金刚诞生记：API、奋斗与AI基础设施的未来

他打开一个橱柜，然后，就像，所有的香蕉都掉出来了，因为有个人只吃超大香蕉。他又打开另一个橱柜，里面还有一张床垫，另外几个人正从第二个橱柜里出来。我说，好吧，这是真的。我们当时身无分文。

He opens a closet, and there's, like, all the bananas felling off because there was one guy who's only eating huge bananas. He opened another closet, there's another mattress with other guys leaving a second closet. I said, okay. This this is real. We had no money.

我们我们我们我们只能带着剩下的600多美元搭乘联合航空的航班，我们有90天的时间，要么成功要么失败。我们知道如果无法参赛，就只能破产回到意大利，就这样了。

We we we we just take $600 plus left to go to a United flight, and we had ninety days to just, make it or break it. We knew that if we couldn't race, we would have go back, to Italy broke, and that was it.

我想很多人不知道或者可能没意识到Kong真正发展起来时有多快。那是经历了七年的饥荒期。对吧？我的意思是，基本上就是一直没起色。

I think a lot of people don't know or maybe don't appreciate how fast Kong grew when it actually happened. So there was seven years of starvation. Right? I mean, it was just basically wasn't working.

每年我们都会颁发创始人奖。创始人奖是2555股股票，授予公司最佳员工。对。为什么是这个数字？它象征着255天的奋斗。

Every year we do the Founders Award. The Founders Award is 2,555 stock to the best employee of the company. Yeah. Why that? It's two hundred and fifty five days of struggle.

这只是象征性的。但为了铭记七年的奋斗历程，每年都举行这个仪式。如果现代软件的基石是用每月一千美元在...

It's just symbolic. But to remember seven years of struggle, every year is a ritual. What if the backbone of modern software was forged on a thousand dollars a

借来的沙发上铸就的？今天的嘉宾是Kong公司的CEO兼联合创始人Augusto Agui Mariotti。他将与16z的普通合伙人Martin Casado一起，追溯从米兰车库到打造全球领先API平台的不凡历程。我们将深入探讨突破前的七年磨砺、迫使重生的濒死时刻，以及API如何成为软件的装配线。Auggie还将解释Kong如何融入从微服务到AI代理的一切流程，以及为何下一波自主系统将使API变得比以往任何时候都更重要。

month in a borrowed couch? Today's guest is Augusto Agui Mariotti, CEO and cofounder of Kong. He joins a 16 z general partner Martin Casado to trace an improbable path from a garage in Milan to building one of the world's leading API platforms. We get into the seven year grind before Breakout, the near death moments that forced reinvention, and how APIs became the assembly line of software. Auggie also explains how Kong sits in the flow of everything, from microservices to AI agents, and why the next wave of autonomous systems will make APIs more essential than ever.

让我们开始吧。

Let's get into it.

Agi是Kong公司的创始人兼CEO，Kong前身是Mashape。我们正在讲述他从米兰的一个车库起步，到现在领导一家具备规模的企业并怀揣着IPO野心的故事。

So Agi is the founder and CEO of Kong, which was previously Mashape, and we're covering his background from a garage in Milan to now being the CEO of a at scale company with, I dare say, IPO ambitions at some point.

漫漫长路。我们还有很长的路要走，但这只是沿途的一站。

Long way. We have a long way to go, but that's one of the steps along the way.

野心啊。好吧。所以你在做这种API相关的事情？你是持旅游签证来美国的？

Ambitions. Alright. So you're doing this kind of API thing. You come to The US on a tourist visa?

旅游签证。是的。

Tourist visa. Yeah.

当时的想法是融资

And the idea was to raise

九十天期限。

Ninety days.

是打算融资吗？

The idea to raise funding?

是啊，我们当时身无分文。最后只剩600美元，就买了经辛辛那提转机的联合航空机票。我们被带进二次检查室，结果阴差阳错先到了亚特兰大。

Yeah. We had no money. We just take $600 last left to go to a United flight through Cincinnati. We went through secondary room, and we landed up sorry. Atlanta first time.

辛辛那提是第二个二次检查站。之后我们直奔旧金山，只有九十天时间背水一战。我们知道如果比赛失败，就只能身无分文回意大利，就此结束。

Cincinnati was second secondary room. And then we went we're right to San Francisco, and we had ninety days to just make it or break it. We knew that if we couldn't race, we would go back to Italy broke, and that was it.

那你们是在那时候融资的吗？他们当时

So did you raise in those? They were

在出发前两周完成的融资。

raised two weeks before departure.

你们是在那时候

You raised in those

天使轮。那可是十年前的事了，小额支票。总共5万美元。天使轮融资定义了YouTube团队。哦，真的吗？

Angel rounds. We're talking about a decade ago, right, small checks. It was 50 k. Angel rounds was was defining YouTube teams. Oh, really?

就是这样。有个有趣的故事——你们是怎么结识投资人的？你提到两个有意思的点，能再重复一遍吗？我可能没法再复述了。

There we go. And here's the funny story. How did you get introduced to them? There's two interesting thing that you say, can you redo it again? I won't be able to redo it again.

是的。首先，我们参加了斯坦福创业周。那时斯坦福在二月份举办创业周活动，非常酷。所有创业者都来了，所有风投也都到场，还有个大型创业交流派对。

Yeah. So number one is we went to the Stanford Entrepreneurship Week. At that time, Stanford was doing the Stanford Entrepreneurship Week in February. And it was cool. You made all entrepreneurs, all the VC were coming there, and there was this entrepreneurship mixer, big party.

我们到得晚走得也晚。那时候所有报名信息都是纸质的，我们拿走了那些带着邮箱和注册信息的纸张带回家。那天晚上我熬夜到凌晨5点，写了整整400封邮件。哇。

We arrived, we left late. At that point, it was all paper with all the email and the registration. We steal the things with all the email and the registration, and we walk home with that. And that night till 5AM, I write all the 400 emails. Wow.

内容大概是'嘿，你需要了解米歇尔。派对上没来得及聊，但我很乐意明天给你做个推介'。哇。大概470人没回复，30人回复了，10人表示有点兴趣。最后有五六个人来见了我们。

About, hey, you need to know about Michelle. We didn't have time to catch up at the mixer, but I'm happy to give you a pitch tomorrow. Wow. Think four seventy didn't reply, 30 reply, and 10 were kind of interested. And at the end, five or six went to meet us.

哇。然后有个叫凯文·多纳休的人，他是YouTube的副总裁还是创始团队成员来着？凯文·多纳休。他后来创办了一家做婴儿书的公司。

Wow. And then this one person, Kevin Donahoo, who was one of the VP or founding team of YouTube, came to our What's his name? Kevin Donahoo. Kevin Donahoo. And then he started a company for baby books.

他把公司卖了。但他说：'听着，我觉得这个项目有戏'。于是我们签了张17,000美元的支票。17？千美元。

He sold it. But he said, look, I I think this is something. So we wrote a $17,000 check on our earmatches. 17? Thousands.

是啊，对我们来说17,000美元就是命

Yeah. And for us, 17,000 was life

嗯。

Yeah.

与死亡相比。是的。因为我们从600美元和几百美元起步。然后他带来了另外两个谄媚的牙齿，实际上是六个，总共是16,000、16,000、16,000。我们拿到了15,000。

Versus death. Yeah. Because we came from $600 and few $100. And so then he brought two other fawning teeth or rather six actually was 16,000, 16,000, 16,000. We got 15,000.

问题在于15,000分成三份是6.06、6.06、6.06美元。六太多了。我们不想在资金里出现太多六。所以我们凑整到十七、十七、17，最终得到了51,000的支票。第二个有趣的部分是，我们是在特拉维斯·卡兰尼克的家里谈判并敲定这笔交易的。

Problem The with 15,000 divided three was $6.06 $6.06 $6.06. It was too many sixes. Like, we don't want too many sixes in the captive. So we rounded up seventeen, seventeen, 17, and we got 51,000 check. Now the second funny part is we negotiated and closed the deal at Travis Kalanick house.

特拉维斯·卡兰尼克的房子。我记得有篇文章提到你睡在他沙发上。

Travis Kalanick's house. I remember there was an article about you sleeping on his couch.

