与Dax Raad一起构建OpenCode

你能说说吗

Can you tell

增长情况怎么样？

me about the the growth?

你们是2025年6月上线的，当时用户增长和团队规模都是怎么发展的？

Like, you launched in June 2025, how the growth was both for usage but but also for the team?

刚起步的时候，团队到底有多大？

Like, when you started, how how big was even the team?

是大部分来自Animalia Labs的原有成员在做这个项目，还是只有

Was was it most of the the existing sort of Animalia Labs working on this or just a

一小部分。

small Yeah.

不是。

No.

当时我们只有三个人。

So we were just three people at the time.

当时就只有三位创始人。

So it was just the three founders.

对。

Yeah.

后来我们雇了我一个朋友，他也对做这个感兴趣，帮我们搭起了最初的产品。

And then we hired one of my friends who was also interested in working as and helped us build, like, the initial thing.

他加入之后，团队就变成了四个人。

But he joined, so it was four.

接着我们说服了一位我们一直想合作的优秀设计师，也加入我们。

And then we convinced one of our, like, a really good designer that we would always wanna work with to join us as well.

那是上线之后的事了，大概在秋天。

And that was after we launched, that was more like in the fall.

不过呢。

But yeah.

我们上线后，增长马上就很不错。

So we we launched and the growth was really good, like, immediately.

这比我们之前做过的任何东西都要好。

It was better than anything that we'd ever done.

到了十二月，我们的月活跃用户达到了六十五万。

But by December, we hit, 650,000 monthly actives.

那时候我们真的惊呆了。

And at that point, we're like, wow.

这太棒了。

We're this is great.

去年秋天，我们跟人说的目标是明年年初突破一百万用户。

We back in the fall, we kinda were kinda telling people our goal was to hit 1,000,000 by next early next year.

大家都觉得我们疯了。

Everyone thought we were crazy.

当时大家就说，哦，是吗。

Like, that was like oh, okay.

好吧。

Sure.

到了一月，我们的月活跃用户达到了二百五十万。

Then in January, we did 2,500,000 monthly actives.

从六十五万一下子跳到了二百五十万。

So that went from six fifty to 2.5.

上个月，我们已经到了六百五十万。

Last month, we're at 6.5.

这个月才过了一半，我还不太确定具体数字。

I think this month, we're still halfway into the month, I'm not sure.

可能快到八百万了。

Maybe close to eight.

我们的下一个目标是突破一千万。

Our next goal milestone was 10.

当时出现了

There was

巨大的飞跃。

a massive jump.

所以我们十二月是六百五十万，到了一月就涨到了两千五百万。

So we went to six fifty in December and 2.5 in January.

那次暴涨是什么原因？

What was that jump?

是因为寒假期间，大家突然意识到这些模型真的特别棒，所以才爆发的吗？

That the the winter break jump where everyone started to realize that these models are really great?

对。

Yeah.

我们一直在做开发者工具，知道一月总会有个小高峰，因为很多人十二月放假时忙着学习新东西，假期一结束就迫不及待地尝试新工具。

So having been at DevTools for a while, we knew that in January, there's always a bump because like I think people take time to learn new things in December with a time off and they have time to try new stuff.

通常情况下，我们都会看到节日期间使用量下滑，一回来第一周就猛增。

So we what we typically see is we see a dip into the holidays and like a giant spike in the first week back.

但这次不一样，节日期间使用量还在持续上升，这在我们其他产品上从没出现过。

For this, we saw growth into the holidays which we've never seen with any other product before.

明明大家都放假了，使用量却比以往任何时候都高。

But, like, despite the fact that people were off, it was higher usage than ever.

到了一月，Anthropic 帮了我们大忙，他们居然想在 OpenCode 里封禁 Anthropic 的订阅。

And then January came around and Anthropic helped us helped us a lot by, you know, they they they, like, wanted to ban Anthropic subscriptions in OpenCode.

这事一下子闹得很大。

That blew up into, like, a huge thing.

我们几乎都没怎么回应，但用户群体简直炸了。

Like, we barely even commented on that, but, like, just the user base was so upset.

Anthropic 不小心把我们和他们放在一起说，这其实不公平，他们可是大得多、成功得多的公司。

Anthropic accidentally put us and them in, like, the same sentence, which we don't deserve because they're a much bigger, more successful company.

但就在那一周，情况突然变了，

But that week, that kind

这纯属意外，结果我们的数据直接飙升。

of happened by accident, and that, like, spiked our numbers like crazy.

所以现在回头看，事情是这样的：Anthropic 暗中禁止了在 OpenCode 里使用 Claude Code 订阅，而当时所有工具，包括 OpenCode，都这么干，因为这涉及一堆第三方，但他们偏偏不许 OpenCode 这么做。

So so I guess in in hindsight, because what happened is Anthropic silently banned being able to use Claude Code subscriptions inside of OpenCode, which which was every tool including OpenCode did that because it was a bunch of other third parties, but they didn't allow OpenCode to do this.

结果就引发了这场轩然大波。

And then this led to this, like, outrage.

他们的问题在于

The thing with them

似乎刚接触开发者群体不久。

is saying that they're kinda new to working with developers.

做些必要的事来保障业务可持续，这完全没问题。

It's okay to do things like like, of course, you need to do what you need to do to make your business sustainable is totally fine.

但偏偏在晚上九点突然封禁，这只会惹人反感。

But randomly dropping a block at at, like, 9PM at night, that just sets you up for, like, people to hate you.

如果能花一个月时间分阶段通知，提前打个招呼，情况会好很多。

Doing, a phased communicator rollout over a month, I think they would have Giving a heads up.

是的。

Yeah.

对。

Yeah.

没错。

Yeah.

我觉得会有人不高兴。

I think people would have someone upset.

当然，不管怎样大家都会不爽，但不会像现在这样，一下子所有人都被激怒了。

Of course, Everyone's gonna be upset no matter what, but it wouldn't have been this concentrated moment of like everyone being being really really annoyed.

这让我想起我们之前聊过的，AI让你能非常快地完成事情。

This reminds me of what we were talking about, how AI allows you to do things really quickly and fast.

你觉得在这种情况下，Anthropic是不是陷入了这种陷阱：他们能快速做事，就真的快速行动了？比如这次封禁，PR流程可能就花了几十秒或几分钟。

Do you think, in this case Anthropic might be stepping in this trap of they can do things really fast and they do things really fast but for example in this case they can just implement a block like this, know, like the PR probably took like what, like a few seconds or a few minutes.

做这件事的人可能根本没想清楚后果。

Whoever did that might have not thought through the implication.

我觉得这其实是任何快速增长的公司都会遇到的问题。

I think it's just a consequence of any kind of fast growing company.

你会忘记自己现在拥有的影响力有多大。

You forget the amount of leverage you now suddenly have.

你随便做一件小事，影响就会波及数百万人。

Like, you do one small action and it ripples through millions of people.

我们自己也在经历这种情况。

Like, we're dealing with this ourselves.

我们以前从没在这么大的规模下工作过。

Like, we've never worked at this scale before.

前几天我们发布了一个bug，几乎所有人都用的是深色主题，但一打开OpenCode，它却以浅色模式启动。

We put out a bug the other day where almost everyone has their terminal in dark mode, and we open OpenCode, it opened up in light mode.

结果一下子闪到好多人。

So we, like, flash banged a bunch of people.

