在人工智能时代，培养自主性比培养技能更重要 | 马克斯·朔宁（Notion 产品负责人）

我们以前会说，给我个能回应的东西，意思是，如果你要写产品需求文档，不如直接写一份用户会读的更新日志或者博客文章。

And then we would say, give me something to react to, which is okay, if you're gonna write a PRD, just write the change log or the blog post that a user would read.

现在要让人有东西可以反馈，容易多了。

Now it's much easier to give people something to react to.

比如说，好，这是产品的这个版本。

As in, yeah, here's the version of the product.

然后大家就会想，要是我们换种方式做呢？

And it's like, okay, what if we did it this other way?

哦，对，这个版本也在这儿。

Oh yeah, here's that version.

我觉得这简直太棒了。

And so I think that is just amazing.

它让迭代真正提前融入了产品开发过程，对吧？

It sort of builds in iteration into the product much earlier, right?

像瀑布式开发，那还搞个啥劲儿啊？

Like waterfall is sort of why bother?

你觉得接下来在产品开发方式上，最大的突破或转变会是什么？

What do you think is the next kind of leap or shift in how we build?

你看到什么新趋势正在兴起，可能会彻底改变我们的工作方式？

What are you seeing as like, okay this is now the new thing that's emerging that is going to change how we operate?

这个问题我其实挺纠结的，一方面我真觉得‘永远别小看纯文本’，这话以前有个论坛帖子说得特别好。

I'm very conflicted on this because on one hand I do wanna, like I believe the never bet against plain text, so a famous forum post at some point.

比如纯文本的 Markdown，还有代码，这些东西都特别持久。

Plain text markdown, like it's just such a durable thing, code is such a durable thing.

我觉得用代码来表达想法，可能真的是个很棒的做法。

I think that expressing your thoughts in code is probably a really good thing.

我们当然可以聊聊为什么，但另一方面我又在想，我们真会变成光靠聊天来协作吗？

We can talk about why, but at the same time, I'm like, are we really going to just be chatting back and forth?

所以像 Figma 这样的工具未来会怎样，我觉得特别有意思——一方面我确实看到有些 Notion 的设计师用得少了。

And so what is the future of Figma for example is like a really interesting example to me because on one hand I do see like sort of a drop in usage of Figma in some designers at Notion.

但另一边呢，又有人坚持说：不对，这些 AI 工具简直太棒了。

And then others are like, nope, these AI tools are wonderful.

我很难预测，直接操作会不会因为智能体替我们做了直接操作就消失了？

I, it's very hard for me to predict of like, is direct manipulation going away because the agent is doing the direct manipulation?

我还很好奇，自动化和增强之间这条分岔路该怎么走。

The other thing that I'm curious about is there is this automation versus augmentation fork.

我看那些超快的模型，比如 Spark，还有那个 Anthropic 的版本叫啥来着——嘿，Coo？

If I look at the really, really fast models like Spark and I forget what the Anthropic variant is where- Hi Coo?

不好意思，说岔了。

No, sorry.

但你还是会得到一个聪明的版本。

It's, you still get a smart one.

是 Opus，不过像是 Opus Fast 之类的版本，

It's Opus, but like Opus Fast or something where

你就能直接，嗯，搞定它。

you can just, yeah, like figure You

一天的费用轻松飙到三千美金。

very quickly run up a bill of like $3,000 a day.

但推理速度真的会改变一切。

But the speed of inference really changes things.

如果推理很慢，你得一堆任务排队，然后到处走走，想想别的事，再回来检查；但如果几乎是瞬间完成的，你还会这么做吗？

If the inference is slow, then you're queuing up a bunch of jobs and then you're walking around the building thinking about other things and then come back and review versus if it's nearly instant, are you still going to do this?

我们现在这种忙乱的多任务处理，真的能带来心流状态吗？

Is this sort of multitasking, the frenetic kind of thing that we currently have going on actually the thing that is sort of, you know, gives us flow state?

不，不会。

Well, no.

但如果推理变得瞬间完成，我们是不是又回到直接操作了？

But if the inference becomes instant, do we get back to direct manipulation, right?

你能立刻像捏陶土一样调整代码吗？

Do you instantly sort of like mold the clay that is the code, right?

我不确定。

I don't know.

我觉得这取决于模型的能力——人的智能会不会有饱和点？

I think it depends on model capabilities, which is do people, is there a saturation on intelligence or not?

我常打个比方，就像视网膜显示屏——你根本看不出来像素点了，完全看不出来。

The analogy I like to give is a retina display, which is after I can't see the pixels, I can't see the pixels.