是的。还有Airbnb的那些人。但在和这些人谈判时发生了什么呢？他们正在谈判但不愿意。那时特拉维斯在卡斯特罗区有栋房子，叫Jampad，大家都去那儿。亚伦·李也会去。

Yes. And Airbnb guys as well. But what happened during the negotiation with these guys were negotiating and they didn't want. So Trevis had, at that time, was this house up in Castro, called the Jampad, where everybody would go. Aaron Lee would go.

Drawbaugh的德鲁是领头人。我们周末都去那里，而我没地方住。实际上，特拉维斯让我在他那儿住了几周，条件是我每周要给他的另一半做一次卡邦尼拉面。我照做了。后来我们需要帮助。

And Drew from Drawbaugh was the leader. We're all going there on the weekends, and I didn't have a place to stay. So actually, Trevis gave me his place to stay a few weeks, as long as I would cook carbonara for his better half once a week. And I did that. And then we needed help.

他们说，来我厨房，我们谈判。于是我们坐在桌前，我和特拉维斯·卡兰尼克，还有三位YouTube的投资人，谈判这些可转换债券，折扣高达50%，很疯狂，但我们走投无路。我说，好吧，我不想接受这交易，去他的。我说，行吧，我们要走了。那时我很天真，才20岁。

They said, come to my kitchen, we negotiate. So we sit on the table, it was myself and Travis Kalanick, and they were the three YouTube investors, negotiated these convertible notes are like 50% discount, something crazy, but we're desperate. So I say, okay, well, I don't wanna take this deal, screw that. I say, okay, okay, we're gonna leave. And say, and I was very naive, 20 years old.

好吧好吧，交易会回来的。特拉维斯走过来，把手搭在我肩上说，如果这家伙走了，你们就再也见不到他们，也拿不到钱了。哇。所以让我想别的办法。

Okay, okay, deal will come back. And Travis come to me, put their hands and say, if this guy leave, you're never gonna see them again and nor the money. Wow. So let me do something else.

哇。

Wow.

所以你们留在这里商量对策，处理还价的事。我和Agi去洗手间，敲定他的薪资，十分钟后回来。于是Travis带我去洗手间，锁上房门防止他们离开。明白吗？我们在洗手间坐了十分钟，回去后又进行了一轮谈判。

So you guys stay here and figure it out, the counter offer. Agi and I go to the bathroom, we close his salary and we'd come back in ten minutes. So Travis brought me to the bathroom, locked the door of the house so these guys don't leave. Happened? We went to the bathroom and we sit there ten minutes, Go back, another round of negotiation.

又回到洗手间。这样来回三四次后，我们终于握手成交了。然后

Back to the bathroom. So three, four time, at the end, we handshake on the deal. And what

是多少？

was it?

勉强42%可转换债。相当于替代50%股权。哇。而且上限更高。就这样。

Barely, like 42% convertible. It's kind of instead of 50. Wow. And it was a higher cap. And that's it.

然后我们握手，交易就完成了。

And then we shake the hand, and that was it.

太疯狂了。我记得你在Quora上有篇热门帖子，讲如何在旧金山靠14,000美元年收入生活。你当时筹了51,000美元。现在你得让这笔钱撑得越久越好。

That's crazy. So I remember you had the Quora, like, the top Quora post for living on, like, 14,000 a year in San Francisco. You'd raised 51,000. Yeah. And now you had to make that last as long as possible.

那么也许谈谈接下来的计划。是的，继续。

So maybe talk about the next Yeah. Stretch.

实际情况是，显然我们第五次获得了机会。我们必须回去。两周后我们就非法滞留了。所以我们回到意大利，然后带着B1签证回来，这样我们就能在这里待六个月。但具体怎么操作呢？

What happened is, obviously, we got the fifth time. We gotta go back. We are illegal in two weeks. So we went back to Italy, and we come back with a b one, which you could be there we can be here six months. But how?

我们没有社保号，没有信用评分，什么都没有。你得从意大利体系直接进入美国体系，必须适应全新的系统，而我们当时一无所有。因为没SSN，我们甚至无法给自己发工资。

We have no social security number. We have no credit score, nobody. So you have into this American system already from the Italy system. You had to enter into a new system, and we had zero anything. So we couldn't pay our salary because we didn't have SSN.

我们没有合法签证来支付自己，诸如此类。所以公司——你们那时有员工吗？没有，完全没有。

We didn't have a legal visa to pay ourself, all that. So the company- Did you have employees at this point? No. No.

当时就只有你们两个人。

It was just you two.

是我们三个人。当时还有另一位土耳其联合创始人Michele。但我们签了每月1900美元的本票协议，由公司支付，我们三人每月靠1000美元生活。情况就是这样。

It was the three of us. It was Michele was the Turk co founder back then. But we did about a thousand 900 monthly promissory note that the company would give us, and we would have to leave in three with a thousand dollars a month. That would be So

你们当时每月就靠1000美元生活？

you were living off a thousand dollars a month?

在'3。

In '3.

三个人靠一千美元生活，是的。

Three three people were living off a thousand dollars Yeah.

是的。两千美元

Yeah. For 2,000

旧金山。是啊。你们是怎么做到的？

San Francisco. Yeah. How do you do that?

那时更便宜。

It was cheaper.

那这是哪一年的事？

Then what year was this?

2009年2月，2010年2月。

02/2009, 02/2010.

即使在2010年2月那时。

Even then 02/2010.

即使在那时。

Even then.

这不可能。三个人每月一千美元？

It's impossible. A thousand dollars a month for three people?

我来告诉你我们怎么做到的。我们把房子以100美元挂到爱彼迎出租，然后所有人挤在同一张床垫上。

I'll tell you how we did it. So we leave it in on Airbnb somewhere else at a $100 and all in the same mattress.

哦。

Oh.

我们当时要去瓦伦西亚。不知道你们会不会去瓦伦西亚。

And we were going to Valencia. I don't know if you're doing Valencia.

你们有办公室吗？

Did you have an office?

不，我们当时在星巴克办公。我们住在瓦伦西亚，买的是大米、豆子、金枪鱼和意面。原因是必须找到碳水化合物含量合适且可能最便宜的搭配。我想这就是在碳水化合物含量和产品价格之间找到的最经济组合。

No. We were working out of Starbucks. We were living in Valencia and we were buying rice, beans, and tuna, and pasta. And the reason is that we had to find the right amount of carbs and probably at the cheapest way possible. And that was the combination of, I guess, amount of carbs and product the cheapest way possible.

我们就是这样熬过来的。比如，我们从不外出就餐，也不买现成的。你知道，所有事都在住处解决。我们在Airbnb的厨房里做饭，煮米饭、豆子或金枪鱼意面。

And that's how we made it. Like, we never eat out. We never buy it. You know, we did everything in house. We cooked in the kitchen, Airbnb, rice, beans, or tuna pasta.

实际上，我们吃了太多番茄金枪鱼意面，以至于我和马克后来看到这种食物就反胃，因为我们每天都靠这个度日。

In fact, we eat so much tuna pasta with tomato that Mark and I, when we seen our tuna pasta, we almost threw up because we just running on tuna pasta every single day.

这种情况持续了多久？

So how long did this last?

持续到...2010年4月，大概一年零几个月吧。

That last, so March, it was April 2010, a year a year and few months.

这段时间你们在做什么？我知道马可当时在写代码，那你

And what were you doing this time? I can I know what Marco was doing this time? He's writing code. What were

在做什么呢？

you doing?

我当时

I was

写博客、建网站、联系专业人士注册、与网站上每个和我们交流的注册用户沟通，尝试了几个小时。所以混合了前后端开发、HTML、CSS、切片，同时也在和投资人洽谈，因为你们开始筹备种子轮融资。是的。尝试做些招聘但从未成功，因为没人愿意为随时可能消失的非法意大利人工作。所以犯了很多错误，就这样，好吧。

writing blog, building the website, calling pro the the sign up, talking with every sign up on the website that was talking to us and try a few hours. So a mix of b b devs, HTML, CSS, slicer, talking with investor too as you start to build the seed rounds. Yeah. Trying to do some recruiting that never worked because nobody wanted to work for illegal Italians that will disappear. So a lot of mistakes and this, yeah, figure Alright.

事情

Things

带我们进入种子轮。所以最终有人决定给你们正式投资。

take us to the seed round. So somebody decided to give you a proper round at some point.

