one conversation. This is relaxed. But going to a conference, speaking in front of a couple hundred people — I've always had a fear of public speaking. So I'm an introverted person that didn't want to put myself out there.

现在我们坐在这聊天——挺放松。但站在两三百号人面前讲话——我一直怕公众演讲。我是个内向的人,本能不想把自己摆出去。

But what I realized is — now that we're playing at higher stakes tables, the company's getting bigger, we need to be out there. We need to be conveying what we do so that people can understand the business model and the prospects of the business model.

但我意识到——我们现在玩的牌桌赌注更大、公司也更大了,我们必须站到前面。我们需要把"我们到底在做什么"讲清楚,让别人理解这个商业模式和它的前景。

So I started doing more conferences over the last couple years. I think they've been rewarding because it's challenged me to do something that's naturally uncomfortable for me.

所以最近两年我开始多去会议。挺有收获——因为它逼着我做一件天然让我不舒服的事。

What else, other than public speaking — I'm just intrigued — remains uncomfortable but you have to do it all the same?

除了演讲,还有什么对你一直不舒服、但你必须做的?

Maybe the only other thing that comes to mind right now is — it took me a long time to learn how to delegate. And this is something I actually committed to in that dark year, '22, as well.

现在想得起来的另一件事——我花了很久才学会授权(delegate)。这也是 2022 那个黑暗年我决定要改的事。

I was a very controlling, hands-on CEO for a very long time. I ran it. It was almost like all roads of the company went up to me.

我很长一段时间是个控制欲很强、亲力亲为的 CEO。我自己管。公司里所有路最后都通到我这里。

And when we fell, and I realized I'm not making great decisions for the business, I also realized other people are smarter at other aspects of the business than I am.

大跌之后,我意识到我没在为公司做最好的决策——同时也意识到,在某些业务环节,别人比我更懂。

So why am I not deferring to them? Why am I not delegating?

那我为什么不让位、不授权?

And so where I got to was — I started stripping away my own roles. Actually it was almost not that I was handing things off — it was that the rest of the team said, "I'm going to come in and just take these things away."

我开始一项项放下自己手里的角色。实际操作上更接近——不是我递出去,而是团队的人说"我来接管这一块"。

Again, this Giovanni I'll give another example — he just started taking the product role that I had run and owned for a decade at the company. He took it away from me. And it was great, because now I can ride side car. I can see what the team does, but I don't have to be in the weeds.

再举 Giovanni 的例子——产品这块我做了十年,他直接把它从我手里接走。这件事特别好——因为现在我可以"在边上跟车",看着团队在做什么,但不必自己陷在细节里。

And so it freed me up to do more strategic thought for the business long term. It freed me up to do more investor relations.

这样我就被解放出来,可以做更多长期战略思考,可以做更多投资者关系的事。

But it's very hard as a controlling founder-type business to have the founder go: "I'm going to hand things off."

但对一个控制欲强的 founder-CEO 来说,自己说出"我要放手了"——非常难。

So those two things, I think, were — one internal, one external — important for me to really see as flaws and try to grow and develop through.

所以这两件事——一件内部(授权),一件外部(演讲会议)——是我必须看清是自己的弱项,并且去突破的。

I think we've seen this prevailing trend of anti-delegation now — Paul Graham's "founder mode" and the importance of being in the weeds on certainly a number of things that traditionally would be delegated. How do you think about the power of delegation, the importance of delegation, with the rise of founder mode and founders being told "go back"?

现在有一股"反授权"的风——Paul Graham 的 Founder Mode、强调一些过去会被授权的事 founder 应该亲自下沉。在 Founder Mode 兴起、所有人都告诉 founder"回去亲自干"的语境下,你怎么看授权的力量?

The whole notion of founder mode is an extreme reaction to extreme bloat that got created in most Silicon Valley companies over the last decade.

Founder Mode 整个概念,是对过去十年硅谷大多数公司"过度臃肿"的一种极端反应。

If you're in a company with a bunch of layers and a bunch of process — how do you reverse it?

如果你在一家堆满层级和流程的公司,你怎么把它反向修过来?

We talked earlier about how a team of mediocrity, you can't reverse back to a team of high output.

我们刚才聊过——平庸的团队,你没法直接"逆转"成高产出团队。

And so in large part, the only way to reverse is to have a founder that takes control back.

大多数情况下,唯一能反转的方式,就是 founder 重新拿回控制权。

But once you get to that lean team of highly exceptional doers — if you're then controlling and not delegating, then what are you doing?

但一旦你已经回到"小而强的干活者团队"——这时候你还在控制、还不授权,你到底图啥?

You have a whole bunch of exceptional talent around you, who in theory, on their own roles in the business, are going to be more of a subject matter expert than one individual who runs the business can be. In that case, delegation is very powerful.

你身边一堆顶尖人才,理论上在他们各自的领域里,都比你这个跑整盘的人更懂。这种情况下,授权是非常强大的杠杆。

I have to ask — there was a day when the stock fell 23%. The caption here is "the short seller war." I'm just really intrigued. Short sellers have come after you multiple times now. Is there a flaw in the mechanics of the market?

我必须问——曾经有一天你们股价跌了 23%。我这里写的标签是"短空袭之战"。我非常好奇:做空机构反复来打你们。这件事是不是市场机制本身有缺陷?

First of all — when you go down 92% in a year, you sort of learn to take your beatings.

首先一件事——一年跌 92% 之后,你就学会了挨打。

I've gotten to the point — that was a massive blessing. Go public and immediately take that beating, you realize that the public markets are volatile.

那次大跌其实是一个巨大的"礼物"。上市后立马挨这种打,你会真正意识到——公开市场就是波动的。

There's things outside your control.

有些东西你根本控制不了。

The short seller attacks were not particularly surprising to me, because we went from a low point of $9 a share to a high point of $750 a share in two and a half years.

短空袭其实没让我太意外——因为我们在两年半时间里,从每股 9 美金涨到 750 美金。

That kind of a run-up from under $4 billion market cap to around $250 billion market cap — I don't know if any other company has ever seen that kind of value creation in that short amount of time in history.

这种涨幅——市值从不到 40 亿涨到 2500 亿左右——历史上我不确定有没有别的公司在这么短时间里完成过。

And then you looked at the companies that were at our market cap — all names that people would recognize — and then goofy-named AppLovin.

你再看跟我们同市值梯队的公司——全是大家耳熟能详的名字,然后突然冒出一个名字奇怪的 AppLovin。

Nobody knows what advertising businesses do, let alone the goofy name.

没人真懂广告生意是怎么回事——更别提我们这个奇怪名字。

And so we sort of expected it, because we weren't out there promoting ourselves. We were just executing the business.

所以我们大致预料到了——我们没出去做营销,我们只是在埋头跑业务。

Because we grew so quickly — both revenue, profit, and stock price — we sort of failed to put our story out there proactively.

因为我们涨太快——营收、利润、股价同时涨——我们没主动把"我们的故事"讲出去。

And then therefore we were sitting ducks for people that wanted to create manipulation or narrative to cause the stock to go down.

于是我们就成了那些想编叙事、做空操纵股价的人的"活靶子"。

Now, the thing I don't like about short sellers in the way the market's constructed today is — they can take a large bet on puts, or sell their research to hedge funds who take a large bet on puts, and put overly dramatic articles out there to try to spook investors to sell off a stock.