以前这种事顶多影响一百人，但这次我一不小心，

And I'm like, in the past, that would have been, like, a 100 people we hit, but, like, I did this, like,

这周影响了一百万人。

a million people this week.

是啊。

So Yeah.

不过回头来看，这个改动的结果其实还不错。

But but in in inside this block, obviously, we know it worked out well in in hindsight.

但当时出事的时候，Opus 4.5 或 4.6 是市面上最强的编程模型，你看到 Anthropic 在媒体曝光前就直接停了 Claude Code 的订阅。

But when it happened, I mean, at the time, Opus 4.5 or 4.6 was the most powerful model out there best for coding, and you saw that Anthropic just blocked, you know, like, Claude Code subscriptions before you knew that this press would happen.

那圈里的反应怎么样？

Like, was the scene's reaction?

大家就是冷静点，继续干？

Was it just like, stay calm, keep going?

对。

Yeah.

奇怪的是，我们早知道这一天迟早会来，可真到了那天，大家反而特别兴奋，因为其实我们早就琢磨这事了，心里有数它迟早会发生。

What's weird is we knew this day would happen at some point, and for some reason when it happened, we all just felt excited because I think we were kinda, like, thinking about this for a while, and we knew this would happen.

我们也知道，咱们活在一个挺离谱的泡泡里，尤其是 Twitter 上，人人都在用 $200 的 Cloudmax 订阅。

We also knew that, like, we live in this crazy bubble, like, especially on Twitter where everyone has a a $200 Cloudmax subscription.

那时候我们月活跃用户是六十五万。

At that point, we were at 650,000 monthly actives.

怎么可能有六十五万人全都在用 Cloudmax 订阅？

There's no way 650,000 of them have Cloudmax subscriptions.

普通人每个月花两百美元买任何东西，都太疯狂了。

Like, it's crazy for the average person to spend $200 a month on anything.

所以我们知道，真正用的人只是其中一小部分。

So we knew it's it was like a smaller subset.

而且当时，我们一直在跟几乎所有其他公司谈合作，想让他们的订阅服务在OpenCode上得到官方支持。

And also at the time, we had been working on deals with pretty much every other company to officially support their their subscription in OpenCode.

就在那之前几周，我们刚搞定微软，让GitHub Copilot正式获得OpenCode的认可。

So the previous weeks leading up to that, we got Microsoft degree for, GitHub Copilot to be officially blessed by OpenCode.

我们还跟好几家其他公司都在推进合作。

We We got a bunch of different companies in the works.

不过这些都还没对外公布。

We haven't announced any of them yet.

唯一还没接触、甚至都没去谈过的，就是OpenAI。

And the big one we hadn't gotten or we haven't even approached was OpenAI.

所以那天晚上出事的时候，我忘了是星期四晚上。

So when this happened that night I forgot what it was a Thursday night.

我记不清具体是哪天了。

I don't remember exactly when.

我在X上收到了上百条标记，都说他们给封了。

I got, like, a 100 tags on on x being like, oh, they they banned it.

他们给封了。

They banned it.

我当时就想，行吧。

And I was like, alright.

该动手了。

It's go time.

于是我给OpenAI发了条消息，说：嘿。

So I messaged OpenAI being like, hey.

明天一早，所有人都会醒过来。

Tomorrow, everyone's gonna wake up.

所有人都会对Anthropic气得不行。

Everyone's gonna be really angry at Anthropic.

你们现在有机会通过持相反立场并正式支持OpenCode来赢得一次公关胜利。

You guys have a chance to score, like, a PR win by taking the opposite stance and officially supporting OpenCode.

第二天早上，他们确认了，没错。

The next morning, they confirmed, like, yeah.

就这么干吧。

Let's do it.

一大早，大家都说，Anthropic封了OpenCode。

So in the morning, everybody was like, oh, Anthropic blocked OpenCode.

他们完蛋了。

They're screwed.

我看到不少人说，他们现在一文不值了。

I saw a bunch of people being like, they're worth nothing now.

反正他们就要归零了，诸如此类的话。

Like, they're they're gonna go to zero, whatever.

我自个儿在那儿偷着乐。

And I was, like, laughing to myself.

我当时就想，行吧。

I'm like, okay.

等一天结束再说吧。

Just wait till the end of the day.

结果我们把集成这事儿给忘了。

And then we forgot the integration.

但我们确实已经实现了。

We implemented it.

当天我们就宣布了，OpenCode正式支持OpenAI。

And then in the day, we announced, like, OpenAI is officially supported in OpenCode.

我们早料到会这样，结果又绕回了OpenNext那套老套路。

We knew what was gonna happen and, like, it going all the back to OpenNext thing.

对吧？

Right?

我们最拿手的策略就是，先挑一个临时的反派，然后让所有竞争对手联合起来，一起对抗他。

The strategy that we know how to play really well is to pick one temporary bad guy and then galvanize all their competitors to push something forward against them.

所以Anthropic当时处境挺被动的，我们就顺势让整个行业都来支持OpenCode，支持在不同地方接入这些模型，因为他们都在和Anthropic竞争。

So Anthropic was unfortunate in that position, so we kinda got the rest of the industry to, you know, support OpenCode, support, like, access to these models in different places because they were all competing with with Anthropic.

这些其实都是很巧妙的策略，哪怕你是个小公司，也能在这样的局面里用上。

So these are, like, nice strategies that you can run-in these situations even as a small company.

我开始觉得，你过去这十年做开发者工具，尤其是最近五年做开源、理解这些生态动态，真的帮了你大忙。

I'm starting to get a sense that that past ten ish years of building dev tools or at least five of open source and understanding dynamics is really helping you.

就连你之前跟我讲怎么构思OpenCode的时候，都在谈策略——说这里有个缺口，一堆模型提供商在竞争，而随着时间推移，在很多生态里，开源的最终都会胜出，厂商们就像围绕Linux那样互相较劲。

Even when you told me about how you were thinking of OpenCode, were talking about strategy, about how there was this gap and there's all these competing model providers and with an open alternative over time in a lot of ecosystems the open one wins and the vendors compete for example with Linux.

Linux是开源的，但它的发行版都是商业公司做的，比如Red Hat、Ubuntu、Canonical这些。

Linux is open source but the distributions are for profit companies.

它们彼此竞争，反而把整个生态做得越来越好。

Red Hat, Ubuntu or Canonical and so on, and they all compete and and they make it better.

所以这次也一样，听起来你其实早就有了一套预判：只要厂商们按他们一贯的方式去打，事情自然就会朝着对的方向走。

So even in this case, it sounds like you, you know, like, it all goes back to you, you actually had a strategy that you expected that if the players play, the vendors play as they play, it will make sense.

对。

Yeah.

没错。

Exactly.

只要你定位准确，世界就会不断给你带来意想不到的胜利。

If you get your positioning right, the world just keeps handing you wins that you didn't even expect.

我们从来没料到会是这么个具体情况，但关于我们定位的核心判断是对的。

So, like, we never predicted this exact scenario, but, like, our fundamental understanding of the position was right.

只要有个中立的第三方，那些动辄几十亿美元的大公司都会倾向于借助它来推进自己的利益。

Like, if if there is a neutral party, all these companies with billions of dollars will kind of use a neutral party to advance their own, like, company's interest.

所以，做这个中间角色其实是有利的。

So it's beneficial to be that thing in the middle.

是的。

Yeah.

然后现在，CoCodex 就站出来了。

And then right now, co Codex came forward.

他们觉得这是提升品牌知名度和使用率的好机会。

They saw it as a way to increase brand awareness usage, etcetera.