一年后我们重新架构。我们去了檀香山的莫阿纳海滩，思考着，所有这些API都在整合。我想现在终于明白了这个词，但十年前、十五年前时机还不成熟。当时想通过API拖拽即时构建应用，我们太超前了。

So then a year later, we rearchitect. So we went to Honolulu on the beach, Moana Beach, and we thought about, okay, all these APIs were aggregating. I think the word is actually, we see now, finally now, but ten, fifteen years ago, it wasn't ready. They were to like build apps on the fly through APIs with drag and drop. We were too far ahead.

我们缺少关键部分。但我说，嘿，我们正在整合大量API。实际上那条可视化装配线的概念仍然成立，我们只需调整方向，建立一个让所有API生产者和消费者汇聚的市场。

We're missing pieces. But I say, hey. We are wrapping a lot of APIs. Actually, that assembly line visual still hold. We just have to pivot it and build an API marketplace where all the APIs producer and consumers, they can come together.

于是我们就这样做了，重新上线产品，压缩了所有内容。好吧。

And so we people that, we relaunch it, we then crunch all the stuff. Okay.

最初我们做的是这种拖拽式、可组合的构建工具。看到这么多公司都经历同样的历程真是有趣，然后你就会觉得——好吧。

So so originally, we're doing this kind of drag and drop Drag and composable. Builder thing. It's so funny how many companies go through this exact journey. And then you're like, okay

用的是HTML5。

With HTML five.

哇，然后你就会觉得——好吧，不，我们需要建立一个市场平台。

Wow. And then you're like, okay. No. We need to build a marketplace.

没错。因为API经济即将到来，不只是应用，API才是未来。于是我们搭建了这个市场平台。

Exactly. Because, okay, the economy this API economy will come. It's not just apps. It will be APIs. And we build a marketplace.

平台上线后，我们获得了长尾流量上的互动。那已经是2011年了，在Piero de Noir之后。我们快速学习、快速迭代，然后就开始融资...等等。

We launch it. We got these interactions on long tail. And that was already the 2011 Yeah. After Piero de Noir. So we learned very fast, boom, boom, boom, and and then we raised Wait.

夏威夷那笔是最后的资金吗？

Was the Hawaiian the last of the money from

不，不是。因为当时从旧金山到檀香山的机票大概300美元。我们拿着5万1千美元，在米慎区那栋历史建筑里靠金枪鱼意面、米饭和豆子度日，那地方真是与世隔绝。

No. No. Because San Francisco to Honolulu, I think at that time, it was like $300. So I say with our 51 k, you know, we're living like on tuna pasta, rice and beans, in Valencia, in Mission, in this historic building, middle of nowhere in Mission.

你们是安德吗

Are you guys And

就像，这个正在以同样的方式运作，事情就是这样

like, this is working in same and things is like

这简直难以置信。

It's unbelievable.

某个时刻，记得吗。就记得早上醒来，我必须去檀香山。我们可以在海滩上散步，思考我们的未来或没有未来，因为我们知道这会没有未来。我们的资金开始耗尽，我们需要转向。就在那时我们有了这个想法，让我们尝试转型。

Some point, remember. Just remember waking up in the morning, I need to go to Honolulu. We could do our walk on the beach and thinking about our future or no future because we knew this was going no future. Our money were starting to drain and we needed to pivot. And that's where we had the idea, let's go for the pivot.

然后我们回来，做了件疯狂的事——我记得我们建网站时，迈克尔快疯了。第三位联合创始人负责所有Java后端开发，快速上线。我们还得到了TechCrunch的一点报道。那时候还是读写网时代。没错。

And then we come back, we did the crazy I I I remember us building the website, Michael was going crazy. The third cofounder was building all the Java back ends and the launch it fast. And I've got a little bit of TechCrunch press. At that time, it was read write web. Yep.

另一个是Mashable。

This other one was mashable.

对。对。

Yep. Yep.

就这样，我们莫名其妙地被这一切所覆盖，现在情况更是加剧了。然后突然之间，他们就启动了种子轮融资阶段。谁投的呢？当时我们经历了无数次反复，但最终是由NEA领投种子轮，接着Index跟投280万，之后我们又获得了一大堆投资，然后——

And so all the somehow we got covered by all of that, which now, like, increased. And then and boom. And then they start the seed round phase. From who? So at that time, we went through a lot of lot of iterations, but where we ended up was NEA leading the seed round and then index co leading 28, and then we got a bunch and then-

至于种子轮，其实是Volpe负责的——

And the seed, so Volpe did the-

没错。实际情况是，最早的投资支票来自当时CRV的George Zacari。他虽然没有领投，但他是第一个承诺10万美元的人。就这样，他表态说CRV投10万。那时候他们还有个种子计划，这在当时非常罕见。

Yeah. So what happened is, actually the first checks were George Zacari from CRV back then. He didn't lead the round, but he was the first 100 ks commit. And so with that, he say, we have CRV, 100 ks. At that point, they had a seed program, very unusual back then.

所以我们融了280万，他说要领投这轮。那时还启动了一个500万的种子计划。后来我遇到了刚搬到旧金山的Mike Volpe，他负责伦敦Index美国业务。当时他和Danny Reimer应该在某处。他们也启动了天使种子计划，并投了几笔大额支票。

So we went 28, and he said, we're gonna lead this round. At that point also, start a seed program, 500. And then I met Mike Volpe that just moved to San Francisco to London Index US. It was like him and Danny Reimer should have somewhere. And they also started Angel, a seed program, and they wrote on other big checks.

接着我们获得了一长串跟投，主要是杰夫·贝索斯和埃里克·施密特通过基金投的。所以我们得到了...这个你是怎么看到的

And then we got a long tail, which was largely Jeff Bezos and Eric Schmidt into the funds. So we got How this did you just see

贝索斯和埃里克·施密特？

Bezos and Eric Schmidt?

所以这又是他那套DPD理论（我不知能否复制）。具体是这样的：杰夫·贝索斯——我知道他对市场平台业务、API和开发者有独到见解。AWS那时刚开始腾飞，还是个不为人知的秘密。

So again, that's his sitting DPD that I don't know we can replicate. So what happened is Jeff Bezos. So Jeff Bezos, I knew he was special about marketplace as a business, APIs, and developers. AWS was just starting to take off. It was this hidden secret.

现在是2011年吗？

Is it 2011?

是的。当时Alenso大概是一家500亿的公司，而不是800亿的公司。于是我就雇用了他们家族办公室的律师来联系Meshe。我们雇人安排会面，在会面过程中顺便提到，既然你也是家族办公室和Bezos Expeditions的律师，能不能

Yeah. So Alenso was maybe a 50,000,000,000 company than an $80,000,000,000 company. So what happened is I hire the lawyer of his family office for call, for Meshe. And after we hire for doing the sit down, as the sit down was going along, say, by the way, since you are also the lawyer of the family office and Bezos Expeditions, can

帮我引荐一下？能把我介绍给杰夫·贝佐斯吗？

you introduce me? Can you introduce me to Jeff Bezos?

我们就是这么做的。这就是PG式的陌生电话。当时他正和他兄弟在德克萨斯州公路旅行。你其实很想加大投资力度。砰的一声（就成功了）。

That's what we do. So it's a PG cold call. And then he was on the road with his brother, going around with his brother on the road in Texas. And you actually really want to put more and invest. Boom.

那就是杰夫·贝佐斯。你能从杰夫·贝佐斯那里得到什么？我们获得了品牌资源，包括年度战略晚宴等。但显然最重要的是这个大品牌。

So that was Jeff Bezos. What do you get from Jeff Bezos? You get the brands that we get a dinners, year strategy of things. But obviously it was the big brand.

然后

And then

第二个人是埃里克·施密特。我们当时共用这个联合办公空间。我们是唯一工作到深夜的初创公司，其他做邮轮版Expedia的初创公司都走了。他们刚获得NEA和埃里克·施密特投资。最后他们看到我们工作到凌晨三四点，会在CHU离开。当他们知道我们在融资时，投资人问：'谁比你们更拼命工作？'

second, Eric Schmidt, we were working this co working space. And what happened is we were the only startup that stay the latest in the night after these other startup that was doing Expedia for cruise ships. They just raised from NEA and Eric Schmidt. And at the end, they saw us working till 3AM, 4AM, they will leave at CHU. When they knew we were fundraising and the investor asked, who's the hardest worker than you?