我对当前市场结构里做空机构这件事的不满是——他们可以自己重仓 put 期权,或者把"研究"卖给重仓 put 的对冲基金,然后发出夸张的文章,把投资人吓到卖掉这只股票。

At the beginning of their short report, they'll say: "We've most likely covered our short position by the time you're reading this report."

而且他们做空报告开头会写一句:"等你读到这份报告时,我们很可能已经平仓离场了。"

So there is not only this massive financial incentive to make the post much more dramatic than necessary or real, there's also not really any downside or protection against what they post.

所以他们不仅有巨大的财务动机去把内容写得比真实情况夸张得多,而且这种发布几乎没有任何下行风险或制衡。

They don't have to be accurate, because they go disclose everything they do.

他们不需要写得准——因为他们已经先披露了"我们在做空"。

On the other side, myself as a public company executive, we operate within the boundaries of the SEC. We have to be accurate in everything we say. We cannot go out and be misleading in any statements.

而我作为上市公司高管,要受 SEC 监管。我说的每句话都必须准确。我不能发任何误导性陈述。

So it's very difficult to be in that position — unable to address these types of reports, and get attacked by people that don't have any sort of downside to what they're posting and have quite a bit of financial gain to come from attacking companies that people know less about.

所以处境很尴尬——我没法正面回应这些报告,而打我们的人没有下行风险,反而有巨大的财务回报,尤其是打那些大众不太熟悉的公司。

What it did for us were two things. One, our team understands volatility.

这件事给我们带来两个结果。第一,我们团队理解什么叫波动。

You go down by as much as we did in the first year post-IPO — the team that's still there doesn't have a problem with volatility, and most likely they have a lot of conviction in the business model and the path to the future.

上市第一年跌成那样,留下来的人对波动早就免疫了,而且他们对商业模式和未来路径有非常强的 conviction。

Some of the team — it was funny, you don't usually want people posting responses, but some of our leaders were posting: "it's funny that the short sellers, because we're so good at what we do and the model that we built, they can't come up with anything other than 'we're cheating'."

有些团队成员——一般你不想让员工自己出去发回应——但当时几位 leader 发了帖说:"挺好笑的,做空机构因为我们做得太好、模型也搭得太强,他们除了说'我们在作弊',真的找不出别的。"

There was a lot of pride in what we built coming out from the team. So I knew that the team was sound and we were going to be able to recover from any sort of attack.

团队里那股"我们做的东西是真的厉害"的骄傲,被这波短空袭逼出来了。我心里知道——团队稳,我们能挺过任何攻击。

The second piece that it did, as a forcing function, was — it required us to go out to investors and the market and do more things about marketing the company and explaining the business.

第二个结果是——它成了一个 forcing function,逼着我们走出去,做更多"对市场和投资人解释公司"的工作。

In a way, it was like a rip-off-the-Band-Aid moment for us when it comes to marketing.

在营销这件事上,短空袭对我们就像"一把撕掉创可贴"的瞬间。

We had to make ourselves more available, and we had to be able to articulate what it is that we do in a clear way.

我们必须更愿意出来,也必须能用一句话讲清楚"我们到底在做什么"。

Was it a mistake to not invest in brand marketing, brand awareness before that?

那在那之前不做品牌营销、不做品牌认知,是不是一个错?

Look, it's easy to say in hindsight, it could have been a mistake. We grew really fast. When you're heads-down, you're working in a lean organization.

事后看,挺容易说"那是个错"。但我们当时增长太快——你埋头干活、又是个精简团队。

What was the revenue growth year one, year two, year three — kind of ballpark?

大致讲一下,你们头几年营收增长是多少?

Honestly, I don't remember exact numbers, but near triple digits each year.

具体数字我记不清,但每年接近三位数百分比的增长。

Our rule of 40, sorry, our rule of 50 — in the last quarter, I think was like 150.

Rule of 40——我们最近一个季度大概是 150。

So, not only are we growing — we grew like 70% year-over-year. We have 84% EBITDA margins.

所以不光是在增长——我们 YoY 增长大约 70%,EBITDA margin 84%。

The revenue growth since we launched Axon 2 model has been astounding. The profitability profile of the business is crazy. The business is expanding without adding heads.

Axon 2 模型上线之后的营收增长,数字非常夸张。利润率也疯狂。业务在扩张,但人头没增加。

We have a very odd financial profile, because when you look at it, you go: how can a business have 84% EBITDA margins? There's not another comp in the world that looks like it.

我们的财务结构非常古怪——你看的时候会想:一家公司怎么可能 84% EBITDA?世界上找不出第二家同行业可比公司。

And so a lot of the things that we've been able to accomplish just don't make sense to people. And in a world where things don't make sense, people think you're cheating.

我们做成的很多事,对外人来说讲不通。在一个讲不通的世界里,人们的第一反应是:你在作弊。

Instead of realizing you built one of the cooler technologies the world's ever seen — and then you go, as a team and me as the CEO of the business, it's my responsibility to go out and explain the business.

而不是意识到"你做出了世界上最酷的几项技术之一"。所以这件事就落到我和团队身上——作为 CEO,我有责任出来把这家公司讲清楚。

I owe it to my team who's built this really cool technology to go explain the business — and we owe it to our partners in the industry. Because when people take shots on us, on the other side, you've got advertisers who are buying on a performance basis.

我欠团队的——他们做出了真的很酷的技术。我也欠行业里的合作方——因为当有人开火打我们的时候,另一边坐着的是按效果付费的广告主。

They're spending billions of dollars a year. We put out that a year ago the scale of investment on our platform was $11 billion run rate. We've grown a ton since then.

这些广告主一年花几十亿美金。我们一年前披露过,平台上广告投入的年化规模是 110 亿。从那之后又涨了不少。

So you're talking well over $10 billion a year of dollars spent on a performance basis. So a shot on us is effectively calling all these advertisers, who are spending at that large scale, a bunch of morons.

所以这是一年远超 100 亿美金、纯按效果付费的钱。任何针对我们的攻击,本质上是在说"这些花钱大手大脚的广告主,全是傻子"。

Not only did I owe it to my team, I owed it to our clients to go out and explain our business and explain why some of the world's best marketers are buying on the other side.

所以不只是欠团队,我也欠客户——我必须把公司讲清楚,讲清楚"为什么全球最强的营销人在我们这买广告"。

Some of the world's best businesses are growing really quickly and profitably on the other side. And our engineers have built really exceptional technology.

他们里面包括一些全球做得最好的生意,跑得快、赚得多。而我们这边的工程师,做出了真的不一般的技术。

Two things. You speak with such confidence. You said "shots on you." Do you give a [shit] what other people think about you?

两件事。你说话特别有底气。你刚说"针对我们的攻击"。你他妈在乎别人怎么看你吗?

A long time ago, I realized you can't control that. So, no.

很久以前我就想通了——这事你控制不了。所以——不在乎。

Okay, you don't? Do you ever doubt yourself? If you speak with such confidence and such assuredness, dude, I want to follow you. No, I'm being serious. I'm like — what?

行,你不在乎。那你怀疑过自己吗?你说话这么笃定,我都想跟着你干了。说真的,我有时候听完就想"啥?"。

This is very rare, but there are moments when you're just you and your wife in the kitchen and your head is in your hands like — [shit]. Do you have a doubt or not?

这种很少见,但总有那种时刻——就你跟你妻子在厨房,你头埋在手里——靠。你到底有没有怀疑过?