你没必要再把像素做得更小了。

I don't need you to make them smaller.

很多认知任务不也是这样吗？到了某个智能水平，我就不再需要更强的算力了，我想要的是不同的交互方式和更快的响应。

Is it not the same for a lot of cognitive tasks, which is at some sort of level of intelligence, I don't need more and I instead I want a different modality and faster.

这些事儿吧，我其实特别期待。

So don't know, those things I'm excited about.

挺有意思的。

Interesting.

你最后说的这点是， smarter 的模型其实不会太大改变团队的工作方式，因为现在模型已经够好了，真正卡住我们的反而是 UX 这类问题。

So the last point you're making is it's like smarter models will not significantly impact how teams operate because they've gotten so good and it's other blockers now like, like UX essentially.

总的来说，我其实特别好奇。

I think in general I'm actually very curious.

我觉得那些实验室都默认，人永远会想要最聪明的模型。

The labs sort of operate, I feel like they operate under the assumption that people will always want the smartest model.

你就想要最前沿的模型。

Like you want the frontier model.

在某些领域，这可能确实没错。

And I think for certain domains that is probably true.

比如搞癌症研究之类的，要是砸几百万进去，那确实可能需要最牛的模型。

I think if we're gonna do cancer research and so on and if we're gonna spend millions of dollars on something that's likely true.

但公司运转又不是这么回事，对吧？

But that's not how we run companies either, right?

总不能每件事都非得找个博士来干吧。

Like we don't have a PhD for everything.

我觉得很多知识型工作，其实做到够用就行了。

And so I think for a lot of knowledge work tasks probably sometime we'll get to good enough.

一旦够用了，我们就可以优化别的方面——比如能本地跑、成本低、速度快。

And once you get to good enough, then you can optimize other things like they run locally, they're cheaper, they're faster.

我真的不太明白，为什么非得追求极致智能，这对我没什么吸引力。

And I don't know why the absolute intelligence thing doesn't interest me very much.

我觉得社会的瓶颈根本不在智商高低。

I think society is largely not capped by intelligence.

泰勒·科文好像也说过类似的话。

I think Tyler Cowen says something similar.

我可不想曲解他的意思。

I don't want to put words in his mouth.

所以我更关心的是外骨骼——而不是指望有个神一样的AI躺在某个数据中心里，咱们就干坐着等它帮忙。

And so I'm much more interested in the exoskeleton versus the, I have a god in a box in some data center somewhere and we're all sort of, you know, twiddling our thumbs.

我这边有几个相关的问题。

I have a bunch of questions along these lines.

这真有意思。

So interesting.

你刚才说下午一点是Notion里消耗token最多的时段？

You talked about how this 1 pm is the highest token spender.

就是整个Notion平台范围内。

This is across all of Notion.

我猜这应该不包括我们自动的安全漏洞扫描，还有像漏洞分类这种，得靠人工触发的任务。

I would assume this may not include our automatic security vulnerability scanning and like bug triaging is like when human kicks off jobs.

对，没错。

Yeah, yeah.

你们对token使用有什么政策吗？

What's your policy on token spend?

是想用多少就用多少，还是设个上限？你们会监控使用情况吗？

Is it spend as much you want, here's a limit everyone, do you keep track?

因为我也不清楚具体政策，我觉得应该是没限制的。

Given that I don't know what the policy is, I think it is unlimited.

我的意思是，理论上总得有个上限，但眼下这根本不是该优先考虑的事。

I mean, you can imagine at some point there would be, but right now I think it's the wrong thing to optimize for.

就像新东西刚出来的时候，不如先让大伙儿自由探索。

It's like when something new comes along, it's worth letting people explore.

我估计再过六到十二个月，很多公司就会开始问：这玩意儿的投资回报率到底在哪？

I do suspect in six to twelve months from now, a lot of companies are going to actually start asking questions around ROI.

我觉得这对很多人来说都会是个挺尴尬的对话。

And I think that will be an uncomfortable conversation for a lot of folks.

说到开销，像埃里克那边，或者总体来说，大概是个什么数字？

In terms of spend, what are the like the numbers for say Eric or broadly in terms

这事儿你问我可真问错人了。

of I am the wrong person to ask.

哦，就是特别多。

Okay, it's just a lot.

反正我也不清楚具体数字，但肯定比不上OpenAI和Anthropic那些人，毕竟人家干的活儿规模摆在那里。

That is just the, I don't, I would assume they pale in comparison to the folks at OpenAI and Anthropic just by the nature of the work they do and so on.