现在他们正在分享增长数据，我敢说，从逻辑上讲，等他们真的占领市场、成为行业领头羊之后，也可能做出同样的选择——不再支持这个东西。但只要还有其他感兴趣的角色存在，这种情况就还有转机。

Right now, they're sharing, like, how how much is growing, and I'm sure, logically at some point, they might turn around at some point say, let's say they win the market or they become market leaders, they might do the same thing, okay, we no longer want to support this thing, at that point there might be other players as as long as there's multiple interested parties.

对。

Yeah.

没错。

Exactly.

所以，说不定哪天，OpenAI 就成了我们要联合起来对抗的反派。

So, like, at some point, maybe OpenAI becomes a bad guy that we have to, like, kind of galvanize everyone around.

这正是为什么竞争是好事。

So it's it's a I mean, it's why competition is good.

这简直就是竞争的典型剧本。

This is kinda exactly how competition plays out.

我觉得大家可能低估了一点：我们其实是个小公司。

I think the thing that people maybe underestimate is, we're a small company.

我们根本算不上什么大企业。

We're not, like, a huge company at all.

只要在对的地方施加压力，事情真的就能推动起来。

You can apply if you apply pressure in the right places, you can actually make things happen.

我们经常被人批评，甚至被骂，说你们对Anthropic太刻薄了。

We get a lot of criticism and and, like, anger from people being like, oh, you guys are being so mean to Anthropic.

他们挑我们做事的方式，可人家可是几十亿美元的大公司啊。

They criticize, like, the way we approach things, but these are billion dollar, you know, huge companies.

你知道的。

You know?

像我们这样的小公司，想施加一点影响都很难，现在看到的就是真实情况。

Like, us being able to apply any amount of pressure to get something to happen, it's very difficult, and this is what it looks like.

有人觉得这些大公司高高在上，不把人放在眼里，但其实让更多人能用上自己心仪的工具，这是好事。你喜欢云代码，觉得永远不换，这没问题。

People might think that it's, you know, they're above it or whatever, but, having more access to these things, having more people be able to use the tools they want is a good thing that you can love cloud code, be like I'm never switching out cloud code, that's great.

但你也该接受别人喜欢用别的工具这个事实。

You should probably still be into the fact that people can use other other tools that they like.

对吧？

Right?

再回过头来看看，为什么OpenCode能这么成功？结合我们对开发工具的整体了解，你觉得当初存在什么空白？

And going going back to why OpenCode is so successful and that there was a gap using what you've learned about DevTools in general, like, why was there a gap?

除了刚开始写出这个工具之外，你觉得你们做对了什么关键的不同之处？

What do you think was the difference that you did outside of just, you know, like writing the writing the tool to start with?

开发工具团队最大的优势在于，所有成员都是程序员，而程序员通常不擅长做面向消费者的產品。

The the biggest advantage to in DevTools is the fact that everyone working at DevTools is a programmer, and programmers are horrible at b to c products.

他们根本没意识到，开发工具其实也是面向消费者的產品。

They don't realize DevTools are b to c products.

这里的B2C，指的是企业对个人用户。

You basically B to c as in business, a consumer.

对，就是这样。

Yeah.

没错。

Exactly.

所以，你得用一种极端的思维去对待它——就像你在推出Instagram或者一款社交应用那样去打磨它。

So, like, you treat them like, the most extreme mindset to have is, like, you treat them as though you're launching Instagram or social media app, something like that.

是的

Yeah.

你可以采用自上而下的方式，确实有不少公司这么做，搞一套严格的大型企业流程，效果也不错。

You can do a top down process, and there's plenty of companies that do, like, an intense top down enterprise process, and that works and that's great.

但那些被广泛采用、甚至成为行业标准的开发者工具，全都是自下而上成长起来的。

But the DevTool products that are, like, massively adopted to the point where they're basically a standard, they're all bottom up.

通常是单个开发者先开始用，觉得好用，然后慢慢渗透进公司，一点一点扩散开来。

Like, they're individual developers start to use them, start to like them, and they kinda creep into companies and they they grow in that direction.

要实现自下而上，你就得像做消费类产品那样去思考。

And to do bottom up, you have to basically think like consumer company.

但程序员普遍根本不擅长这种思维方式。

And programmers generally are very, very bad at thinking this way.

所以我们特别注重，你一打开 OpenCode，它的体验就必须明显不同、而且比其他选项更好。

So we very much focused on the moment you open OpenCode, it should feel very different, and better than some of the other options.

为了做到这一点，我们得从零开始，构建一套全新的终端渲染框架。

And to do that, we had to build like a ground up, terminal rendering framework.

Claude Code 并不是这么做的。

That's not what Claude Code did.

其他那些编码终端代理也没这么干。

That's not what any of these other coding terminal coding agents did.

他们就是用了 Ink 之类的工具，反正能用就行，满足自己的需求就够了。

They just used ink or whatever and, you know, they made it work for what they need.

但我们从一开始就投入了大量精力，因为只要你一用，就能感觉到完全不一样。

But we invested on it upfront because from the moment you use it, it feels like something different.

可能这不适合你，也许你觉得信息太满、太刺激了，但你还是会觉得，做这东西的人肯定很专业。

It might not be for you, like, maybe you don't like the overwhelming experience, maybe it feels like too much, but you still walk away thinking the people that did this are are competent probably.

我们要做的，就是让人一用就立刻有好感觉，而且能马上开始提问，中间没有任何障碍。

So immediately getting people a good feeling and then immediately getting them prompting without any layers of friction.

所以我们一心一意地减少所有摩擦，比如考虑这种情况：我用的是公司锁死的笔记本电脑。

So we focus on just reducing that friction as much as possible and focusing on things like, I'm on a lockdown enterprise laptop.

我该怎么用 OpenCode？

How do I can I use OpenCode?

就是说，我们要把所有这些都优化好。

Like, optimizing for all of those things.

在OpenCode最初的五个月里，我们的框架其实并不怎么样，但勉强够用。

And our harness wasn't very good for, like, the first, like, five months of OpenCode, but it was good enough.

够用到大多数人根本察觉不出区别。

It was good enough that most people couldn't really tell the difference.

等我们占稳了市场份额，才回头去把框架真正做精、做智能、做优化。

And once we won enough share, then we went back and, like, tried to make our harness, like, good and smart and optimized and all those things.

这跟别人的做法完全相反。

But was inverted from what everybody else was doing.

别人都觉得，你得先做出最智能的框架，才能赢。

Everybody else was to being like, you have to build the smartest harness, and that's how you win.

我们呢，用的是个中等水平的框架，但却是用得最多的。

We did we have, like, a mid level harness, but we are the most used.

现在我们终于能慢慢升级，既有了最好的框架，也成了使用最广的。

So and now we're able to, like, you know, creep up and actually have the best harness and are also the most used.

所以我觉得我们采用的这种逆向策略挺不错的，但很多人

So I think it was a good inverted strategy that we did that a lot of

并没有完全理解这一点。

people didn't fully understand.

说到这个逆向策略，从技术层面来看，你们最初发布的是一个叫 Tauri 的框架。

And to get into the inverse strategy, just the technical level, like, you you launched with a framework called Tauri.

对吧？

Right?

哦，那是针对桌面应用的。

Oh, that was a for a desktop app.

所以关于桌面端，我觉得这基本上是我们这边的一个失误。

So the desktop, I would say, I wouldn't even say that was that was mostly a mistake on our part.