然后说说我们旁边这些人。他们把我们介绍给了埃里克·施密特基金，埃里克·施密特就是这样投资的。

And say, these guys next to us. And so they introduced us to the Eric Schmidt fund and that's how Eric Schmidt invest.

哇。

Wow.

完全没关系。

Totally unrelated.

好的。那么现在这就把我们带到了A lot

Alright. So now this brings us to A lot

很多运气。

A lot luck.

那你总共筹集了多少种子资金？

So what was your total seed fund you raised?

那时候金额很大，是150万。感觉就像——哇，超级大的种子轮。我得说是在2011年2月。

At that time, it was very big. It was 1.5. It's like, wow, huge seed round. I gotta say In 02/1011.

不得不说这很乐观。用这种乐观的回顾视角来看，三个人挤一张床垫睡了一年，然后还说自己是幸运的。

Gotta say there's a lot of optimism. It's an optimistic retrospective view to be like, was three of you on a mattress for a year, and then you say you were lucky.

非法地。听起来

Illegally. That sounds

挺倒霉的。非法地。

pretty unlucky. Illegally.

进错街区的监狱。

Wrong hood jail.

好了，现在你系好安全带，开始匹配形状了。

Alright. So now you have your seatbelt and match shapes going.

于是我们调高了座椅。好了，现在来真的了。我们得搞定这些签证，得去见他们。

So then we raised the seat. Alright. Now it's real. We gotta get these visas. We gotta meet them.

所以我们回到意大利，去了美国大使馆，巴拉巴拉，弄推荐信。其实萨姆·奥特曼当时还是Looped的CEO时，就是给我写推荐信的人之一。

So we went back to Italy, go to the American embassy, blah blah blah, do their letter of recommendations. Actually, Sam Altman was one of the guys that wrote me a letter of recommendation when he was CEO of Looped.

哦，开玩笑的。

Oh, kidding.

那次是在我们去参加NEA静修会的路上，我在大巴上坐在他旁边，我们一起去圆石滩参加NEA静修会，共处了三小时。我随口说，其实我在这里是非法居留，你能帮我写封O1签证推荐信吗？没开玩笑。然后他真的写了，信里把我夸得天花乱坠。

Which I met on the we were going to the NEA retreat, and I see next to him in the bus, and we went to Pebble Beach to the NEA retreat, and we spent three hours together. And I said, by the way, I'm illegal here. Can you write me a letter of recommendation for my o one visa? No kidding. And he wrote me so I have this big thing of how great I am.

山姆·奥特曼写的？太棒了。

From Sam Altman, that's great.

山姆·奥特曼写的推荐信，确实厉害。信里说无论如何都要让我入境。我们凑齐五封推荐信后，终于拿到了O1签证。有了签证就能租WeWork办公室，开始招人，团队发展到七人。马可也拿到了签证——他因为法律书籍的事连大学都没上过，他的签证是最难办的。

From Sam Altman, that's great. You gotta get this guy in the country no matter what. So we got these five letters, and and I get this o one visa, and finally I got the visa. So if I can come here, I can get a WeWork space, and we start to hire, and we got seven people, and Marco also got the visa. Marco has also, because of the law book, he never got to college or nothing, so it was the hardest visa to do.

后来我们想办法给他弄到更多推荐信，把他带过来了。我们在这里安定下来，诸如此类。那年我们搭建业务，逐步发展。但你知道，当时想融A轮起码需要百万美元营收额，对交易平台来说要百万总交易额。

So we figured it out, more letter of recommendations for him, and we brought him there. We stay here, blah blah blah. We build that year, we grow. But you know, at that time, like you needed like a million, let's say a million revenue to take a good series A. For a marketplace, you need a million in gross volume.

我们当时只有5万美元，用户增长很快但收入不高。但我们还是经历了三四个月的A轮融资拉锯战，最后CRV领投，接着是Index...

We were like at 50 ks, so a lot of traction, but not a lot of revenue. But we went through a three, four month series A race, and CRV led and then index

是不是那个开发...

Was it was it dev. That

是的。那个？对。因为乔治·扎卡里更偏向消费者领域，所以他通常把链接转给Dev那边，那边更侧重企业级业务。

Yeah. That? Yeah. Yeah. Because George Zachary was more a consumer, so he passed the the link to Dev dot just generally more on the enterprise side.

我记得Dev。他周日早上6点打电话约我去散步，就为了决定投不投资。他知道我是夜猫子，却偏偏选周日清晨6点让我去帕洛阿尔托，我还得租辆Zipcar，那时候还没有Uber。我赶过去后散了几次步，最终决定投资。

I remember dev. Call me, like, Sunday at 6AM to go for a walk, to decide to invest or not. And he knows I'm a night owl, and and he just Sunday at 6AM to go to go down to Palo Alto, which I had I had to rent a Zipcar, they just weren't Zipcars. There was no Uber. And and go down there and and and then go for a few walks, I decided to to invest.

因为他一直坚信API就是生产线，这点显而易见。虽然不确定Marketplace是不是最佳落地方案，但这个主题他非常认同。而且他就像我们一样，亲眼见过我们挤在同一张床垫上睡觉。

Because it was it was always actually a big believer on APIs as assembly line. You could see that. Yeah. Maybe it wasn't sure Marketplace was the right execution, but the theme, he was a big believer. And I think he like us, like, he came to see us sleeping in the same mattress.

他想确认我们不是编造悲情故事。那确实是真的。他来到我们在南公园的Aker House——当时我们整个七人团队吃住工作都在那里——打开衣柜时，熟透的香蕉哗啦啦掉出来，因为有个家伙只吃超大号香蕉。

He wanted to really see that we're not, like, bullshitting the big drama story. It was really drama. Yeah. Actually, he opens he came to the Aker House because we moved to Aker House in South Park, Quebec at that time where we're sleeping and working there with the seventy team. And he opens a closet, and there's, like, all the bananas felling off because it was one guy's only eating huge bananas.

接着打开另一个衣柜，又发现一张床垫，还有人在里面睡觉。我说：好吧，这确实是真的。

Yeah. You open another closet, there's another mattress with another guy sleeping in a second closet. I said, okay. This this is real.

哇，你们当时A轮融了多少钱？

Wow. And how big was your a?

六百五十万。

Six point six point five million.

哦，对你来说那肯定是一大笔钱吧。

Oh, so for you, that must be must have been a lot of money.

650万。是的。我用这个PPT去融资。我们打算筹集1000万系列。在当时那可是件大事。

6,500,000.0. Yeah. I went out with this power point. Let's raise a 10,000,000 series. At that point, that was, like, the big thing.

我们最终拿到了650万，我说这个数字时感觉就像当初看到它时一样，觉得可能有戏。

We ended up at 6.5, and I say that science it was like when we saw it, it was like, yeah. Like, maybe we got a shot.

所以某个时刻你认定市场失灵了。

So at some at some point, you decided that the market wasn't working.

市场确实失灵了。于是我们融了这笔钱，扩招团队，搬了新办公室，25个人，就像典型的POS赛道蜜月期。我记得大概半年后，业务还是毫无起色。有次开董事会，我们只能给董事们煮意面吃，因为那会儿比你加入我们还要早两年。

The market wasn't working. And I think So we raised this and we hired more. We moved to a new office, 25 people, like the usual POS race honeymoon. And I remember this thing like six months after, we the business wasn't really doing anything. So it was one board meeting that we just cooked pasta for the board members at that time, because that was like two years before you joined us.

而且业务上根本没什么可汇报的。

And it was there was nothing to talk about from business.

问题在于你们无法变现，或者说存在根本性问题，这类平台型生意本来就很难做。

The issue was you is you couldn't monetize or there was a problem, like these marketplaces are very tough.

当我与Airbnb创始人同住时，我领悟到市场平台可能是世界上最大、最具威力的商业模式。一旦获得流动性，你甚至无需持续创新，它本身就难以被颠覆。看看eBay，它坚不可摧。

So the idea of the marketplace that when I was staying with Airbnb founders, I learned that the marketplace are could be the biggest, more powerful business in the world. Like, you don't even have to innovate once you get liquidity. It's just you can't disrupt. Look at eBay. You can't kill.