但对个人来说，肯定几千是有的，说不定上万，具体多少我也说不准，看情况吧。

But it is definitely for an individual in the, you know, I don't even want to put numbers in, but like thousands for sure, but like maybe tens of thousands, I don't know, it depends, yeah.

你作为产品负责人居然都不清楚，这本身就说明咱们现在根本不用操心这个，先放手干，六个月后再说是不是值得，就像你说的。

I think just the fact that you're as head of product or not on top of that means that it's just, let's not worry about this, let's just see what we can do and then we'll, you know, in six months, as you said, we'll figure out if this is ROI positive.

没错，我现在确实有资格不去管它。

Yes, I have the luxury to right now not care.

是的，我肯定有人在盯着，不会失控的。

Yeah, and I'm sure, you know, someone's looking at it, it's not gonna be out of control.

没错。

Correct.

我觉得有个挺关键的节点，就是什么时候token的花费会超过一个人的工资。

I think there's like this big, I don't know, milestone of when does token spend exceed someone's salary.

现在大家越来越常聊这个了：这笔开销该不该比工资还高？

That's something people talk about now more and more, just like, should that be higher than your salary?

还是该更低？

Should it be lower?

这中间到底该怎么权衡？

How does that all connect?

我觉得把token花费当成就来炫耀其实挺危险的，就跟有人炫耀自己一天写了多少行代码一个道理。

Yeah, I think there's a real danger in sort of making the token spend metric to like boast about, which is the same as when people boast about how many lines of code they've written in a day.

对。

Yeah.

我就想问，你怎么写了这么多代码？

And I'm like I, why do you have so many lines of code?

你这代码量，我都说不上来。

You have, I don't know.

世界上最大的软件项目，也没那么多百万行代码啊。

The largest software projects in the world have not that many millions of lines of code.

咱们干嘛要拿这个吹牛？

Like why are we bragging about that?

我根本不关心谁花了多少令牌。

I don't actually care about how many tokens someone spends.

对啊。

Yeah.

这根本不是个有用的指标。

It's not a metric that's useful.

没错，说得太对了。

Yeah, such a good point.

我知道Meta最近因为搞这么个排行榜，比谁用得最多，被不少人批评。

I know Meta got some flack for this recently where they're trying to create a leaderboard of who's doing the most.

说真的，我能理解公司为啥这么干——我真没想到，要让人搞清楚自己工作的核心流程，再找个代理工具，搭个类似‘工厂’那样的系统，居然这么费劲。

To be fair, I do understand why companies do that, which is I am surprised by how much work it takes to get people to identify the outer loop of their work and enlist an agent and build sort of the, I don't know the term right now is like factory, right?

就是那种用来处理他们日常工作的软件工厂。

Like the software factory for the work that they do.

让我惊讶的是，要让人跳出他们惯常的工作方式，得花多大劲儿去推他们。

It is surprising to me how much prodding you need to do to get people out of their, way they're used to working.

所以像Meta这种几万人的大公司，我多少能理解，干脆搞个排行榜，激励大家参与，也算个不错的起点。

And so if you're dealing with tens of thousands of people at the scale of meta, I have some sympathy for, okay, a good way to do this is just start a leaderboard and encourage people to do it.

等他们玩起来，自然会发现哪些东西真有用，对吧？

They will find good things and useful things to do with that as they learn, right?

所以这招，其实挺靠谱的。

So it's a, yeah.

这话说得太对了。

It's such a good point.

你得使劲推一把，才能改变人们那种随随便便的惯性做法。

Like you have to over index to change people's default easy behavior.

我就想照老样子写这些产品需求文档，没别的。

I'm just gonna do the I'm just gonna write these PRDs the way I've always done them.

会议我也还是按以前的方式开，没打算改。

I'm gonna run the meetings the same the same way I've done it.

我觉得这挺有道理的。

I think that makes a lot of sense.

你见过Notion里真能让人彻底改变工作方式的案例吗？

What have you seen actually work with Notion to get people to significantly change the way they work?

这得看是什么岗位。

Depends on the role.

那些跟工程离得远的岗位，其实根本不用多劝，他们自己就惊了：天啊，我突然有超能力了！

So roles that are perhaps further away from engineering, actually you don't have to convince them all that much because they're like, woah, I have superpowers now.

你看我刚搞出的这个东西，太牛了——以前和现在的能力差距太大了，简直让人上瘾。

Look at this amazing thing I've just built because the capability gap of what they were able to do before versus after is so huge that it's intoxicating.

但你得反过来想，没错，但你明白为什么我们不能合并这个PR吗？

And then you have to actually almost do the opposite, which is like, yes, but do you understand why we can't merge this PR?