我特别喜欢终端操作。

So I love terminal stuff.

我所有的工作都是在终端里完成的。

I do all my work in the terminal.

所以我们的数据预计会达到八百万月活跃用户。

So our numbers are gonna so we're gonna hit, like, 8,000,000 monthly active users.

我觉得八百万人没必要都用终端。

I don't think 8,000,000 people should be using the terminal.

我觉得用户太多了，跟产品的定位不太匹配。

I think it's a little over, we have, like, too many users for what the product is.

很多人用图形界面体验会更好。

I think a lot of them would probably have a better experience on a on on a GUI.

我们很早就意识到了这一点，所以一直在尝试做网页版和桌面版。

And we understood that very early, and so we were kind of experimenting with, like, a web app with a desktop app.

我们觉得，未来大概率会朝这个方向发展。

And so we're like, this is probably gonna be the direction things go.

可惜的是，我们猜对了。

Unfortunately, we were right.

事实确实如此，但我们没把这个项目当回事，应该更认真地推进，早点上线，把技术细节想得更透彻一些。

Like, the the that is, like, where things are going, but we didn't, like, treat that project too like, we should treat it a lot more seriously and try to get it, like, out there way faster and thought a little bit harder on on the technical stuff.

所以我们当时就是随便挑了些东西，也没怎么认真考虑过。

So, yeah, we just kinda picked stuff and we didn't, like, think too hard about it and,

结果这确实是个错误，我们现在又回过头来用 Electron 了。

it ultimately was a mistake and we're moving back to Electron now.

不过 OpenCode 的增长速度真是惊人。

But OpenCode is is growing amazingly fast.

有意思的是，有些团队在 GitHub 的 PR 里会特意注明自己用了哪个工具——比如 Cloth Code、GitHub Copilot，有的没用，但在 OpenCode 里，你明明可以这么做，这几乎是免费宣传，却偏偏不提。

Interesting enough, there's this growth hack that some harnesses are doing, I would say growth hack where in the PR on GitHub, actually put which harness that Cloth Code is doing it, GitHub Copilot is doing it, some others are not doing it, and in OpenCode, you are not doing it even though this could be, like, almost like free marketing.

是的。

Yeah.

一开始我们其实是有的。

So initially, we had that.

因为最初我们基本上就是照着 CloudCode 做的。

Because initially, we're effectively a CloudCode clone.

只要 CloudCode 能做，我们就跟着做。

If CloudCode did it, we did it.

所以我们之前确实有这个功能。

So we had that.

每次提交时，日志里会写着‘通过OpenCode提交’，但其实没走GitHub。

Like, with the commit, it would say, know, committed with OpenCode without running GitHub.

后来有很多人来找我们，说：嘿。

And then we got a bunch of people being like, hey.

能不能关掉这个功能？能不能加个开关？

Can I disable can I have an option to disable this?

我仔细想了想，觉得这事儿来得太晚了。

And I thought about it, I was like, this is so late.

我自己都觉得这做法挺没劲的。

It just felt lame to me.

你完全可以把它做成像赌场一样。

I'm like, you can be like a casino.

对吧？

Right?

赌场就是用各种小花招，千方百计把你困在里面。

Like, casino just tries to trap you in there with every little trick trick to, like, get you to stay.

这已经是在做消费产品时走到极端了。

That's at the extreme of doing a consumer product.

我觉得我们没必要做到这种地步。

I felt like we didn't have to go to that extreme.

这么明显的增长套路，感觉有点low，太直白了。

It felt like a little bit lame to, like, try to it's like too obvious of a of a growth hack.

对吧？

Right?

谁都看得出来，而且都知道这玩意儿为啥这么设计。

Like, everyone sees it and everyone knows why that's there.

是啊。

Yeah.

所以我根本不喜欢这个设计，干脆就直接默认关掉了，也不提供开关选项。

So I I just wasn't a big fan and so instead of having the option, we just turned it off by by default.

你知道吧，归根结底你还是在运营一家营利性公司，你总得有资金吧。

You know, you're still running a a company that is a for profit company in the end or or you need some, you

得有本钱才行。

know, you need bankroll.

OpenCode 的商业模式是什么？

What is the business model behind OpenCode?

我们其实有两条业务线。

So, we have two lines of business basically.

当初刚推出 OpenCode 的时候，大家遇到的一个大问题——或者说我们自己也意识到——就是使用起来太麻烦，得先连账号。

So initially, when we first launched OpenCode, a big problem that people had was or that we had, again, thinking about reducing friction, they had to connect something.

得连上 Anthropic 的账号。

They had to connect their Anthropic account.

还得连上 OpenAI 的账号。

They had to connect their OpenAI account.

那时候你刚注册 Anthropic，根本拿不到足够的调用额度，根本没法用 OpenCode 这样的工具。

At that time, when you signed up for Anthropic, couldn't even get enough rate limits to even use something like OpenCode.

所以很多用户根本用不了。

So a lot of our users couldn't even use it.

我们就想，好吧，至少得做个推理服务，让用户注册后就能用上所有模型，还能获得所需的速率限制。

So we're like, okay, we have to at least build some kind of inference service that like you can sign up for and get access to all the models with the rate limits you need.

于是我们做了这个服务，叫OpenCode Xen，一开始只是想简化新用户的入门流程。

So we built that and we called it OpenCode Xen, and we initially just built it as like an onboarding thing to smooth out onboarding.

但后来这板块增长得特别快，再加上越来越多开源模型流行起来，我们发现正确部署这些开源模型其实特别麻烦。

But that grew like a ton, and with the amount of open source models that are becoming popular, we also found there's a ton of difficulty in hosting open source models correctly.

所以Xen现在成了汇聚最佳模型的平台。

So Zen has become a place to aggregate the best model.

我的意思是，前沿模型当然好办，但要把所有开源模型的推理做到最好，才是真正的挑战。

I mean, course, frontier models are easy, but then the best inference for all the open source models.

这些东西很快就火了。

All that became really popular.

这个业务现在增长得非常快。

That business is growing a ton.

几个月前，我们宣布这个服务的月使用量在五到六个月里就突破了五千万次。

I think a couple months ago, we announced, like, that hit 50,000,000, run rate, within, like, five or six months.

天哪。

Wow.

这个业务的利润率其实挺不错的，因为开源模型的部署成本相对低，利润空间大。

And the margins there can be pretty good because open source models, you can host that at a decent margin.

这部分的增长速度简直惊人。

That is growing like crazy.

我们当初根本没料到会这样，但它现在成了业务的重要组成部分。

We didn't really expect that, but that's that's a good big part of it.

另一块业务就特别枯燥了。

The other side of it is extremely boring.

如果你是一家有上千名工程师的公司，总不能让每个人都自己去下载OpenCode，再手动配置一个OpenAI的API密钥吧。

If you are a company that's using OpenCode and you have a thousand engineers, you can't just tell them all to go download OpenCode and, like, add an OpenAI API key.

你得有个统一的管控平台，来集中配置所有服务商、权限、预算和使用限额。

You need some kind of control plane to, like, set up all the providers, permissions, budget controls, rate limits.

所以我们为此开发了一个产品。

So we have a product there.

我们很快就会把它公开，但目前只在企业客户中部署使用。

We're gonna make that publicly available soon, but right now, it's just been, like, enterprise deployed.

如果你公司大规模使用 OpenCode，那就需要一套管理软件来支撑它。

So just if you're a company that's using OpenCode at scale, you need some administrative software to to run it.