I'll say, like, building the business — almost every morning I'd wake up thinking: I got to check stats, make sure we're still operating, or are we going to go bankrupt today?

我这么说吧——做这家公司的过程里,几乎每天早上一睁眼,我想的都是:得去看数据,确认我们还在运转——还是今天就要破产了?

And so in a way, I've always had this doubt that this is real, that what we're building is going to last, that what we're building is going to be really big.

所以某种意义上,我一直怀疑——这一切是真的吗?我们做的东西真的能长久吗?能做大吗?

And in essence, like that fear of blowup is one of my big motivators.

本质上,那种"怕炸"的恐惧,就是我最大的动力之一。

So I feel like I always have that doubt. I don't ever feel like we've made it.

所以我一直带着这种怀疑。我从来没觉得"我们成了"。

And that pushes a lot of us to want to keep pushing forward, because we are in a very tough space. Advertising is very competitive.

这股劲推着我们继续往前——因为广告这门生意非常残酷,竞争极其激烈。

There's a lot of technology that's improving in terms of capability for our performance stack, but also that forces us to continue to be innovative — otherwise we'd fall behind peers.

技术一直在进步,我们的 performance 栈也越来越强——但这也逼着我们必须持续创新,否则就会被同行甩开。

And so if we ever get complacent, we're almost certain to lose.

所以我们一旦自满,几乎一定会输。

I always tell investors or team: if at any moment I sound like I don't have conviction in our future path, we're sort of reeling. That would be a moment to doubt us.

我经常跟投资人和团队说:任何一刻如果我听上去对未来路径没 conviction 了——那就说明我们真的在打转。那时候才是该怀疑我们的时刻。

But I don't feel that way, because I've been doing this a very long time.

但我自己并没那种感觉,因为我做这件事已经很久了。

And with the team that I've got working on these technologies, this product, this platform, and the opportunities in front of us, I've always had conviction that the future was going to be better than the past.

我手里这支团队、这套技术、这个产品、这个平台,以及我们眼前的机会——让我一直相信未来会比过去更好。

And that has kept me in a position where I can voice confidence in what we're doing.

这种 conviction 让我有能力把"我们在做什么"讲得有信心。

One of my very dear friends has built up a half-a-billion-dollar position in ByteDance — obviously TikTok's parent company — and he said one of the reasons is they have the most advanced targeting engine in the world. Would you agree?

我一个非常好的朋友在字节跳动(TikTok 母公司)上压了 5 亿美金的仓位。他说原因之一是字节有全世界最先进的定向引擎。你认同吗?

When it comes to engagement — creating the ability for a social network to not need any social interaction and still be able to deliver you fantastic content — the TikTok recommendation algo is quite phenomenal.

就"engagement(用户留存与黏性)"这件事而言——能让一个社交网络不靠任何社交关系、仍然给你推出极棒的内容——TikTok 这套推荐算法非常惊人。

If you think about recommendation systems — what's the world we operate in?

想一下"推荐系统"这件事——我们身处的是怎样一个世界?

On the one hand, the content you see on Instagram, the content you see on TikTok is very dialed in to what you're interested in. It is a constant loop.

一方面,Instagram 和 TikTok 推给你的内容,已经精准对到你的兴趣点上。它是个不停的循环。

It's very interesting — the advertising systems too. The ads you see on Instagram have become very much like content. They're highly relevant.

有意思的是——广告系统也一样。Instagram 上你看到的广告,越来越像"内容"本身。相关性极高。

The ads that we're able to show consumers now are getting very relevant and they drive action.

我们现在能给消费者展示的广告,相关性非常强,而且能驱动行动。

As the technology and recommendation system models, and just generally AI models, have gotten better — the capacity to serve more relevant, more targeted ads to the consumer, even knowing less about the person, has gotten so good that people are really able to use the advertising to discover the products they want.

随着推荐系统模型和 AI 模型整体变强——即使对用户了解更少,我们也能投放更相关、更精准的广告。结果是,用户真的开始通过广告"发现"自己想要的产品。

We mentioned TikTok and we mentioned Meta there. AppLovin currently valued circa $150 billion market cap — whatever it is precisely, but give or take. For AppLovin to be a trillion-dollar company, do you have to be a social network as well?

我们刚提到了 TikTok 和 Meta。AppLovin 现在大约 1500 亿市值——具体数字不重要。AppLovin 要做成万亿市值公司,你们是不是也得做成社交网络?

No. If you think about what creates a trillion-dollar business — and I sort of said cash flow minus SBC is a real important metric — like if we ever got to generating $30-35 billion of cash a year, we'd probably be a trillion-dollar business, right?

不需要。想一下万亿公司是怎么撑起来的——我之前讲过 cash flow 减 SBC 是关键指标。如果我们一年能产出 300-350 亿现金,我们大概就是一家万亿公司了。

So you think about what can get us to that point. There are a couple of things that can get us there.

所以问题变成:什么能把我们带到那个数字?有几条路。

One is continued execution in the domain that we're in.

第一,在我们现在这个领域继续执行。

We think we can get much bigger just by better monetizing the gaming audience. It's a billion-plus daily active users who play these games. Adult audience, a lot of heads of household.

我们认为光是更好地变现游戏用户,我们就能再上一个量级。这是一个 10 亿+ DAU 的群体,成年用户多,很多还是家庭决策者。

The next thing you think about is how do you expand what you have? In the past I've talked about connected TV as one of the holy grails of advertising.

第二条是:怎么扩展已有的能力?我以前讲过——CTV(联网电视)是广告业的圣杯之一。

If you can port the performance ad we serve on mobile to the television, and allow small and medium-sized businesses to serve there, and make it all performance-based — that's a really big unlock.

如果你能把我们在移动端做成的 performance 广告,搬到电视上,让中小企业也能在那投放、而且全部按效果付费——那是一个巨大的解锁。

So it's something we still take seriously.

所以这件事我们一直认真在做。

Then you think about what are other applications of the technology. We're really good at advertising models. We have yet to have a chance to have our team work on an engagement model.

第三条路是:这套技术还能用在哪?我们做广告模型很强。但我们还没有机会让团队去做一个"engagement 模型"。

So a social network for us is not a requirement to get to a trillion dollars. It's an interesting play to recruit talent and continue to tune our skills in modeling.

所以对我们而言,"做社交网络"不是奔向万亿的必要条件——但它是个有意思的牌:用来吸引人才,继续打磨我们的建模功力。

As you think about the research labs and any company that's building models — they better have things that are interesting for new researchers to come in to work on, new applications of technology.

任何一家研究实验室、任何在做模型的公司——你最好有一些有意思的新课题、新场景,能让研究员愿意来。

For us, a lot of these bets will also be means to go hire some of the best people in the world. If we execute on it, obviously great. But not a requirement.

对我们来说,这些押注的另一层意义是吸引顶尖人才。跑通当然好,跑不通也不是必需。

Eli Gil just tweeted actually — compute is the currency of the future, and compute will be one of the defining factors that the best talent looks for when deciding which company to join. Do you agree?

Elad Gil 刚发推说——算力是未来的货币,顶尖人才挑公司时,算力会是关键决定因素之一。你同意吗?

Depends on the space. So large language models obviously have the ability to scale with more compute, and therefore it is attractive to researchers to join companies that can invest a lot in compute.

看领域。LLM 显然吃算力——这种情况下,愿意大手笔砸算力的公司,对研究员就有吸引力。

If you look at right now — we can all say probably Anthropic is doing the best in terms of releasing models and product in the large language model space as of this moment. Anthropic probably does not invest the most in compute.