是的。当时我感觉有这么多API接口，完全可以构建一个具备流动性的市场。它本可以成为像AWS那样的存在。但我们注意到，我们做的是开发者API市场平台——要让市场运作，你需要聚集大量低影响力的长尾用户。

Yeah. So I felt like, there's all this API. You can build a market with liquidity. It'll be like the AWS or And so we we noticed though that in our case, it was a developer API marketplace. So to a marketplace to work, you need to have a long tail of low power people.

如果资源集中在少数高影响力用户手中，市场就无法运转。就像Airbnb需要数百万普通房东参与。这是第一点。第二，市场需要一定排他性——就像获取房源必须通过平台门户。而API呢？你可以直接搜索官网，也能通过市场获取。

If you have concentrated in a few high power marketplace doesn't work. Like Airbnb houses, like millions of people with low power. So that was one. Two, you need to have exclusivity somewhat like the door to access that marketplace supply has to be through the marketplace. APIs, you could Google and go to the website, you can go through the marketplace.

当时的公开API呈现长尾效应：推特或Stride等30个主流API举足轻重，另外3000个影响甚微。第三关键是质量——运营云市场时你无法确保供应质量，总会因此受指责，且缺乏信任机制。

APIs at that time, power of low for public APIs was on the Tweet or the Stride, or 30 that matters and 3,000 that didn't matter much. So it was also this long tail power. And third thing is was quality. Like you couldn't, you were running a cloud marketplace, you couldn't actually maintain the quality of the supply, you would always get blamed for it. And there was no trust.

正因如此，虽然总营收达到150万美元，但它始终未能成为Airbnb或谷歌那样的巨头，利润率甚至不到10%。由于AWS上的持续亏损，我始终无法让这个模型的财务逻辑成立——即便规模扩大也无济于事。

And so I think because of that, it got to a million and a half in gross revenue, but it never became this Airbnb That or Google or is like 10%. Yeah. It was it was also losing margin on AWS. So they I could never make the economics in that model. I could never make the economics works, even if this thing would have scaled.

于是我们面临抉择：资金即将耗尽，团队精疲力尽。当时我们已为这个长尾市场构建了强大的API引擎——能处理2万个API的网关系统，涵盖计费、限流、路由、缓存、认证授权和日志等全套功能。经过三次迭代，最终版本堪称完美。这时我们突然意识到：每家公司都将成为API公司。

So then we go there. It's like, okay, this is where we're, we're burning out, we're running out of money. And so we build this massive API engine behind the marketplace, API gateway that was powering 20,000 APIs of this long tail, doing billing, rate limiting, routing, caching, authentication, authorization, logging, all of that. We built it three times, and the third time was the great one. And we say, wait a second, every company will become an API company.

何不将这个引擎开放给全世界？这就是MongoDB开源故事的开始。于是——轰然一声——我们开源了Kong。

Why we don't take this engine and we give it to the whole world? And that was the beginning of open source, Mongo. And so boom, we we open source Kong.

当时你们开源Kong时的资金跑道有多长？

What was your what was your runway at the time that you open sourced Kong?

我们在2015年4月15日开源。当时不得不接受200万美元的过渡贷款来维持一年运营，因为

So we open sourced in April 15. We had to take a bridge of 2,000,000 to go another year because

是内部过渡贷款吗？

was it an insider bridge?

是的。我们不得不申请延期，因为资金已经耗尽。

Yeah. We we had to take an an extensions because we were out of gas.

是来自DevTutt和Mike的吗？

From from DevTutt and Mike?

对。他们给了我们200万的延期，因为我们资金链断裂，否则公司就撑不下去了。

Yeah. They gave us an extension on day two millions because we're out of gas like we were would have died otherwise.

哇。我记得Kong发布时确实引起了巨大轰动。那是2015年2月对吧？

Wow. So you you released I remember when Kang released. It was actually a really big deal. When was that? 02/2015.

'15年。

'15.

是的。2015年2月。

Yeah. 02/2015.

那可是件大事。我记得当时突然就爆发了。对。就是那时候我们开始讨论你加薪的事。你在提高标准。

That's a really big deal. I remember Took off boom. Yeah. So this is when you and I started talking about your raise. You're raising the b.

你来得正是时候。之前这些人完全没道理可言。好吧。现在

You came at the right time. So before, like, these guys make no sense. Okay. Now it

开始说得通了。不过那次加薪对你来说也很艰难，毕竟公司已经运营七年了。

starts to make sense. Well, that was even a tough raise for you just because the company had been around for seven years.

你得

You had

看看市场情况。比如Kong刚推出，正在起飞阶段。但也就几周时间，可能几个月。说实话，让我下定决心的原因是，我们做了所有工作，GitHub星标数也验证了。你的故事非常了不起。

to take look marketplace. Like, Kong had just come out, and it was taking off. But, like, it only been a couple of weeks, maybe a couple of months. And so, honestly, I'll tell you what what did it for me is, you know, we were we're doing all the work and, like, the GitHub stars checked out. Your story is phenomenal.

你对业务的掌控很棒，但这还不够。然而在我们做尽职调查时，不断遇到这些巧合，比如有人用过Kong并且很喜欢。我记得有人住在你的Airbnb里，爱上了Kong，然后你转发给我。所有这些氛围都围绕着Kong，结果我就觉得，这显然是一种现象。

Your command of the business is great, but it just wasn't enough. But then while we were doing the diligence, like, kept getting these kind of serendipitous, like, somebody had used Kong and loved it. I remember didn't like somebody stay in, like, your Airbnb and, like, loved Kong and, like and then you forwarded that to me. And so they're just kinda like all of this you know, zeitgeist around Kong. And as a result of that, I'm like, oh, this is clearly a phenomenon.

Gary，我记得当时我一直在用邮件轰炸你，每封都是潜在客户或用户对Kong的评价。你不停地说Kong、Kong的使用、Kong的使用，没完没了。简直要把收件箱挤爆了。

Gary, I remember I was spamming you with emails of every prospect or customer user they were saying. You just you say, Kong, Kong usage, Kong usage, just nonstop. Like, it was blowing up the inbox.

是啊，是啊。那么你们离资金耗尽还有多久？

Yeah. Yeah. Yeah. So how how close how close are you to running out of money for it to be?

违约后，只剩几周时间了。

So after breach, there were weeks.

天哪。

Jeez.

是的。但我非常

Yeah. But I was very

这辈子再没什么能让你紧张了，因为你至少两次在生意上濒临绝境。太多次了。因为我记得我们去办公室时，感觉死气沉沉的。基本上就像没救了。

Nothing nothing is ever gonna stress you again your entire life because you've been so close to at least business Two times. So many times. Because I I actually remember when we went to the office, it felt dead. It felt like basically No.

不，我们当时已经完蛋了。

No. We were dead.

是啊，只是我们当时不知道。你们离完蛋就差两周了。我是说，感觉我们当时

Yeah. We just didn't know about it. You were like two weeks from being dead. I mean, it felt We were

就像大黄蜂一样，你知道，它能飞，但按理说它不该会飞。

just like Bumblebee, you know, that it flies, but it's not supposed to fly.

我们不断...我们不断前进。不，我记得非常非常清楚

We keep we keep going. No. I remember that very, very

没错。你当时还说，伙计们，这家伙已经在那儿待了很久。我觉得他们可能会，你知道的，筹到钱然后就会放弃。你当时...我说过，我说这很艰难。我觉得如果我们回头说，不知道我们能否复制发生的整个流程，所有发生的那些事。

Yeah. And you were like, you're like, guys, this guy's been there a while. I think they're gonna, you know, whatever, they raise money and then they're gonna give up. You were I said, I said, it was tough. And I think I think if if we go back and say, don't know we were able to replicate all the sequence that happens, all all the things that happens.

我记得我们确实去了。然后我们和马克·安德森一起吃了顿超棒的寿司，就你和我，我们敲定了协议。还有马可呢？

And I remember we went yeah. And then then we had that that great sushi with with Marc Andreessen, you and I, and we sealed the deal. And Marco?

是的。2016年我们完成了交易。自那以后，公司表现就非常出色。我是说，最近还有个公告说突破了100亿，其实那已经是挺久之前的事了。

Yeah. At the 2016, we closed. And since then, the company's just been remarkable. I mean, there was kind of an announcement recently that across a 100,000,000, which is now actually quite a while back.

一年半前。是的。

A year and a half ago. Yeah.