如果没有这套工具，实际根本没法大规模使用 OpenCode。

You can't practically use OpenCode at scale something like that.

这个工具也是开源的，但大多数人还是选择购买我们的托管版本。

That's also open source, but, you know, most people just pay for our hosted version.

另外，我觉得现在终于到了大家开始认真关注大模型使用成本的时候了，都在问：我们到底在花这些钱做什么？

The other thing is I think, it's finally the time has finally come where people are looking at how much they're spending on on LM, and they're like, what are we doing?

我们真的因此提高了效率吗？

Are we actually getting anything any more done?

现在不少公司都在审视自己的开支，想办法稍微优化一下。

Like, we so, like, companies are now looking at their costs and trying to figure out how to optimize it a little little bit.

这时间点真是太好了，因为开源模型现在已经非常有竞争力了。

It's great timing because OpenSource models are now very competitive.

成本只有原来的十分之一。

They are 10 x cheaper.

把开源模型整合进来，提供优质的推理服务，也成了我们业务的一部分。

Blending that in and having good inference for open source models is becoming a part of our business as well.

这些大公司需要控制平台，而我们顺便给他们提供对其他模型的推理访问，他们也就自然而然地开始用了。

So these big companies, you know, they need the control plane, but then we kinda just give them inference access as well to the these other models, and they end up just kinda naturally starting to use it.

如果这部分成了我们业务的主力，我们可能会不再单独收费控制平台，只收推理服务的费用。

If that ends up being a main part of our business, we might stop charging for the control plane itself and just charge for the inference.

达克斯刚才提到的是那些枯燥但至关重要的工作，帮助企业顺利接入系统。

Dax was just talking about the boring but essential work to get enterprises onboarded.

单点登录、权限管理、控制平台——每一家认真做事的公司迟早都会需要这些。

SSO, permissions, control planes, all the stuff every serious company eventually needs.

这时候我得提一下我们经验丰富的赞助商 WorkOS。

Which is where I need to mention our seasoned sponsor WorkOS.

早晚会涉及到这些企业级功能的开发，不只是控制平面，还包括应用和智能体的认证系统。

Sooner rather than later you will need to get around building these enterprise features, not just control planes but things like auth for apps and agents.

WorkOS 能帮你搞定这些。

WorkOS handles it.

单点登录、SCIM，还有针对智能体实际运行方式设计的精细权限控制，全都内置了。

SSO, SCIM, fine grained authorization built for how agents actually operate.

增长最快的那些AI公司，比如Anthropic、OpenAI、Cursor、Perplexity，早就靠WorkOS来解决这些问题了。

The fastest growing AI companies, Anthropic OpenAI Cursor Perplexity, they already trust WorkOS to solve these problems.

详情请访问 workos.com。

Check it out at workos.com.

我还想聊一聊我们的冠名赞助商 Antithesis。

I'd also like to talk about our presenting sponsor, Antithesis.

我们刚才说了，慢下来反而能跑得更快，认真思考该做什么，别一头扎进错误的方向。

We just talked about slowing down to speed up, thinking hard about what to build so you are not sprinting to the wrong destination.

但质量本身，也是你要抵达的目的地之一。

But quality is part of the destination too.

你肯定不希望自己的产品像OpenCode那回一样，一下子炸出上百万用户。

You don't want your product to flash bang a million people like the team at OpenCode did that one time.

Antithesis是一个基于属性的测试平台，帮你定义系统应有的特性，然后验证这些特性在生产环境的混乱中是否依然成立。

Antithesis is a property based testing platform that enables you to express the properties your system should have, then verify that those properties will hold in the chaos of production.

有了Antithesis，规范制定和全面验证就变成了无缝流程，让你清晰把握质量，安心交付。

With Anthesis specification and thorough verification becomes a seamless workflow, giving you clarity so you can deliver quality.

想去了解更多？请访问antithesis.com/pragmatic。

Check out antithesis.com/pragmatic to learn more.

好了，我们接着聊Dax和OpenCode。

And with this, let's get back to Dax and OpenCode.

我们最近发了一条推文，讲推理服务其实非常赚钱。

We had a recent tweet about how inference is actually really really profitable.

你当时转引了一个人的说法，说那些AI模型提供商可能正面临财务压力。

You were quoting someone who was saying like oh, you know, these AI model providers might be having financial difficulties.

你能跟我们这些搞软件工程的人解释一下吗？我们平时只用模型，根本没碰过推理，总觉得这事儿理所当然是别人的生意。

Can you explain to those of us software engineers who, know, we don't we don't do inference, mean, we use these models and we we just assume that this this must be our our business.

这怎么还能赚钱呢？

How can it be profitable?

为什么能赚钱？

Why is it profitable?

你看到的是什么情况？

What are you seeing?

对。

Yeah.

我觉得这其实涉及业务的几个不同部分。

So I think this is it's kinda there are different parts of the business.

如果你只看纯粹的推理服务部分，想想成本底线在哪里，那最低就是电费。

So if you look at the pure inference part of a business, if you think about what's the floor on the cost, the floor is the cost of electricity.

买硬件确实有资本投入。

There's a capital cost to acquire the hardware.

但一旦设备到位，生成一个token的最低成本，其实就是让它运转所需的电费。

Once you have it to deliver a token, the cheapest it can get is the electricity to power it.

当然，还有其他基础设施方面的问题。

And obviously, there's, like, other infrastructure.

还有运营团队之类的开销。

There's, like, operation staff.

诸如此类的支出。

There's stuff like that.

但我们见过一些模型，看它的标价时，心里清楚——因为我们大规模租用GPU来运行这些模型。

But we have seen models that we look at the sticker price of it and we know like the, because because we rent GPUs at scale to run the models.

顺便说一句，我们还是得通过中间商。

And we still use middlemen by the way.

所以我们并没有把成本压到最低限度。

So we're not like going all the way down to the down to the floor.

就连我们自己，有些模型的标价和实际成本之间，利润率能高达80%。

Even for us, there are some models, the sticker price and the cost to us, there's like an 80% margin in there.

我觉得还有个大家没注意到的事，那就是价格其实涨了。

And I think the other thing that people don't notice is prices have gone up.

这让人困惑，因为看起来他们根本没涨价。

It's confusing because it seems like they haven't.

但实际上，从我们的角度看，价格已经涨了——以前我们默认用Sona，因为Opus太贵。

But they've gone up from the point of view as we used to all use SANA as our default because Opus was too expensive.

后来他们把Opus降价了，我们就开始默认用Opus，但还是比Sona贵得多。

Then they made Opus cheaper so that we started to use Opus as our default, but it was still way more expensive than Sona.

所以价格涨了，但Hostess模型的成本却没变。

So the prices have gone up and the cost of Hostess models haven't changed.

这也是为什么利润这么高的一个原因。

So that's one aspect of it being very profitable.

Anthropic当然有OpenAI那种超大规模的采购能力。

And Anthropic, of course, they have an OpenAI crazy scale.

他们签下的GPU采购合同，规模是最大的。

They have the biggest GPU deals that they've done.

所以我不奇怪他们现在的利润率可能高达九成。

So I wouldn't be surprised if they're looking at, like, 90% margin at current prices.

我觉得这种利润率长期来看根本撑不住。

I don't think that's, like, a defensible margin long term.

这么高的利润，投入的资源却没多少，简直不可思议。

That's crazy to be able to make that much with the amount of

而且流水还这么大。

money going through as well.