但你看现在——LLM 这块,Anthropic 目前在"模型 + 产品"双发布上做得最好。可 Anthropic 在算力上的投入,大概率不是最多的。

So if you think about that — how did they actually get really good researchers creating the best product output? They have really good culture and they have really good people, and they really tuned what they were going after.

这就有意思了——他们怎么招到顶尖研究员、做出最好的产品输出?答案是:很好的文化、很对的人,以及对方向的精准打磨。

Recommendation system space does not need as much compute to create the output that's necessary to succeed.

推荐系统这块对算力的需求,远不如 LLM 那么疯狂。

So it's quite different. You're looking for people that still want to solve really big problems and are very mathematically inclined — but there's different spaces in modeling. There's vision models, there's LLMs, there's recommendation system, there's others.

所以这是另一个游戏。你要找的依然是"想解决大问题、有强数学直觉"的人——但建模本身分很多领域:视觉模型、LLM、推荐系统等等。

Depending on the product that someone is interested in, they'll go to a different company. You've got people that like working on recommendation system models, and they're not bound by compute — they're bound by curiosity and application of techniques to create a better output.

研究员去哪家公司,看他对什么产品感兴趣。喜欢推荐系统的那批人,瓶颈不是算力——瓶颈是好奇心,以及把技术应用到产品上做出更好输出的能力。

You mentioned the buyback that you did in 2022 — when the stock was very low. We've seen a wave of buybacks, whether it's Wix or ServiceNow among many others, Salesforce huge. How should we read these buybacks — a sign of internal confidence?

你提到 2022 年股价很低时你们做的回购。现在出现了一波回购潮——Wix、ServiceNow、Salesforce 等等。怎么解读这些回购?是内部对自家信心的信号吗?

Buybacks are interesting because if you look at history, the concept of a buyback doesn't usually pan out. It's not usually a good financial bet.

回购这件事很有意思——看历史的话,回购这件事大多数情况下并不划算。从纯财务角度看,它通常不是好赌注。

It's tough. Here's why it's tough — it's easy when you're inside to think you're cheap, but you sort of trade where you deserve to trade.

为什么难?——你在公司内部,很容易觉得"我们被低估了"。但市场往往就是把你交易在你该交易的水平上。

And it's really hard to know — when is it cheap enough?

真正难判断的是——什么时候才算"足够便宜"?

The reason why our buyback was very successful is — when we went public during COVID, we didn't build a really big roster of blue chip investors. So we had a very flimsy cap table.

我们那次回购很成功的原因是——我们 COVID 期间上市,没建起一批真正的蓝筹投资人。我们的股东结构非常脆。

And this led to the stock collapsing much more than it should have.

这导致股价崩盘程度远大于"它本该跌的"程度。

When we went public in '21, we had $700 million of EBITDA. $28 billion IPO — company goes to $40 billion.

2021 年上市时,我们有 7 亿 EBITDA。IPO 估值 280 亿,然后冲到 400 亿。

In '22, we cleared a billion of EBITDA. So we grew 40%-ish in '22. Yet the stock fell 92%. We got to under 4× EBITDA.

2022 年我们做到了 10 亿 EBITDA——同比增长大约 40%。但股价跌了 92%,EBITDA 倍数跌到 4 倍以下。

So why did that happen? We went public in COVID, didn't attract blue chip investors. Our cap table was basically the private market cap table that needed to sell.

为什么?因为我们 COVID 期间上市,没招到蓝筹投资人——我们的股东表本质上就是一批"必须卖"的私募时代股东。

That's a real big problem. Most companies that are private probably have half their cap table as sellers.

这是个大问题。大多数私募公司,股东表里大概一半人是"该卖了"的那种。

So when we went and did our buyback, we didn't say: "Hey, we're just going to go to the market and take float out, take a share back from every single shareholder." That would imply that even partially part of my shares are getting bought back, right?

所以我们做回购时,不是简单地"去市场上买回流通股、从每一个股东那里都买一点"——如果那样,连我自己的股份也得部分被回购掉。

Instead we said: "If you are a seller, please work with us to sell back to the business."

我们换了个思路——告诉股东:"你想卖的话,跟我们合作,直接卖回给公司。"

We went and deployed every dollar that we made — and we raised some debt to deploy even more — and took back a lot of the shares on the cap table that were going to inevitably sell into the public markets over the coming months.

我们把当时挣的每一块现金都投进去回购,还借了一点债加大力度,把那些"接下来几个月几乎一定要在公开市场抛售"的股份接过来。

By doing that, we were able to get liquidity to company folks, investors, and old ex-co-founders and other folks on the cap table that needed liquidity. We were able to get them liquid — no problem. We were happy to do the trade.

这样一来,公司员工、投资人、老联合创始人、其他需要流动性的股东都拿到了出口。我们也乐意做这笔交易。

And we were able to take out the selling pressure with these folks — the fact that they were willing to work with us was a gift.

这样我们清掉了卖盘压力。他们愿意配合我们,本身就是一份礼物。

So we were able to take out the selling pressure, and then as the business started accelerating — you remove the selling pressure and overhang, then you're set up in a position where you can now go attract the right investors.

把卖盘和悬而未决的抛售压力清掉之后,业务加速时——我们就能开始去吸引"对的投资人"。

Give or take, how much money did that buyback make you?

大致估一下,那波回购给你们赚了多少?

Based on where we're trading today, roughly a third of the company's value came from that buyback. So you said $150 billion roughly — let's call it $50 billion around.

按今天的市值算,公司价值大约三分之一来自那次回购。你说 1500 亿——那大约就是 500 亿。

Well done. Yeah, it was a good buyback.

漂亮。是,那是一笔成功的回购。

Now, now just buying out of the market — just buying your float back — it's not a good bet usually.

但单纯地"从市场上回购流通股",通常不是个好赌注。

Wix — I really like the Wix team. Great and lovely people. But gosh, you only do the buyback, a big-ass buyback, and then down like 25% in a [short time].

Wix——我真心喜欢他们团队,人很好。但你做了这么大的回购,股价然后又跌了 25% 出头。

That's the problem. You start doing the buyback and if you're not right, you don't time it well — and we're not, none of us are day traders when we're running businesses. You can burn the capital that you made really quickly and then you're in a much worse spot.

问题就在这——你启动回购、判断错、时机不对——经营公司的我们,没人是 day trader。一不小心就把辛苦挣的现金烧光,反而陷得更深。

You said people trade where they deserve to trade in a lot of cases. Is the "SaaS apocalypse" fair, then?

你刚说"市场往往把你交易在你该被交易的水平"。那"SaaS 末日"这个说法公平吗?

As an investor — if I was one — when you get into an unpredictable outcome in the future, it's very easy to sell businesses. And the rapid rate of product delivery in the large language model space makes a lot of traditional enterprise SaaS companies hard to bet on years into the future.

如果我是个投资人——未来一旦变得不可预测,你最容易做的事就是卖。LLM 这块产品发布的速度太快,让很多传统企业 SaaS 公司"五年后还能不能押"变得难判断。

So what happens? Terminal value is dicier. So you value the company less, you get out.

那后果是什么?公司终值变得更不确定。你估值打折,你撤资。

Their stock-based comp was high, but it was an acceptable percentage of total value. Stock tanks, stock-based comp becomes too extreme.