从工程角度来说，西蒙经常提到，任何代码里的人工干预其实都不太好。

Think on the engineering side something that Simon last talks about a lot is sort of any manual intervention in code is kind of bad.

你很可能在验证流程或者软件生产线上哪里出错了。

You probably did something wrong in the verifiability loop and in sort of the software factory.

当然，代码审查是例外，这个不算。

This excludes obviously reviewing code, right?

我本人还是坚定支持代码审查的。

Like I am still very much in camp.

你真该多花点时间在代码审查上，比你现在做的还要多。

You should probably review more code than, put more effort into reviewing code than you do.

但在写代码这一块，每次有人工干预，都该让人感觉像出了个bug似的。

But at least on the writing side, every time there is an intervention, a human intervention, it should feel a little bit like a bug.

我觉得这其实是个很好的测试标准，能看出来你到底有多依赖人工操作。

I think that's a good litmus test for how, I don't know, agent filled you are.

我想再聊聊你们用的那些工具。

I wanna come back to the tools that you use.

你说Figma在设计团队里好像不太火了，这挺有意思的。

You mentioned Figma is kind of trending down within the design org, which is really interesting.

有什么工具最近反而越来越流行吗？

Is there anything that's trending up?

除了Figma，还有别的工具在你们的工具栈里用得越来越少吗？

Anything else that's trending down in terms of tools in the tool stack

你们团队里？

of your team?

其实我不确定Figma是不是真的在走下坡路。

So I'm actually not positive that Figma is trending down.

我觉得更可能是分成了两个阵营。

I think it's more that there is a, there's two camps.

我觉得Jevan悖论完全说得通——Figma可能其实越来越火了。

I could totally believe the Jevan's paradox, which is Figma is actually going up.

当然，Vibe Coding 这个趋势正在上升。

And then of course Vibe coding is going up.

说实话，我特别不喜欢硅谷那种非此即彼的对立叙事——好像Anthropic赢了，OpenAI就得输，反过来也一样，这种心态我真的挺反感的。

Like I don't want to create, in general I really, really dislike the rivalry discourse that exists in Silicon Valley which is for Anthropic to win, OpenAI needs to lose and vice versa and like that kind of thing.

所以我不希望把这种对立思维带到Figma和编程工具的讨论里。

So I don't want to perpetuate that with sort of the Figma versus coding.

其实终端工具的兴起挺让人意外的，一开始大家觉得这玩意儿吓人，但其实能干的事特别多；现在产品经理们一旦用上Cloud Code或者Codex，就发现根本没那么难。

I think the terminal is actually surprising, which is it's initially kind of scary for people and you could do so much, but now PMs are slowly the, once they're in Cloud Code or Codex, everything is fine.

我一般都会建议他们别总依赖图形界面。

And I generally encourage them to not use the GUIs.

我鼓励他们用Tweez，因为我很清楚，时间一长，他们肯定会好奇，忍不住去扒别的东西。

Encourage them to use the Tweez because I just know that over time they're going to be curious and like pull at other threads.

总有一天他们会突然醒悟：哦，原来我懂了点计算机底层是怎么回事了。

And one day they wake up and they're like, oh, I understand more of the substrate of what how how computers work.

这真的太有意思了。

That is so interesting.

所以设计师们都在用终端？

So the designers are using the terminal?

是的。

Yes.

没错。

Yeah.

还有啊，Conductor也是其中一个例子，我也不太确定。

That's so And then, I I don't know, Conductor is another one.

他们基本上就是主要用开发工具。

They're they're basically just mostly using developer tools.

跟开发者用的东西其实差不了多少。

It's not that different from what, developers, use.

我特别想跟你们聊聊本季的赞助商Vanta。

I am so excited to tell you about this season's supporting sponsor Vanta.

Vanta帮超过一万五千家公司，比如Cursor、Ramp、Duolingo、Snowflake和Atlassian，赢得并证明了客户的信任。

Vanta helps over 15,000 companies like Cursor, Ramp, DuoLingo, Snowflake, and Atlassian earn and prove trust with their customers.

现在团队借助AI，产品开发和上线的速度比以往快多了，但随之而来的风险也达到了前所未有的水平。

Teams are building and shipping products faster than ever thanks to AI, but as a result, the amount of risk being introduced into your product and your business is higher than it's ever been.

我接触的每一个安全负责人，都感到肩上的担子越来越重——既要保护公司、业务，更别提客户数据了。

Every security leader that I talk to is feeling the increasing weight of protecting their organization, their business, and not to mention their customer data.