有意思的是，当初布莱恩·坎特雷尔上节目时，说过他以前在做一款想跟AWS竞争的云服务，那时候情况也是一样。

It's interesting because when Brian Cantrell was on the podcast, he used to work at building a cloud service that would compete with AWS, and he said that back then this was the same thing.

AWS那时候隐瞒财务数据，大家都以为云业务是赔钱的，血本无归。

AWS back then hid their financials, everyone thought, and they told everyone that cloud is a terrible business and it's like you're red blood everywhere.

但他亲自做了之后才发现，其实云业务

And then he started to do it and he said like, actually, it's

赚钱赚得飞快。

a really freaking profitable business to run

但这个秘密大家心照不宣，毕竟谁会去宣传自己在印钞票呢？

a cloud, but it's kind of a welcome secret because why would Amazon or any other provider advertise the business that it it's printing money?

任何被炒得火热的行业，总会伴随着负面情绪。

There's always negative sentiment that exists for any business that's getting hyped.

他们根本没有动力去纠正这些看法。

They have no incentive to correct it.

所以，这事儿挺复杂的，因为我清楚训练成本占了很大一部分。

So, again, it's complicated because I know that training costs are a big part of it.

研发部门的开销确实惊人，但从长远来看，推理服务作为一项业务是完全说得通的。

The r and d department is is hugely expensive, but long term inference makes sense as a business.

我觉得它永远都会如此。

I think it it always will.

是的。

Yeah.

话说回来，你之前在公开场合提到过GPU的事，我引述一下你的话：‘GPU的数量刚刚好够用。’

This being said, you also said something in in public about GPUs about you you I'm quoting you, there are just enough GPUs.

连我们这样的公司都被GPU卡住了，这简直太离谱了。

It's crazy that even a company our size is being bottlenecked by us.

是的。

Yeah.

那是什么意思？

What does that mean?

对。

Yeah.

现在整个GPU产业链，从生产到硬件支持再到人力，全都特别紧张。

So across the whole stack of GPUs, so everything from like producing the GPU to supporting hardware to like labor, everything everything is like super tight right now.

推理需求正在增长。

The demand for inference is growing.

我觉得这增长不是线性的。

So, like, I don't think it's linearly growing.

可能是呈指数级增长。

I think it might even be exponentially growing.

但我们GPU的产能还没跟上这种指数级的增长。

But we haven't made our production of GPUs grow exponentially.

这基本上是个线性过程。

That's, like, kind of a linear process.

所以当这两条线交汇时，供应就会越来越紧张。

So as those lines intersect, there's gonna be tightening.

对我们来说，得提前预留一些GPU。

So for us, like, there's we have GPUs that we need to reserve.

upfront就得付一大笔钱。

We have to pay a lot upfront.

现在大家都忙着囤货，因为大家都觉得这种紧张局面还会持续下去。

Everyone is now hoarding because everyone's kind of expecting this crunch to kinda continue.

想拿到推理算力太难了。

It's very hard to get capacity for inference.

还有另一件特别离谱的事，我记得我之前提过。

And, the other thing that's crazy, I think I posted about this.

你看，有些公司一融资就是二十亿美元，就为了做点AI相关的事。

You know, we see things like, oh, a company has raised $2,000,000,000 or whatever to do something in AI.

这数额听起来简直太离谱了。

And that feels like a crazy amount.

我的天啊。

Like, wow.

这真是个天文数字。

That's that's a huge amount.

你得是那种超级巨头才行，大科技公司每年花的钱可是几十亿级别。

Like, you have to be, like, a crazy star to do The big tech companies are spending, like, tens of billions in a year.

比如亚马逊、Meta，还有其他的。

Like, Amazon, Meta, whatever.

它们的投入完全碾压了初创公司的一切。

Like, they dwarf anything that's happening in the startup space.

它们把所有需求都吸走了。

So they are just vacuuming up, like, all the demand.

任何在供应链上的公司，谁还顾得上跟你谈？他们忙着跟亚马逊、微软或者谷歌打交道呢。

Any company that is in the supply chain, don't they wanna talk to you because they're busy trying to get something with Amazon or Microsoft or or Google.

所以，现在日子确实挺紧的。

So, yeah, it's very tight times.

我觉得最终会解决的。

I I think it'll get resolved.

在我整个职业生涯里，每次遇到短缺，都会经历一段紧张期，然后很快就迎来供过于求。

I think any in my whole career, anytime I've seen any kind of shortage, there was a tense period and it was met with like crazy oversupply.

这次可能不一样，但一般来说，事情都是这么发展的。

It might be different this time, but generally it's how how I see things go.

但眼下确实很紧张。

But right now it's tight.

我想聊聊AI工具，特别是AI代理，给工程团队带来的所谓生产力提升，以及现实情况。

I I want to talk about the the hype of what productivity gains these AI tools, specifically AI agents are giving to engineering teams and the reality.

你发过一条现在被广泛引用的推文，我摘几句说一下：大家都说自己的团队效率已经到顶了，瓶颈在于代码产出能力。

You wrote a now very heavily quoted tweet, which was I'll quote just some parts of it, everyone's talking about their teams like they were at peak of efficient and bottlenecked by the ability to produce code.

但实际情况是，就像你提到的，你们团队和整个组织其实不缺好点子，大家用AI并不是为了效率翻十倍，而是为了少花点力气把任务做完。

But the way things actually look like is, and you listed a few things like your team, your org greatly has good ideas, people are not using AI to be 10x more productive, they're using it to churn out their tasks with less energy to spend, and so on.

这事儿是几个月前的事了，大概两三个月前你发那条的时候。

So like, this was a few months ago, think, two or three months ago when you wrote it.

对。

Yeah.

我觉得这件事涉及的层面太多了。

I think that there's so many dimensions to this.

我那篇帖子讲的重点是，我们常常忘了软件工程行业到底有多大。

The thing I was talking about in that post is the majority again, we forget how big the software engineering industry is.

世界上几乎每家公司都在某种程度上雇佣着软件工程师。

Like, every company in the world employs software engineers to some degree.

对吧？

Right?

但大多数这类工作环境，并没有那么有激情、那么让人兴奋。

The majority of these environments aren't like the most motivating, exciting environments.

那里大多数人只是想把工作做完，早点回家陪孩子，过个安稳日子。

Most people there are trying to do their job, go home to their kids, like have a have a reasonable life.

你给他们一个能让他们更快完成工作的按钮。

You give them a button that lets them do their work faster.

他们最自然的反应就是拼命按这个按钮，干同样多的活，然后把省下来的时间拿去休息。

The natural place for them to do is hit that button as much as possible, do the same amount of work, and just cash in that extra time.

对吧？

Right?

这完全说得通。

Which makes total sense.

如果你没有动力去超额付出，那你是不会用这些节省下来的时间去推动公司更进一步的。

Like, you're not if you have no reason to be above and beyond motivated, you're not gonna really use that to, like, you know, push your organization harder.

是的。

Yeah.

这些工具或许能提升效率，但你得现实一点，想想你的员工。

These tools may make you more productive, but, like, be really realistic about your employees.

他们到底会把省下的时间用到哪儿去呢？

Like, where are they gonna, like, cash in cash in those gains?

当然，有些公司并不是这样。

Obviously, some companies are are not like that.

他们的员工本身就有动力。

Their the employees are motivated.

他们有充分的理由这么干。

They have good reason to be.

他们的报酬方式也合理。

They're compensated in a way that that makes sense.