原来 SBC 不低,但占总市值的比例还能接受。股价一崩,SBC 比例瞬间过高。

Now they're in a position where not only are they going to lose their heads, they're also competitively challenged.

这时候这些公司不光要被迫裁员,而且在竞争上也受冲击。

So you're in a really bad downward spiral.

直接进入一个非常糟的死亡螺旋。

In a way, I would say not only is it fair because of the risk that exists — I'm not sure it's actually done yet.

所以某种意义上,不光是"公平"——风险确实在那里——而且我不确定这一波是不是已经结束了。

Again, I'm not a trader of businesses, but I do think we're going to go through material changes in the market, especially when it comes to enterprise SaaS over the coming years.

我不是炒公司的——但我觉得未来几年市场会有实质性的变化,尤其是企业 SaaS 这块。

It may not be that these companies that we have today as some of the SaaS leaders are completely going to wipe out — because I don't think that happens. Companies, once they're embedded with you utilizing a certain software, usually don't change.

这并不意味着今天这些 SaaS 龙头会被彻底抹掉——我不觉得会发生那样。客户一旦用上某套软件,通常不会换。

But it may be that a lot of the growth opportunities are gone for these businesses.

但很可能这些公司未来的增长机会大幅消失。

And you strip out growth opportunities in businesses — I mean, the reason we went and traded down to under 4× EBITDA is because investors did not believe in our future growth prospects.

公司一旦失去增长前景——我们当年掉到 4 倍 EBITDA 以下,就是因为投资人不相信我们的增长前景。

When you're a public market investor, you only like to bet on companies where you have sound belief that their future is going to be a lot rosier than the present.

公开市场投资人的本能是——只押那些"未来明显比现在好"的公司。

It is very hard to believe that right now in traditional enterprise SaaS, when these large language model businesses, the frontier models, continue to get so much more powerful.

在 LLM、frontier 模型越来越强的当下,你很难对传统企业 SaaS 抱有"未来明显更好"的信念。

You mentioned the material changes there — what do you think the most material changes will be in the next few years?

你说"未来几年会有实质性变化"——你觉得最实质性的变化会是什么?

The rate of advancement is astounding over the last few months. You see the amount of products that are rolling out — it's like every day there's something new.

最近几个月技术进展速度惊人。每天都有新产品发布。

So I think the coolest thing that we're seeing right now is — for people who know how to utilize it, the ability to just launch an army of agents to do certain tasks. And obviously coding is the most obvious utilization today.

我觉得当下最酷的事是——对那些会用它的人来说,你可以直接"放出一支 agent 大军"去干特定任务。最明显的应用就是写代码。

If you see that today and believe we're already at a point where the army of agents can continue to start improving the code that's available to them and the products that are available to them in a recursive way — the rate of acceleration of technology and R&D, and our imaginations becoming products, is only going to get faster.

如果你相信我们已经到了"agent 大军可以递归地不断改进自己能写的代码和能搭的产品"这一步——那技术、研发、"想象变产品"的速度只会越来越快。

Where does that lead us to? I don't know. But I think it's going to be a much more productive future than the present.

这会把我们带到哪儿?我不知道。但我觉得未来一定比现在生产力高得多。

We mentioned engineering again. You said you don't have a product team, so to speak. How do you think about the org chart today, and how that changes over time? Do we lose product as a function?

回到工程话题。你说你们不设产品团队。你怎么看今天的组织架构,以及未来怎么变?"产品"这个职能会消失吗?

We chose not to have it because we wanted to have exceptional engineers that understood the product.

我们之所以不设,是因为我们要的是"懂产品的顶尖工程师"。

The belief was — if our engineering team is writing the product that delivers revenue, our sales team and all other teams are effectively cheerleading for the engineering team, making sure they have what they need and then eventually going out and selling their product.

我们的信念是——既然工程团队写的就是"产生营收的产品",销售和其他团队的角色就是给工程加油、确保他们有他们需要的东西、然后把产品卖出去。

But we can only sell the product if it's good enough to be sold.

但产品得先够好,才卖得动。

The engineers, if exceptional, better be good enough at understanding the product they need to build to go build it.

所以顶尖工程师,必须强到能"看懂自己该建什么产品",然后去把它建出来。

I do think the role of product should end up looking a lot like it does at our company over time — either your product people become engineers or your engineers become product people, but you don't need both.

我觉得"产品"这个职能未来会越来越接近我们的样子——要么产品人变成工程师,要么工程师变成产品人,你不需要两套人都存在。

So what usually happens — whoever becomes AI-native and knows how to utilize these tools will become those powerful 10x, 100x output folks who know how to use the tools to create that kind of output.

规律是——谁先成为 AI-native、谁先掌握这些工具,谁就成为那批 10 倍、100 倍产出的人。

I do think for some time still, though, you're going to need an engineer doing the work and still making sure that the code is up to security standards, the code's not slop, the code is good enough to contribute to your main codebase.

不过这两年里,你还是需要工程师亲自审——确保代码达到安全标准、不是 slop、足够好到能进主干 codebase。

So there's a lot that still comes from having a traditional engineering background that's valuable in today's world.

所以传统工程背景仍然很值钱——尤其在今天这个环境里。

Every day it feels like we have another major security breach. We've seen Lovable in the last 24 hours. We saw Vercel in the 24 hours before that. It goes on and on and on in the last month.

每天都有大型安全事故。过去 24 小时是 Lovable,再之前 24 小时是 Vercel,过去一个月接连不断。

To what extent are you nervous that models will unravel vulnerabilities that were previously unseen, and we have security be the biggest problem?

你担心吗——模型会把以前没人看到的漏洞挖出来,让安全成为最大问题?

Look, obviously there's a risk there. You could ask: is Anthropic slowing down the rollout because they don't have the compute, or because they're really concerned about the risk? It's probably somewhere in between.

显然这是真实的风险。你可以问:Anthropic 放慢发布,是因为算力不够,还是因为他们真的在担心风险?——大概两者都有。

These models — one of the things they're built for is to audit code and expose any vulnerabilities or bugs, and solve them.

这些模型——它们被训出来的能力之一,就是审代码、把漏洞和 bug 暴露出来并修掉。

So you would hope that we will be a lot more buttoned up on security in the future than we are today.

所以你应该期待,未来安全水平会比现在高得多。

But because of how quickly these models are just getting exceptionally good — it's almost certain companies are going to be releasing code faster. When you release products faster, you ship fast, you break things, and because of that, you're going to have more security breaches most likely.

但由于模型变强的速度极快——公司几乎一定会更快地发版。发得快,就 ship fast、break things,中间出现更多安全事故几乎是确定的。

But once you get past that point, you're probably going to be in a point where the technology is a lot more buttoned up than it was before.

但一旦熬过那段窗口期,整体技术的安全水位会比之前高得多。

What is no one talking about that you think everyone should be talking about?

现在没人讨论、但你觉得所有人都该讨论的是什么?

I do think there needs to be a lot of honesty around what is the world going to look like as these AI technologies continue to get more powerful.

我觉得我们必须更诚实地讨论——AI 技术继续变强,世界会变成什么样。

If every technology company could stand to lose 75-80% of their talent and get more efficient — what does that actually mean?

如果每一家科技公司都可以裁掉 75-80% 的人、反而效率更高——这句话到底意味着什么?

Does it mean that there's going to be 10 times more startups? So the startup funds are going to be crushing it, and people are going to be way more productive and we're going to get way more product in the world. Plausible.