一年半前。

A year and a half ago.

是的。我们那时，一年半相当于十年，当你投资时，我们的年度经常性收入还不到一百万。

Yeah. We Yeah, was in a year and a half is ten years where we were like, when you invested, we were ARR less than a million.

是的。是的。

Yeah. Yeah.

年度经常性收入不到一百万。

Less than a million of ARR.

你还记得吗？

Do you remember do you

记得吗？甚至连一百万年度经常性收入都不到。平均K值可能。

remember Not even a million ARR. Average K probably.

你还记得有一次吗？我记得很清楚。你第一年表现非常出色，然后我告诉你，听着，如果你能达到1000万或那个目标，我就给你买辆车。还记得吗？

Do you remember once? I remember so yeah. So you had a great first year. And then I told you, said, listen, if you if if you hit 10,000,000 or whatever it was, I'll buy you a car. Remember that?

我记得。哦对了，我们现在把它放在新办公室了，我得给你发张照片。

I did. Oh, actually, I we have it in new office now. I need to send you a picture.

关于这个有个有趣的故事。你知道，你做到了。我记得你那年的业绩增长了大约10倍。

Well, there's a funny story about this. And so like, you know, you you do it. So like, I think you grow like 10 x that year.

那年，我们的年度经常性收入从200万增长到了1000万。

That year, we went from 2,000,000 of ARR Yeah. To 10.

没错。

That was right.

而原计划，我记得是六七百万左右。

And the plan, I think, was six or seven.

对，对，对，对。然后我

Yeah. Yeah. Yeah. Yeah. Then I

回到了x。

went back x.

我去咨询了合规部门，问能不能给奥吉买辆便宜车。他们告诉我礼物的最高限额是多少来着。所以我心想，糟了。最后我只能给你买了能找到的最好的模型车。

I went and I checked with compliance, and I'm like, can I buy Auggie a cheap car? And they said the maximum for a gift is whatever it is. And so I'm like, crap. So I ended up buying you the best model car I could find.

来自日本的，我想你

From Japan, I think you

对，对，对。我确实花了很长时间挑选这款模型车。

Yeah. Yeah. Yeah. I spent a long time actually looking at this model car.

有'2 80 GT'型号吗？

Is there a '2 80 GT?

没错，就是这样。太棒了。所以我特别想换挡。

Yeah. Yeah. That's right. So great. So so I would love to shift.

所以我认为这是我所知最了不起的硅谷创业故事，我亲身参与的，某种程度上算是经历了

So I I think this is the most remarkable Silicon Valley startup story that I know of, that I'm close to, and I kind of having went

而且看到总有一天会拍成电影。

and seen That that will make a movie one day.

他们总有一天得拍成电影。是的。我很想稍微转向你如何看待产品、如何看待市场，然后逐渐转向人工智能。所以实际上你已经见证了几次转变，对吧？

They will have to make a movie one day. Yeah. I would love to, like, kinda shift towards how you think about product, how you think about markets, and then kind of move towards AI. So you actually have seen a number of shifts now. Right?

你见证了云计算的兴起，见证了向API的转变。所以或许可以谈谈你如何看待当前AI带来的转变。它是根本性的不同吗？还是相同的？

You saw the cloud come. You saw the shift to APIs. So maybe just talk a little bit about how you view this current shift with AI. Is it is it fundamentally different? Is it the same?

这会改变你作为领导者的自我认知吗？比如，你的高层视角是怎样的？

Does it change how you think about yourself as a leader? Like, how how is your view at a high level?

所以关键点在于...我先从API说起。显然，我们把自己打造成了一家API基础设施公司。

So the the the big thing is I I'll start with API first. So, obviously, we we we built we become this API infrastructure company.

也许吧。是的。也许在我们深入之前，你应该先描述一下Kong到底是什么。

Maybe. Yeah. Maybe maybe you should just describe what Kong is right before we actually do that.

于是，我们通过这次转型，开源了Kong API网关，一炮而红，成为了一家以Kong Inc.为品牌的企业公司。我们开始构建各种API基础设施来运行、管理和保护你的内部或外部API。假设你有软件、微服务、成千上万个API，我们就像是确保它们运转的高速公路——提供限速功能。如果把API比作汽车，我们就是护栏、减速带、测速摄像头、加油站、维修站、防护栏等等一切，甚至救护车。我们提供所有基础设施，确保API连接能为大小公司顺畅运行。

So, we left with this pivot and we open sourced Kong API Gateway, took off and we became an enterprise company, became with the brand Kong Inc. And we started to build all sort of API infrastructure to run, manage, and secure your internal or external APIs. So, have you software, you have microservices, you have 1,000, thousand of APIs. We're kind of these highways that makes them run, and you have rate limiting. Like if APIs were cars, you got guardrails, speed bump, speed cameras, gas stations, stalls, bombs, guardrails, everything, ambulance, and that we provide all the infrastructure to make sure that the API connectivity runs for small company, big company, all of that.

现在，在你提到的这次转型中，我认为真正推动云API爆发式增长的是工作负载从本地迁移到云端。第二次转型是将单体架构拆分为微服务，这催生了越来越多的API、大数据等，也带来了事件流处理等。因此，动态数据的规模正变得远超以往静态数据或使用中数据的量级。

Now, in this transition you mentioned, what really drove, I think, this explosion of cloud APIs was the workload moving from on prem to the cloud. The second transition is breaking down the monolith to microservices, it creates more and more APIs, the big data and all that, it creates event streaming, all that. So this just data in motion is much higher than, it's becoming much higher than like data rest or data in use than before.

但从某种角度看，你创业时正是乘着一场重大转型的东风，对吧？就是架构解耦和云迁移这两大关键转变。

But in a way, when you were starting the company, you were drafting on a big transition, right? Which was the breaking down of the model and then the shift to cloud. The key. Right?

没错，那确实是关键。早期的API解决方案都是针对传统架构设计的，根本无法适应向云端转型、架构微服务化以及大规模分布式系统的需求。比如我们在API管理领域首创的控制平面与数据平面分离方案——当时根本没人这么做。

So that was the key. Yeah. I think before there were all these archive legacy solutions in APIs, but they were not built for the transition to the cloud and breaking down the model into microservices and high scale decentralized architecture. I mean, you know, your time on this year, like we kind of invented the control plane, data plane separation at the API management, API control plane level. Nobody was doing it before.

不得不说，很多人可能不了解（或低估了）Kong真正爆发时的增长速度。其实之前经历了七年的蛰伏期，基本上毫无起色对吧？

I gotta say, I think a lot of people don't know or maybe don't appreciate how fast Kong grew when it actually happened. So there was seven years of starvation, right? I mean, was just basically wasn't working.

确实。你知道我们每年都会颁发创始人奖——这个奖项会给公司最佳员工2555股期权。

No. Actually, know, and you told, you know, despite, every year we do the Founders Award. Yeah. The Founders Award is 2,555 stock to the best employer of the company. Yeah.

为什么是这个数字？它象征着255天的奋斗历程（合计七年），让我们铭记每一天。七年啊，这真的很了不起。

Why that? It's 255 days of struggle to remember every day, which is seven years. That's amazing. It's seven years. I

这个典故我倒不知道。

didn't know that.

公司每年都要经历2055这个数字，这纯粹是象征性的。

Every year the company has to go they get 2,055. It's it's just it's just symbolic.

是啊，是啊。

Yeah. Yeah.

当然，为了纪念七年的奋斗历程，每年都是一种仪式。

Of course. To remember seven years of struggle, every year is a ritual.

还有那次疯狂的飞身接球，你知道的，马可做了最棒的功夫动作。他把产品推向市场后，销量就追上来了。一旦成功，公司就以惊人的速度成长起来。

And there was this crazy diving catch, which, you know, Marco does the greatest Kong. He throws it out on the market and it catches up. But once it did, the company grew incredibly quickly.

我们刚开始时有45个竞争对手。

And we had 45 competitors when we started.

哦，我记得这事。

Oh, I remember that.

是啊，现在大概只剩30个了。

Yeah. That is probably 30.