但大多数公司都不是这样，问题在于，这类环境里往往只有少数人特别拼命，因为他们真的喜欢自己的工作之类的，尽管公司其他人没那么有干劲。

But most places aren't like that, and the problem with that is usually in those environments, it's like there will be a couple people that are irrationally motivated because, you know, they love the work they do, etcetera, even though the rest of the company isn't as motivated.

这些人通常是最在意质量、会推动大家多努力的那几个。

They're usually the ones that are trying to make sure everything is good quality, make kinda push everyone to try harder.

现在他们却被大量的代码审查压得喘不过气。

They're all now overwhelmed by, like, slot PRs.

我们团队里有几个人之前就待过这样的公司。

And we we've had a few people on our team that have joined, and their previous company was like this.

他们就是那个还在乎的人。

Like, they were the person that still cared.

公司里其他人都只是点个按钮把任务做完，而他们却被一堆烂事压得喘不过气。

The rest of the organization just, like, hits the button and gets their tasks done, and they're drowning in just garbage.

他们渐渐耗尽了，然后选择离开。

And they're getting burnt out, and they're leaving.

动力、团队、人、情感——这些因素在这整个事情里依然至关重要。

The motivation and team and people and humans and, like, emotions, all that stuff is still a big factor in all of this.

你觉得像中型公司或者小型公司，现在应该怎么重新思考激励、薪酬和工作方式呢？我跟史蒂夫·耶格聊过这个，他说一个特别有动力的工程师，只要投入时间和精力，产出可以多很多，而且如果方向对了，还能带来更好的业务成果。

How do you think companies, like let's say mid sized company or even a small company might have to rethink kind of motivation, compensation, thinking about work, Now that, you know, I was talking with Steve Yegge about this on how an engineer could, if they're super motivated and they put in the task and energy, they can produce a lot more output and if it's well directed that could be better business results.

但这背后要付出巨大的代价和精力。

But it does come at a huge cost and energy.

如果你拿的还是同样的薪水，那这么做就不那么划算了。

And again, if you're just being paid by the same paycheck, it doesn't necessarily make as much sense to do so.

所以我想听听，你到现在有什么初步的想法？

So I'm wondering like what kind of early thoughts you have so far?

当然，对于你们这些在初创公司的人来说，可能相对容易一些

And, of course, for for you guys in in a startup, I guess it might be a

确实容易一点。

bit easier.

我的意思是，没错，对我们来说确实更容易。

I mean, obviously, for us, you're right, it's easier for us.

我们所处的领域非常有前景。

We are in a very exciting space.

加入我们公司的人都很有竞争力。

The people that join our company are very competitive.

他们想冲进去，想赢，这本身就是很大的动力。

They wanna get in there and they wanna win, so that's like a big driver.

每个人都有股权，只要我们把事情做对，迟早会带来相当可观的回报。

Everyone has equity that if we do our jobs right, like, it'll be pretty meaningful for them, at some point.

甚至在AI兴起之前，我就觉得，我只能以初创公司为例，因为那才是我熟悉的领域。

I I even pre AI, I feel like, I get can only speak for startups as, like, that's where my experience is.

我总看到初创公司发招聘帖，说招两名工程师，还列了两个薪资范围。

I'd always see startups post jobs where they're like, we're hiring two engineers and they post two salaries.

这些薪资其实也不算低。

And these are like okay salaries.

我一直想，干嘛不只招一个人，把两份薪资合起来呢？

I always thought, like, why just hire one person and combine the salary?

这样你就能请到一个真正能改变你整个事业方向的人。

You'll get someone that'll, like, meaningfully change the direction of your whole life.

到了这个层次，优秀的人自然会找上门来。

Like, people will start showing up at that at that level.

所以我们一直愿意给很高的薪酬。

So we've always been willing to pay pay quite a lot.

我们不需要招一千个人。

We don't need to hire a thousand people.

我们只需要招二十个出色的人就够了。

We need to hire, like, 20 good people.

所以总的来说，我觉得高薪依然能带来很大影响。

So, generally, I think good salaries still still go a long way.

但话说回来，在规模更大的公司里，我觉得

But, yeah, at bigger scales, like, I

这确实很难。

think it's just hard.

当公司发展到一定规模后，员工根本没理由多做超出岗位职责以外的事。

Like, when when your company hits a certain size, like, there is just no good reason for someone to do much more than what their job strictly ask them ask them for.

所以我也没啥好办法。

So I don't have any good ideas here.

我一直在想，这会不会是因为很难把这种新技术和工作方式融入现有的组织结构？说到底，问题还是在于，像你这样的AI原生初创公司，或者现在新成立的这类公司，到底会有多大改变？他们的工作流程、任务内容，甚至岗位职责，都会怎么变？

And I I wonder if it's a a thing where it is hard to retrofit this technology and the way of working to existing structures which have been, and I guess the question goes back to like how much will like the AI native startup like yours or ones that are starting today change, like how will their workflows change, the type of work, even the roles, right?

比如软件工程师，当然他们会用AI写代码，但产品呢？

Like what does a software engineer even do in terms of, okay, of course they prompt AI to write code, but what about product?

其他那些方面又该怎么办？

What about all these other things?

是的。

Yeah.

对。

Yeah.

我的意思是，瓶颈确实还存在。

I mean, yeah, it's it's I think the bottlenecks are still there.

他们还在摸索该做什么。

They're still figuring out what you should be doing.

可能花上一年时间，就为了弄清楚该做什么。

Can spend a year just figuring out the right thing to do.

一旦想明白了，实际做起来可能很快，但话说回来。

And then once you figure it out, maybe it's fast to actually build it, but yeah.

以前还没用AI的时候，我95%的精力都在想该做什么，只有5%用来动手做。

Like, I I think the the joke I made was pre AI, I would spend 95% of my energy thinking about what to do and 5% of my energy doing it.

现在呢，我96%的时间还在想该做什么，只有4%真正用来动手。

Now I spend 96% of my time thinking about what to do and 4% of my time actually doing it.

所以，没错，效率提升了20%，但每天的感觉还是跟以前一样难。

So, yeah, it's like a 20% improvement, but day to day it feels as hard

一点没变。

as ever.

你还提到，很多公司——我想你是通过OpenCode在其他公司的使用情况间接看到的——现在CFO都在问：你说每个工程师每月多花2000美元是什么意思？

Another observation you made is at a lot of companies, and you will, I guess, see this second hand from from OpenCode being being used at companies, that the CFO is now going, what do you mean each engineer costs extra $2,000 per month?

这方面你听到过哪些说法？

What what are things you're hearing about in that in that area?

因为这事确实在发生。

Because this is happening.

是的。

Yeah.

我觉得每一种新技术都会经历一个炫耀阶段，现在有太多公司都想让人觉得他们走在前沿。

So the with I think with every technology, there's a phase of, like, flexing where there's so many companies now that want to be seen as, like, we're so future facing.

他们的流程比谁都先进。

We're, like, our process is ahead of everyone else's.

你根本没机会追得上。

Like, you do have no chance you're gonna catch up.

他们都到处炫耀自己花了多少钱。

And they're all bragging about how much they're spending.

现在大家流行说，每个工程师每月花上一万美元。

So there's like, right now, there's a push to be like, each engineer is spending like 10,000 a month.

这钱花得实在太值了。

And that's like, that's so worth it.

我们的业务这么成功、这么有用、这么高效，花多少钱都值得。

Like, our business is so successful and and so useful and so efficient that any amount of money is worth it.