是不是意味着会有 10 倍数量的创业公司?创投基金会大爆?生产力暴涨,世界上的产品多 10 倍?——有可能。

I'm a believer that the technology unlocks a lot more output, and our ability to imagine things, create, and then go and create becomes not only cheaper, but much more believable.

我相信这项技术会解锁巨大的产出能力,我们"想象 → 创造"这条链条不仅变得更便宜,而且变得更可信。

But it requires people to really level up.

但代价是——人必须真升级。

And I think we need to be honest about what the path is going to look like — because my guess is, you're going to see a lot more tech layoffs over the next couple of years, as companies really start understanding that not laying people off creates a blockade to actually getting to this AI-native state.

我们必须老实承认这条路会是什么样——我猜未来几年还会有更多科技裁员。因为公司迟早会意识到:不裁人,就到不了真正的 AI-native 状态。

Did you find it hard that in a year where you have triple-digit growth, stellar year, you're laying off such a large portion? Because I really like and respect you, but at that point, that's a choice you don't need to make.

在一个三位数增长的好年景,你还要裁掉那么大比例的人——这件事难做吗?我是真喜欢你、尊敬你,但说实话那个时间点你完全不必那么做。

Look, again — are you playing to win, or are you playing not to lose?

看——回到那个问题:你是在为赢而玩,还是为了不输而玩?

So I feel like we're very transparent with our employees today.

所以我们对员工非常透明。

Anyone asks me, I'll say: "You're here because you're an exceptional talent."

谁问我,我都直说:"你能留在这,是因为你是顶级人才。"

And what does that mean going forward? You use these technologies to create more output. You become AI-native. You're going to have a role here.

那未来呢?——你用这些技术做出更多产出、你变成 AI-native——你就有岗位。

If you avoid utilizing these technologies, you're not. And you're going to get fired, and that's life.

如果你回避这些技术——你就没岗位。会被裁掉。就这么简单。

So we demand that the people who are at the company are adopting these technologies rapidly to create more output.

所以我们要求公司里的每一个人,都必须快速拥抱这些技术、做出更多产出。

But we don't shy away from difficult discussion — because if they're not able to do that, there's a role somewhere for them, but it wouldn't be at our company.

我们也从不回避那种艰难对话——如果他们做不到,他们在别的地方会有岗位,但不会在我们这里。

Totally get that. Final one before we move to a quick-fire is just on budgeting. Token budgeting is one of the biggest questions for leaders — how should I think about it, actually thinking about planning it, forecasting?

完全 get。进 Quickfire 之前最后一个——预算。token 预算现在是 leader 们最大的问题之一:该怎么想?怎么规划?怎么预测?

To my point — I think it's flawed logic. Because if you just throw a budget at people and you create a leaderboard of token usage, what are people going to do? Create a bunch of crap that has no value.

回到我刚才的观点——这个逻辑本身有问题。你给员工设个预算,再搞个"token 用量排行榜",大家会干嘛?——制造一堆没价值的东西。

All of a sudden you burn your budget, you're paying really big checks, and you don't have revenue on the other side of it.

突然之间你预算烧光、支票越签越大,而对面什么营收都没。

Companies need to get to the point of understanding — what are they actually optimizing to, and who's utilizing the technologies and creating token consumption that actually aligns with those KPIs.

公司必须先搞清楚:我到底在优化什么 KPI?哪些人用这些技术、产生的 token 消耗,真正跟这些 KPI 对得上?

When that happens, you won't be in the mindset of token budgeting. You will want to invest in tokens because there's revenue on the other side of it.

一旦想清楚,你就不会再有"token 预算"心态。你会想多砸 token——因为另一边有营收。

But I think today people are just blindly going "spend a bunch of money, get on the leaderboard, use the tools, something good's going to happen." You better be able to measure that, otherwise you're going to get a lot of bad behavior.

但现在大家就是盲目地"砸钱、上榜、用工具,等着好事发生"。你必须能度量,否则只会引出一堆错误行为。

It's no different than companies that staffed up to very large team sizes and bloated teams over the last decade or 15 years in the Valley because they had the means to, and it was "let's just get on a hiring quota."

这跟过去 10-15 年硅谷公司"反正有钱、定个招聘指标、把团队招到很大、最后变臃肿"完全是同一套逻辑。

Token quotas and token budgets are no different than hiring quotas. Until they get efficient, they'll be inefficient, and I think a lot of companies will just burn money.

token 配额、token 预算,跟"招聘指标"没区别。直到他们想清楚什么才叫高效之前,他们就是低效——很多公司会因此烧掉一大笔钱。

I care desperately, before we do a quick-fire, about being the best I can be and being number one in my business. I also want to be a parent. What is the uncomfortable truth that I should hear about being a parent and trying to be the best?

进 Quickfire 之前我得问一句——我极其在意"成为最好的自己、在自己生意里做到第一"。但我也想做个好父亲。关于"既想做最好,又要做父母"这件事,你能告诉我一个不舒服的真相吗?

I think it's really hard. As human beings — to become really good at something — you have to focus on it and you have to put out a lot of effort. At least me, I'm not all that great at multitasking.

非常难。人类要想在一件事上真正做到顶尖,必须专注、必须投入大量精力。至少我,我不太会同时做好几件事。

Being a parent is a really difficult thing.

而做父母,本身就是一件非常难的事。

So if you are a founder running something and you want to become the best — you want to become the best podcaster, I want to become the best in advertising with my team leading us the way there — to do that you need to prioritize that task.

所以,如果你是个 founder 在做事——你想成为最好的播客主持人、我想跟我团队一起把广告做到最好——你必须把这件事放在第一位。

And the second you do that, in essence you're deprioritizing the task of being a parent, being a husband, being a good person in the personal life.

但你一旦那么做,本质上你就在把"做父亲、做丈夫、做生活里有担当的人"这件事往后推。

It requires having a family that understands the commitment you have to the day job, and it requires a balance that is really hard to attain.

这要求你的家人理解你对工作的全身心投入,也要求一种非常难真正实现的平衡。

What have you missed that you regretted?

你错过过什么、让你真正后悔的?

As you do what you do — like I said, a lot of times you're not really connected to reality to what's happening around you, because your mind is wandering. My mind is always on business.

在这种状态下做事——很多时候你跟身边正在发生的事,其实是脱节的,因为你脑子在别的地方。我脑子一直在公司上。

Even when I dream and I wake up, it's something about business. So I think back at moments where the kids were growing up and sort of a blur — and I go: you know, was I just not there?

我做梦都是公司,醒来还是公司。所以回头看孩子成长的那些时刻——一片模糊——我有时会问自己:我是不是根本就没在场?

I was there, but I wasn't there mentally. And it is not a great thought when you have that.

人在,但心不在。这种念头一旦冒出来,不好受。

And on the other side — I do, because this business that became much bigger than I thought was possible means a lot to me. And figuring out that balance is really hard as human beings.

但另一面是——这家公司做得比我想象的大得多,对我来说意义重大。作为人,要找到那个平衡极其难。

I think it is a challenge. It's one that I'm still working on. I think it's very hard to accomplish being really good at all facets of life.

这是一个挑战。我现在仍在调。要在生活的每一面都做得非常好——非常难。

The ultimate challenge. Do you mind if we do a quick-fire round?

终极挑战。介意我们来一轮 Quickfire 吗?

Yeah, go for it.

来。

What have you changed your mind on most in the last 12 months?

过去 12 个月,你改变想法最多的是什么?