但即便如此，在当时确实如此。Kong很快就脱颖而出。你知道，近年来它显然已成为这一领域的领导者。我认为，这种情况只有在市场存在明确需求时才会发生。市场某种程度上也帮了忙，比如MuleSoft被收购，Apogee也被收购了。

But even then at the time yeah. Well, Kong broke away pretty quickly. So I've been, you know, in in in recent years, it's just clearly been the leader on this. And that only happens, I think, if there's just clearly a market need. The market also kinda did you a favor in that, like, MuleSoft got acquired and Apogee got acquired

没错吧？确实是这样。2017年MuleSoft被收购了。

too. Right? It was right. Yeah. '17, MuleSoft got acquired.

Apogee是在2016年被收购的。

Apogee got acquired in '16.

没错。某种程度上来说，市场经历了整合，而你们作为领导者脱颖而出。你们成功驾驭了从云到微服务的转型，构建了API的控制平面。

That's right. Yeah. So, like, in in a way, kind of like the there was consolidation and you broke out this as leaders. So you kind of navigated this transition from, you know, cloud to break up the microservices. You know, you built the control plane for APIs.

现在你们被视为API领域当之无愧的独立领导者。Kong是毫无疑问的独立领导者。但如今你们正面临另一个市场转变——AI浪潮。你认为这对你们是直接影响、相邻领域还是增值机会？

You're considered now like the, you know, the the leader independent for sure. In the API space, Kong is the leader independent. There's no question of that. But now you're facing another market shift, which is kind of this movement to AI. And so do you view this as directly impactful, adjacent, accretive?

作为CEO，你如何看待这一变化？

Like, how are you as CEO viewing this?

是的。我认为这并非我们刻意为之，但市场通过创建更多API将AI浪潮推向我们。这意味着什么？智能体将以与人类完全不同的方式使用互联网——不再有传统网站、滚动点击、上下翻页的应用程序。

Yeah. I think and again, it's it's not like was like any me those things, but the market came at us in AI by creating more APIs. What that means? Agents are gonna consume Internet in a very different way than how Xueman did. There's not any more websites, scroll, click, up and down, applications.

他们仍会存在，但不再像以前那样重要。代理将通过编程方式交换劳动力，通过编程接口完成任务，无论是传统的APRS、MCP协议——就像是API界的多邻国，让它们能说英语。无论采用什么协议。但机器消费互联网的方式将与人类截然不同。人类通过用户界面使用互联网。

They're gonna be there, but not as relevant as before. Agents are gonna programmatically exchange labor, get tasks done through programming interface, whether it's classic APRS, MCP protocols, which is like Duolingo for APIs, makes them speak English. Whatever it is, the protocol. But machines consuming Internet is gonna be very different than human consuming Internet. Human consuming Internet would be through UI.

机器通过编程接口消费互联网。这正是我们正在捕捉、赋能并确保企业思考AI连接性的重大转变。归根结底，幕后的一切都是API。

Machine consuming Internet is through programming interface. That's the huge shift that we we are now capturing and powering and making sure the enterprise think about AI connectivity. And at the end of the day, behind the scene, it's all APIs.

我们昨天开了董事会，对吧？说实话，让我印象深刻的是AI领域竟然有这么多平淡无奇的应用场景。我们可以讨论代理，这确实值得探讨，我也同意你的观点。但同时也存在大量基础性工作，比如密钥管理——很多公司在这些AI模型上投入了大量资金。

We, you know, we had that board meeting, what was it, yesterday? Yes. And what, you know, what I found actually pretty remarkable is how many more banal use cases there are around AI. Right? We can talk about agents and I think it's worth talking about and I agree with you, but it also feels like there's like a lot of just basic stuff like like key management for, you know, like a lot of the companies are spending a lot of money on these AI models.

对吧？所以能够实现监控、计费、密钥轮换等功能确实很棒。那么你现在是否看到了非代理性质的草稿？

Right? So it's great to be able to have like, you know, monitoring, billing, key rotations, all of that stuff. So are you actually seeing, like, a non agentic draft now?

是的。我觉得这个词有点被滥用了。有些AI公司像Poor只专注于构建模型等等。然后还有些包装公司试图通过AI解决人力资源问题之类的。

Yeah. I think I think so what I think the the word is a bit of stock. Okay. You have these AI companies like Poor that are just building models, yada yada. And then, then you have this wrapper company that are trying to solve, you know, HR issue, whatever through AI.

接着是那些提供对话API的大语言模型公司，比如Entropy，通过API获得大量收入。目前缺失的是让这些AI对话运行的基础设施。这正是我们现在通过AI网关等产品在做的事。但核心问题就像你提到的身份验证——就算实现了通用人工智能，代理在需要认证时仍会卡壳。而认证就需要获取API密钥。

Then you have these LLMs that have APIs to talks, companies like Entropy, lot of revenue through APIs, all that. What what is missing is the infrastructure to make those those AI talks and run. And that's what we do now with AI gateway and a lot of products. But at the core, the boarding problem, like you mentioned, authentication, authorization, you can be AGI as much as you want, but at the end, an agent gets stuck if he has to authenticate. And to authenticate, gotta get an API key.

目前还需要人工介入获取API密钥进行认证。但如果你能在一开始就提供基础设施，就能实现密钥配置、自动轮换和认证的自动化。这样你的大语言模型和代理就能自由运行，不必每次都被身份验证和授权卡住，因为所有密钥都已预先配置好了。我认为这正是我们客户需要的解决方案。

You need a human in the loop to get the API key authenticate. But if you can provide infrastructure at the beginning, you can provision and automate key provisioning, key rotation, authentication. So once you're there, you can unleash your LLM, your agents, or just roam free without getting stuck every time for authentication, authorization, because you're already provisioning all these keys. I think that's where we want to our customers.

那么，也许在我们讨论的这一部分，我们可以假设听众对AI API基础设施相当熟悉。对吧？你认为未来六个月、一年或两年内，API基础设施会因为AI而发生翻天覆地的变化吗？还是说我们现在所理解的基础设施会逐步演进，以相当明显的方式服务于智能体？你觉得这在基础设施层面是变革性的，还是更像一种过渡？

So so maybe just just assume for this this this part of our discussion that whoever's listening is is pretty familiar with AI API infrastructure. Right? Do you think that the way API infrastructure looks in six months or a year or two years is dramatically different because of AI? Or is it that the infrastructure that we have in place that we understand today evolves to also service agents in ways that are pretty obvious? Like, to what extent do you think this is like transformational on the infrastructure layer versus like more of a transition?

我认为如果不以API为先，就无法实现AI。这就像模型没有嘴巴和耳朵一样。AI就是通过API交流的。我们所了解的API基础设施会演进而非改变，但API流量和AI流量正在融合。传统的API调用确实如此。

I think you cannot do AI if you're not API first. It's just you don't have the mouth and the ears to a model. And so that's how AI talks. So the way we know about API infrastructure, it will evolve, it won't change, but API traffic and AI traffic, they're converging. So the classic API calls, yes.

但当你传输token时，这本身也是API调用。而智能——比如智能将通过token来售卖。每次财报都在讨论token。但每个token背后都是调用，有效载荷传输着token。所以它们正在趋于融合。

But also when you move tokens is an API call. And intelligence, let's say intelligence will be sold through tokens. Every earning release is about tokens. But behind on each tokens, are calls, the payloads move tokens. So they're kind of converging.

我们正在构建的是一个统一的API和AI连接平台，帮助你在管理传统API流量、智能体流量、MCP流量和LLM流量的过程中平稳过渡。我认为最终这些都将融合成一个统一的程序。这就是我认为智能将如何发展，基于未来两三年内的渐进式演变。但我们已经看到MCP流量每季度都在增长。

What we are calling or building is a unified API and AI connectivity platform that it helps you navigate this transition as you're managing classic API traffic, as you're managing agent traffic, as you manage MCP traffic, as you manage LLM traffic, I think it will all converge into a unified like program. And that's how I think intelligence will move and based on evolutionary steps in like two, three years. But we already see MCP traffic growing every quarter.

我是说，我甚至看到一些非常基础的东西也在快速演变。比如开发者工具，就拿Cursor来说——我以前是职业开发者。过去要了解公司有哪些API，需要阅读大量文档或进行繁琐搜索。

I mean, I see even, like, really basic things. Like, for developers, this is evolving very quickly. So for example, you know, for fun, I develop using Cursor. You know, I used to be a professional developer, you know, in a in previous life. Even knowing what APIs existed in a company required a ton of reading through docs or some weird search.