这种说法现在特别多。

And and there's a lot of that narrative.

但其实这些都是假的。

Then there's like and then that that's fake.

迟早会消退的。

That'll go away.

但还有一种真实的情况是，这些公司确实雇了成千上万的工程师。

Then there's a real narrative of like, yeah, these companies hire thousands of engineers.

如果他们每个月多花一千美元，那整个预算就直接崩了。

If they literally cost a thousand dollars extra a month, that, like, breaks your whole budget.

这种事可不是能随便乱来的。

Like, that's not a a thing that you can kinda casually do.

我们现在正处于一个实验阶段，看看哪些可能性真的可行，大家都在摸索中。

We're in a temporary period of, like, we're experimenting to see what's possible and and and everyone's learning.

我觉得这种状态不可能长期持续下去。

I doubt this is gonna be a sustainable thing.

这些公司本来预算就紧，尤其是当他们没法明确指出具体解决了什么问题时。

Like, these companies were already pretty tight on what they could afford, especially if they can't directly point to clear results of here's the issue.

对吧？

Right?

我说的不是事实，但确实存在一种可能：这些AI编程工具最终带来的结果是，工作总量没变，但工程师们更开心了，因为工作变轻松了。

The net result of this I'm not saying this is the case, but there is a world where the net result of all these AI coding tools is the same amount of work gets done, but all the engineers are happier because their job is easier.

对很多公司来说，这还不够好。

That's not good enough for a lot of companies.

所以他们干脆就回去手动写代码了。

So they're just gonna say go back to typing out the code.

这里还有一个有意思的现象：当别人都在用的时候，哪怕我们只从‘员工更开心’这个角度来说，假设整体代码质量其实没变，

It's also an interesting dynamic where there's also the pressure of well, when everyone else is doing it, even if if we just stick with the the happier analogy, and again, let's let's assume that the the quality of the output overall wouldn't change.

如果别人都在用，而你却说‘算了，不用了’，你最好的工程师就会走人。

If everyone else is doing it and you're like, okay, it's not worth it, let's stop it, your best engineers will leave.

我听到一些CTO会说，我们提供更好的工具，是因为我们是顶尖公司，要是说‘你只能用GitHub Copilot了’，那多尴尬啊。

So I'm also hearing some CTO saying like, well, we're giving access to better tools because we're one of the best companies, like, it would look silly if we said like, oh, you can only use, I don't know, GitHub Copilot anymore.

我们得让员工用OpenCode、ClockCode或者其他任何新工具，不然最好的人才留不住。

Allowing you to get whatever OpenCode or ClockCode or whatever else that is because we like, our best people would leave.

是的。

Yeah.

不是这样的。

No.

这确实是现实存在的问题。

That that that is a real thing.

说白了，这得看情况——如果你有个能力超强、机会多到挑花眼的人，天天用Jira反而会让他烦，最后直接走人。

It's kinda like I mean, again, depends on if if you have someone that's really good that has infinite opportunities, using Jira is a risk because they're gonna be like, I hate using Jira every day, and they're gonna leave.

我认识不少人真就这么干了，但这种情况一直都在发生。

And I know people that have literally have done that, but that that's like always a always a dynamic.

所以这始终是个问题。

So But but that will

在人才市场上，这会是顶尖公司才敢这么做的。

be at the top of the market.

对吧？

Right?

是的。

Yeah.

没错。

Exactly.

是的

Yeah.

所以，以我们现在的规模来看，我们真的能感受到这个世界有多大。

So it's again, I think at the scale we're at, we're just so exposed to, like, how big the world truly is.

而我们平时所接触的这个世界，跟所有其他东西相比，简直微不足道。

And the world we typically play in, it's so small compared to everything else that that exists.

在大多数这些公司里，就是直接用 Copilot。

And at most of these companies, it is just use Copilot.

把 Copilot 接到 OpenCode 上，就这么简单。

Just plug Copilot into OpenCode and that's all you get.

你的限制就是你现有的这些工具，就这么回事。

And, like, your limit's your limit and and and that's what's there.

现在有人在鼓吹，说这些公司迟早会完蛋，因为跟不上时代之类的。

And there's some narrative being pushed that these companies are gonna die because they're gonna fall behind, etcetera.

但我也不确定。

But, I don't know.

我只是觉得

I just don't think

没那么简单。

it's that straightforward.

还有另一篇长帖，我特别留意了，就是你发给OpenCode团队的那篇，讲的是应该做三项调整。

Another longer, post that, that I I paid a lot of attention to is the one that you sent out to your team, to the OpenCode team, which was on on the topic of saying, alright, we should do three changes.

我来引用一下你的话。

I'll quote you.

大意是说，有以下几个挑战，虽然都不是新问题，但我觉得现在被大语言模型放大了。

Basically, have the following challenges, none of them are new but I think they're turbocharged by LMs.

第一，推出了一些根本不值得做的功能。

Number one is shipping features not worth shipping.

第二，迭代功能时，有时候最初的设计就有问题，导致不得不搞些临时方案，但大语言模型处理不了这种临时修补。

Number two, when iterating on features, sometimes the original design design is off and it forces you something hacky but the LM cannot deal with the hackiness.

第三，我们得花更多时间来清理代码。

And number three is we need to spend more time cleaning things up.

你是怎么得出这些观察结果的？情况有什么变化？团队又是怎么回应的？

How did you get to these observations and what's changed since you and how how did the team respond?

首先是第一点，也就是那些我们本不该上线的功能。

So the first thing, right, features that we shouldn't have shipped.

因为很容易就为了应对某个问题

So easy just to respond to

而匆忙上线一个功能。

a problem by shipping a feature.

所以我们要更加克制，尽量把这些点再梳理得更清楚一些。

So having more restraint and I so we we try to, like, flush that out a little bit more.

比如，这是我们要尽快上线的功能类型。

Like, here's the type of stuff that we wanna ship.

而这类功能，我们会先放一放，等思路更清晰了再动手。

Here's type of stuff that, like, we'll wait on until we have more clarity.

这样能让我们多几分克制。

There's probably more restraint there.

关于发布那些临时拼凑的功能，我认为这是最大的挑战，因为一直以来都存在这个问题：你有一个正在运行各种功能的系统。

The thing about shipping hacky stuff, that one is I think the probably the biggest challenge because forever has been a problem where you have a system that's doing a bunch of stuff.

你需要给它添加一个新功能。

You need to add a feature to it.

但这个系统并不完全支持该功能。

The system doesn't exactly support the feature.

所以你的选择有：从第一性原理出发重新思考系统，重新设计以支持该功能，或者暂时接受这个临时方案。

So your options are rethink the system from first principles, redesign it so it supports that feature, or just absorb the hack for temporarily.

你可以根据具体情况做决定，比如这个临时方案有多糟糕，或者把这个功能上线对公司有多重要。

And you can make a judgment call on which one you do depending on how bad the hack is, depending on how valuable to the company, it is to get this feature out.

由你来权衡判断。

You make the judgment call.

但现在这种判断力、这种做决定的能力已经严重扭曲了，因为智能体会直接帮你走捷径。

That judgment, that ability to have that judgment is so distorted right now because the agent will just do the hacky thing for you.

智能体会顺手处理那些后续出现的临时方案问题，让人很容易就想，哦，这只是个临时解决办法。

The agent will kind of deal with the hacky problems that come down the road and it's way easier to be like, oh yeah, it was it's a temporary fix.