I don't know if I've changed my mind much in the last 12 months.

我不太确定我有改过什么大想法。

When I hit that low point in '22, I sort of got to a place where I said — I'm going to think forward about what I do professionally, and maybe this translates to my personal life too. I'm going to plan out 3 to 5 years and work back from it.

2022 年最低点的时候,我决定了一件事——以后工作上(可能也包括生活)我都用 3 到 5 年视角往前看,然后倒推回今天。

When you think about the current year, 12 months, I feel like whatever is happening now is defined by the decisions we made in the past. Nothing that I do today is going to change an outcome in that 12 months.

所以你看现在这 12 个月发生的事——其实都是过去的决策定下的。我今天做什么,改变不了未来 12 个月的结果。

What I'm thinking today, or trying to execute on today, or starting to research today, can change an outcome 1, 2, 3, 4, 5 years down the road.

我今天在想什么、在执行什么、在研究什么——影响的是 1、2、3、4、5 年后的结果。

But because it's undefined still and you're in that moment of "huh, I think this is something interesting" — it's very hard to challenge that thought.

但因为结果还没定,你只是在那个"嗯,这件事挺有意思"的瞬间——很难当下就 challenge 自己的判断。

If you believe in it and you've got conviction in it, you sort of just run with it.

你信、你有 conviction,你就照着跑。

So I don't know that I would change anything that I've thought about in the last 12 months, because I yet don't know what's going to happen from it.

所以我没什么过去 12 个月想法可以"撤回"的——因为我还不知道这些想法会带我去哪。

Who do you not have on your board who you would most like to have on your board?

你最想加进董事会、但目前不在的人是谁?

That is a tough question for me too, because I think we have a pretty well-constructed board, but I don't have a lot of experience with boards.

这个问题对我也难——因为我觉得我们董事会构成挺好,但我自己跟董事会打交道经验并不多。

When we were private from 2011 when we started the business to 2018 when KKR invested and we got our first three-person board, I didn't have a board.

2011 年创业到 2018 年 KKR 投资、组建第一届三人董事会之前,我们都没董事会。

I just ran on my own and we ended up just deciding and making choices as I saw fit. Obviously, I would consult my co-founders, other people on the team, but there was no board — because we were a bootstrap business with just a convertible note round.

那时我就是自己拍板。当然会和联合创始人、其他团队成员商量,但没有正式董事会——我们是 bootstrap 公司,只发过一轮可转债。

Then we had a three-person board. Now, we've got quite a bit bigger than that — I think it's eight or nine people if I recall.

后来是三人董事会。现在大了不少——我没记错的话是 8 到 9 人。

We have a really good composition of people now — the people around the table are a mixture of people who have worked at the company, know me intimately well, and are supportive — or have really good business instincts outside of us, who bring great things to it.

现在董事会构成挺好——一部分是在公司里干过、跟我很熟、支持我的人;一部分是公司外部业务嗅觉极强的人,带来不同视角。

And I actually recently stepped aside as the chairman of the board to hand it over to this gentleman, Craig Billings. He's CEO of Wynn, one of the smartest people I've ever met, very competent at building businesses and understanding corporate governance.

我最近还主动让出了董事会主席的位置,交给了 Craig Billings——他是 Wynn 的 CEO,我认识的最聪明的几个人之一,做生意和懂治理都极强。

I felt like my job is to run the business, and I don't want to be consuming my own time on anything other than day-to-day operations.

我觉得我的本职是把公司跑起来。除了日常运营,我不想把时间耗在别的事上。

Board is something that I've got to really work with and allow to be pulled into the business and contribute back to the business and work with me on the business — but it's not something that I'm going to be good enough to be the chairman on, versus someone like Craig who is exceptionally talented at all aspects of building a big business.

董事会这件事必须用心运作、让他们更好地介入业务、跟我一起把事做大——但我自己不是当董事会主席的好人选,而 Craig 这种人,在搭建大公司的各方面都极其强。

It's not common that you'll see a CEO step aside as chairman — but I've always believed that in every role we all do, whether it's me or someone else on the team, if there's someone better to do it, step aside and let them take over. That's something that allows you to always be leveling up.

CEO 主动让出董事会主席的位置并不常见——但我一直信:任何岗位,无论是我自己还是团队的人,如果有更适合的人,就让位、让他干。这才让你一直往上升级。

How do you feel about founders angel investing?

你怎么看 founder 做天使投资?

So I don't invest anymore for a couple reasons. One is — in order to invest, you've got to sell shares in your own business to have liquidity to go invest.

我已经不投资了,原因有几个。第一,要去投资,你得卖掉自家公司的股份套现。

And I don't know — again, if I didn't start this business for money, I don't know what I need to invest to create more return on.

我创业一开始就不是为了钱——那我去外面投出更多回报,我图啥?

If you're an investor, you hopefully really want to create return or impact or something that is a KPI that you care about. But the second you care about that KPI and you chase it, you're selling from your own core business to go diversify. You're not focused on your day job.

如果你是个投资人,你应该真心想创造回报、创造影响、或者追某个你在意的 KPI。但你一旦在意这个 KPI、开始追它——你就在卖自家公司股份去分散。你的注意力就不在本职上了。

For me, my goal in life is to make my company as good as it can possibly be three years from now, five years from now, 10 years from now, 20 years from now.

我的人生目标是把这家公司在 3 年、5 年、10 年、20 年后做到极致。

If I plot into the future, every second of my available time should be committed to it. Otherwise, there's some loss. I don't know what that loss is, but if I get distracted on other things, there's some loss that I can't measure.

从这视角往前看,我每一秒可用时间都该投在这上面。否则就有损失。我说不清是什么损失,但只要我分心,就有一笔我无法度量的损失存在。

And as those losses start adding up, they can compound and it can make it less likely that you could succeed.

这些损失会复利累加,最后让你成功的概率变低。

What decision with AppLovin would you do differently knowing what you know now? You mentioned the weakness of your cap table there — that struck me, and I was like, do you wish you delayed it then? There's never a good time to go [public].

如果你拿今天的认知回到过去,AppLovin 哪个决定你会改?你刚说股东表很脆,我听到觉得——你是不是后悔当时上市?反正没有"完美的上市时机"。

I don't question the past, because the past makes up where you are in the present. I've made a lot of decisions. A lot of them end up wrong. But we pivot and we learn from them.

我不质疑过去——过去构成了你现在所在的位置。我做过很多决定,很多都是错的。但我们调整、我们学到东西。

We went public in a very difficult time — towards the late of the growth-stock runup during COVID. And the second COVID ended and usage patterns returned to what they were pre-COVID, everyone collapsed in growth stocks, in particular those late-market IPOs.

我们上市时机非常难——COVID 增长股最后一波的尾端。COVID 一结束、用户行为回归疫前,增长股全线崩盘,尤其是后期上市的那批。

So you could say "okay, we didn't time the market right." As you just said, there's no right time to go public.

你可以说"我们没踩对市场节奏"。但就像你说的——根本没有"对的上市时机"。

I also think the learning for us, and anyone going public, is — the moment in time, it's like a Series A, Series B, Series C. It's a fundraising round.

我们的体会(对任何要上市的公司也是):IPO 本质上跟 A 轮、B 轮、C 轮一样,就是一轮融资。

You have a business that has long-term growth opportunities that you have high conviction in, and can be big enough to be owned by anyone in the world, and interesting enough to be owned by anyone in the world.