而现在，特别是如果你有像Kong这样的工具（它拥有所有元数据和目录），并能集成到Cursor里。不仅智能体本身知道这些API，你还能将其暴露给开发者。感觉整个开发范式都在转变，因为我们现在能把这些东西呈现给开发者了。

And it feels like now, especially like if you have something like Kong, which, you know, has all the metadata and has the catalog, and then you can integrate into something like Cursor. And so not only does the agent itself know about the APIs, but you can then expose it to you. I mean, it feels like a lot of like, even the development paradigm is shifting now because we can start to surface these these things to developers.

是的。我认为我们可以成为赋能者，帮助大小企业触达更多开发者或更广泛的群体。这也是我们将提供基础设施的方式。正如你从Yase Bourne看到的，我们有一个非常激动人心的路线图，正全力朝这个方向推进。这些就是全部要点。

Yeah. I think I think you can you can think you can we can be an enabler to help large companies, small companies to to get into the hands of more developers or more coarser. This is also how we're gonna provide you that infrastructures and and and all of that. As you see from Yase Bourne, we have a very exciting roadmap that is going in all in that direction. So those are all the name of them.

我认为关键在于微服务领域发生的变化。我在意大利长大，那里数学学得不多，主要学习历史。但从中我们明白，历史虽会重演，但每次都有新意。总存在某种类比。那么我们在微服务中看到了什么？发生了什么？

I think here's the big bet. The big bet is what happened in micros because I grew up in Italy and you don't really study a lot of math, you just study histories. But what you learn from it is that, even it does exactly repeat, but it reads. It's just there's always an analogy. So what we see with microservices, what happened?

我们最初在Python框架、JavaScript框架、TypeScript框架、Java框架甚至PHP等框架中反复构建了10次、100次、50次的速率限制和认证功能。到某个阶段，这变得毫无意义。于是我们将所有功能抽象为网关模式，把连接逻辑都移到网关，由其分发给正确的服务，无论使用什么语言。我认为LLM领域也会发生同样的情况。最初企业会使用一个大型LLM。

We first build rail limiting and authentication 10 times, a 100 times, 50 times into the Python framework, the JavaScript framework, the TypeScript framework, the Java framework, whatever PHP, blah, blah, At some point, didn't make any sense. Let's abstract everything to a gateway pattern. And we move all this connectivity logic to the gateway and that dispatch to the right service, no matter what language. The same thing I think is gonna happen in LLMs. At first you have one enterprise use one big LLMs.

现在他们会用五个、十个甚至上百个小型、中型或大型语言模型。一旦发展到这个阶段，你就不需要在每个LLM中通过框架或链式结构重复做令牌限流和认证。微软也经历过同样的事，你会再次将其抽象为网关合作伙伴。我认为AI网关将扮演类似角色，通过调度器编写连接逻辑来对接正确的LLM，而非在每个LLM中重复开发。这个重大趋势会像微服务革命一样，改变企业管理和运营数百个LLM的方式。

Now they're gonna use five, ten, hundreds, small LM, medium, large, whatever. Once you get to, you don't do tokens relimiting, token authentication in each LLM using the frame or the line chain, whatever. Eventually, same thing happened in Microsoft, but you will abstract that to a gateway partner once again. And that's where I think AI gateway will have the same analogy and dispatcher to write connectivity logic to the right LLM versus rewriting in each LLM. Think that's the big bet that we make that happens in microservices, happens in how enterprise will run and govern hundreds of LLMs.

太好了。在我们结束前，我想最好能给听众一些建议，比如如何了解更多关于AI如何改变API格局的信息，或者也许...

Great. Well, listen, as we wrap this up, I think it'd be great to kind of end with, like, any sort of recommendations for people that are listening to learn a bit more about how kind of AI shifting the API landscape or maybe, you know, some advice on

别开公司，帕特里克。

Don't start a company, Patrick.

或许可以给些建议，比如...你知道...该如何开始为这个趋势做准备？

Maybe advice on kind of like, you know, you know, how to start thinking about getting ready for that?

我之前说过，现在大量关注点都在模型预训练和调优上。但关键其实是连接层——LLM之间、智能体之间如何交互，如何运行。我认为很多流量会具有战略意义，而初创企业和大型企业都需要构建管理这种AI连接的能力，为下一代应用、内部工具和客户服务做好准备。

I think I said before, there is a lot of focus on model pre training, tuning, all of that. But a key thing is the connectivity layer, how LLMs, agents, whatever will talk to each other, how you run. I think a lot of the traffic will be very strategic and yet to build in the startups of the enterprise to manage that AI connectivity into your next apps, your next internal tools, your next customers.

对于那些正在聆听这段堪称硅谷最硬核创业挣扎故事的未来创始人们，你有什么建议给他们？

And for the, listen, the budding founders that are listening to, like, the most hardcore Silicon Valley struggle story ever, what advice do you give to them?

我想说的是，我从这十多年的人生中学到的是

I think what what I learn over a decade of life

你有没有想过...我必须要问，你是否曾觉得或许应该及时止损，然后重新开始？比如，你是否认为那本应是正确的

Do do you ever think I I just have to say, do you ever think that, like, maybe you should have cut your losses and tried to start it over again? Like, do you ever think that that would have been the right

选择？

thing?

从未。没有。

Never. No.

从来没有。原因是我无法想象自己回到意大利的机场，父亲来接我时问‘情况如何？’而我只能夹着尾巴垂头丧气地说‘失败了’。我宁可饿死在那里也绝不可能那么做。所以这个念头从未在我脑海中闪现过。

Never. Never. The the reason is I could never visualize myself going back to the airport in Italy and my dad picking him up and say, how's it going? And I was just like, would fail with my tails and my legs, like, could never I would have died here without food, like I cannot do that. And so that that never crossed through through the head.

听得人起鸡皮疙瘩。说得太棒了。

Shivers. That was so good.

但绝不是那样，每次我一开始有那个画面，就立刻告诉自己：我绝不会重蹈覆辙。关键在于，任何事总比我们预想的耗时更长。最好选择那些能持续十年、二十年的趋势，因为你有时间成长、试错、积累。你必须坚信这个趋势，而非追逐昙花一现的诱惑——无论是作为领导者、个人、市场、产品还是收入，过程总会比预期漫长。所以选定长期赛道，每天全力以赴，复利效应自会显现。

But never, like, that's that's the visual that every time I even start to I I had that visual and immediately say, I'm not going back like that. Oh. And so, and then, but the thing to advise is it always take longer than what we think. It is good to take a trend that last ten years, twenty years, because you have time to grow into and do a lot of mistakes and building, And just you have to generally believe it in this trend versus falling glitters, because it's gonna take longer than what you think as a leader, as a human, as a market, as a product, as a revenue is always gonna take longer. So take something that lasts and just put a 110% on it every day, and then it will compound.

早期保持低借贷利率。

Keep the borrowing rate low in the early days.

别倒下。

Don't die.

别倒下。我是说，别放弃。这些...这些就是我在实践中领悟的真谛。

Don't die. Like, don't quit. Those those are those those are, I think, the things I learned along the way.

Agi，过去几年与你共事是我的荣幸，非常感谢你参与本期播客。

Agi, it's been a privilege and an honor working with you these past few years, and thanks so much for coming on the podcast.

彼此彼此。谢谢你，Martin。

Likewise. Thank you, Martin.

感谢收听本期a16z播客。若喜欢本期内容，请点赞、评论、订阅、评分或留言，并分享给亲友。更多节目请访问YouTube、Apple Podcasts和Spotify。关注我们

Thanks for listening to this episode of the a 16 z podcast. If you like this episode, be sure to like, comment, subscribe, leave us a rating or a review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow

我们

us

在X平台（原Twitter）@a16z，并订阅我们的Substack专栏a16z.substack.com。再次感谢收听，我们下期节目再见。需要提醒的是，本节目内容仅作信息参考，不应视为法律、商业、税务或投资建议，也不用于评估任何投资或证券，且不针对任何a16z基金的现有或潜在投资者。请注意，a16z及其关联机构可能持有本期播客所提及公司的投资。更多详情（包括我们的投资清单链接），请访问a16z.com/disclosures。

on x at a sixteen z, and subscribe to our Substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode. As a reminder, the content here is for informational purposes only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any a sixteen z fund. Please note that a sixteen z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a 16z.com forward slash disclosures.