你公司有长期增长机会、你对它有强 conviction、它够大也够有意思,所有人都可以买它。

In a world where there's no right time to go public, you can't time the market. Just go public. You take the capital you raise, and you build forward.

既然没有完美时机、又没法 time the market——那就上市。拿到融资、往前建。

What's really important for me running the business is — I really am not focused on where the stock's going to be next quarter. 3 to 5 years from now, we better be higher than where we are by enough so that I feel like people made a good return on investment owning our shares today.

对我经营公司来说,重点是——我不在意下季度股价。3 到 5 年后,股价必须比现在高出足够多,让今天持股的人觉得这笔投资回报够好。

They better make more on us than they can make by owning the basket of the S&P, just putting their money in debt.

他们持有我们,要赚得比 S&P 指数、比把钱放进债券更多。

And if they make a good enough return on us over the next few to five years, I feel like I did my job right as CEO. Then they need the next 3 to 5 years, and the next 3 to 5 years after that — but we owe it to investors to make them a return greater than what else they can put their money in.

未来 3-5 年他们在我们身上的回报够好,我作为 CEO 这份工作就算干对了。然后再下一个 3-5 年、再下一个 3-5 年——我们欠投资人的,就是这种"比其他资产都好"的回报。

Finish this sentence — the advertising business that is most at risk from AppLovin in the next three years is __.

填空——未来三年最受 AppLovin 威胁的广告生意是 __。

So, tough question to finish the sentence on, because I don't think it's any. We build a business trying to help an advertiser reach a consumer, drive a transaction inside this gaming audience — billion-plus daily active users.

这题不好填,我觉得"没有"。我们做的事是:帮广告主在 10 亿+ DAU 游戏用户里,触达消费者、产生交易。

We're trying to create incremental transactions. When you do a performance marketing platform, we're not trying to take from others. We're trying to give an advertiser the chance to —

我们在做的是"增量交易"。一个 performance 营销平台,本质不是从别人盘子里抢——而是给广告主一个机会:

"You spend $100,000 a day growing your business today. Spend an extra $20,000 a day with us and create more transactional volume."

"你今天一天花 10 万美金扩大业务。在我们这多花 2 万美金,做更多交易。"

Don't take from anyone else. Take your $100,000 a day investment business that might have $300,000 a day of revenue, and add another $20,000 of media spend. Get to $120K and get to $360K revenue. Your business grows 20%.

不抢别人。你原本一天 10 万投入、30 万营收——多投 2 万到我们这,变成 12 万投入、36 万营收。业务增长 20%。

By investing an extra 20% in our technology, our platform, our audience that you otherwise weren't accessing in that moment.

这 20% 是花在我们的技术、平台、以及你当下原本接触不到的那批用户上。

I'm going to steal from another podcaster who's actually a friend of mine — but, "great artists steal" is another kind of quote. It's a really nice question and it's the final one. What's the kindest thing that anyone's ever done for you?

我要从另一位也是我朋友的播客主那偷一题——反正"great artists steal"嘛。最后一个问题:别人对你做过的最 kindest 的事是什么?

It's a weird one. Look, maybe kindness is — like in those dark moments — whether it's my wife, close friends, people checking in on me.

这问题挺怪。"kindness"对我来说——大概是那些黑暗时刻,我妻子、亲密朋友主动过来确认我状态那种。

But the reason I say it's a weird one is — we don't tend to push the word "kindness" around very often at the company. We believe you're pushing forward in an aggressive fashion, almost cutthroat.

这问题怪在哪——因为我们公司内部很少用"kindness"这个词。我们的文化是 aggressive 地往前推,甚至有点 cutthroat。

Do you worry that you're too aggressive?

你担心自己太 aggressive 吗?

Candidly, I think some people will listen to this and I get in trouble for this. They would say that it's because it's too aggressive.

说实话——有人听完这集会觉得我"太 aggressive 了"。

I love being aggressive. If you do checks on me and people who've come across me, you'll get half the people that say I'm very aggressive and they sort of like it. Half the people will say I'm an [a-hole]. They'll all say I'm competent.

我喜欢 aggressive。你去找跟我打过交道的人——一半人会说"他很 aggressive,我还挺喜欢";另一半会说"他是个混蛋"。但所有人都会说"他很能干"。

On the one hand, it sort of checks the boxes I care about. People think I'm competent. Great.

从我在意的指标看——我能 hold 得住"competent",已经够了。

But the reality is — aggressive can rub people the wrong way. And the reason I just reacted a little awkwardly to the kindness point — if you're too kind and not as direct, not as aggressive, you're wasting time.

现实是——aggressive 确实会让人不舒服。我刚对"kindness"那题反应有点别扭,原因就在这——你太 kind、不够直接、不够 aggressive,就是在浪费时间。

In a world where time is limited and you can't quantify the loss from sugar-coating things — I'd much rather be aggressive and rub some people the wrong way and surround myself with people that want to push hard, than really be surrounded by people who care so much about kindness that they're willing to slow down.

时间有限、"包装话语"带来的损失又无法量化——我宁可 aggressive、惹一些人不爽,然后被一群"想拼命的人"围着;也不愿被一群"为了 kindness 愿意降速的人"围着。

Have you ever rubbed people up the wrong way and you regretted it?

你有没有让人不爽过、然后自己后悔的?

Not really. I guess I just don't think about it much. I don't live in much of a world of regrets, because I live in a world of almost short-term memory.

没怎么有。我想这事不多。我不太活在"后悔"的世界里——我活在一种几乎是"短期记忆"的状态里。

I make a lot of decisions. A lot of them end up wrong. I optimize to go forward.

我做很多决定。很多最后是错的。我优化的是"继续往前"。

Same thing with interpersonal relationships. And I really do want to be surrounded by people who are great, who I can work with for a long time, who I can become friends with.

人际关系也一样。我真心想被那种厉害的人围着——能一起合作很久、能变成朋友的人。

I would love to be surrounded by a core group of family and friends for a very long time. As long as I'm here around all of that — when you're moving fast, you're certainly going to rub people the wrong way at times and you're going to miscommunicate.

我希望长期被一群核心的家人和朋友围着。只要我能跟他们在一起——你跑得快的时候,一定会偶尔让人不爽、会沟通出岔子。

You're going to do something wrong. But if you live in fear of that and you allow that to impact your pace, you'll slow down.

你会做错事。但如果你怕这个、让它影响你的节奏,你就会慢下来。

I'd rather just go fast, and know that that's a risk, and it is what it is.

我宁可全速跑,知道这是代价——就这样。

So funny — I do so many shows, I've done so many shows, I've done this 11 years. But you feel really good shows when you're doing them.

挺有意思——我做过这么多节目,11 年了。但你录的时候是能感觉到"这是好的一集"的。

It's funny — for the first 5 minutes of this, I was like, "Oh, this is going to be good." And what's really hard, actually, is to keep really good — people don't think about this. It's very hard to keep quality with length of conversation.

头 5 分钟我就在想"这集会很好"。但真正难的是——保持好。大家不太提这点。聊得越长,保持质量越难。

You were exceptional. Thank you so much for doing this with me. Thank you for doing it in person. I've loved this.

你太棒了。非常感谢你愿意来。感谢你愿意线下来录。我太喜欢这一集了。

Awesome. Thanks for having me back. It's cool doing it four years after we first [met].

很棒。谢谢你再次邀请我。距离我们第一次见面已经四年了,这种再聚的感觉挺酷。