因为节奏太快，他们只能不断救火，瞎猜优先级，还得凑合着用那些过时的工具。

Because things are moving so fast, they are constantly reacting, having to guess at priorities, and having to make do with outdated solutions.

Vanta能自动处理合规与风险管理，支持35多种安全与隐私框架，比如SOC 2、ISO 27001和HIPAA。

Vanta automates compliance and risk management with over 35 secondurity and privacy frameworks, including SOC two, ISO 27,001, and HIPAA.

这让企业能快速达标，还能持续合规。

This helps companies get compliant fast and stay compliant.

如今，信任比以往任何时候都更能决定你企业的生死。

More than ever before, trust has the power to make or break your business.

了解更多，请访问 vanta.com/lenny。

Learn more at vanta.com/lenny.

作为本节目的听众，你还能享受Vanta 1000美元的优惠。

And as a listener of this podcast, you get $1,000 off Vanta.

就是 vanta.com/lenny。

That's vanta.com/lenny.

AI 完全改变了软件工程师的工作方式。

AI has completely transformed the work of a software engineer.

两年前和现在比，简直是天差地别。

Like two years ago versus today, it's completely different.

你现在写的代码，几乎全是 AI 给的。

Like almost all your code is now AI.

我们一直在聊，什么时候全球有 50% 的工程师会写 100% 的 AI 代码？

We've been talking about like when will 50% of engineers in the world be writing 100% AI code?

估计再过一年就到了，这工作变化也太快了吧。

It's probably like in a year, which is insane how much that job has changed.

你觉得 AI 下一个会改变哪个岗位？

Which role do you think AI transforms next?

是市场？增长？销售？还是设计？

Is it marketing, is it growth, is it sales, is it design?

你觉得除了工程之外，还有哪个领域也开始发生明显变化了吗？

Do you have a sense of where things starting to really change other than engineering?

这可能有点争议，而且我其实还没想太清楚。

Okay, this is maybe a hot take and I actually don't have enough.

我觉得那些实验室的人可能在说：‘瞧瞧这人。’

It's very likely that the labs are like, look at this guy.

我的看法是，至少从实际数据来看，模型在编程能力上的提升速度明显是指数级的。

My take, it's very clear at least empirically that models are getting better at coding at some exponential rate.

我觉得这趋势不会变。

And I don't think that's changing.

但其他领域的发展，我真没觉得有多惊艳。

Now, I'm not that impressed with the progress in any other domain.

我觉得它们在写作方面并没有明显进步。

It tends to be like I don't think they've gotten significantly better at writing.

我到现在还是特别受不了那种AI写的垃圾文字。

I still very much hate reading sort of AI slop writing.

但关键是软件啊，安德里森，对吧？

But the thing is software, Andreessen, right?

软件正在吞噬世界。

Like software is eating the world.

如果软件开发、把商业流程写成代码的成本——我就说那种老派的软件1.0代码——正在急剧下降，

Well, if the cost of software and creating software and encoding business practices in code, and like I just literally mean the old like software one point zero kind of code.

那我们只会看到越来越多的软件。

Then if that cost is very much going to zero, we will just have a lot more of it.

所以我觉得，真正发生的是软件工程会渗透到各个领域，而不是说HR那边突然自己搞自动化了，因为现在他们不用再麻烦工程师去写代码了。

And so I think then in that case, it's more that software engineering will go into all the other domains, not necessarily that there is sort of some sort of, yeah, like, I don't know, our folks in HR are automating a lot of things because now they don't have to bug an engineering team to write that code.

我觉得这就是趋势。

And so I think that's how it's going.

当你看到那些模型公司说，我们在非编程领域取得了巨大进展，

And like if you look at when the model companies say, oh, we've made great progress in this other non coding domain.

我就会想，你们不过是把编程的思路用到了这个领域而已，这确实不错，但这也恰恰说明它们擅长的还是编程。

I was like, you just applied coding principles to this domain, which is wonderful, but that's what it's getting better at.

对吧？

Right?

所以我觉得，软件吞噬世界这个趋势只会越来越快。

And so I think, I just think software is eating the world is going to accelerate.

这个观点真有意思。

That is a really interesting take.

所以本质上是软件吞噬世界的进程在加速，而不是说它要去做完全不同的事。

So it's basically software, just the acceleration of software eating the world versus it's like, it's gonna now do a different kind of job.

这让我想起Codex的产品负责人也说过类似的话：每个能胜出的智能体，都会是个能自己写代码来实现需求的编码型智能体，而不是只靠预设的几种能力。这真是个绝佳的例子。

This makes me think about the head of product for Codex said the same thing that every agent, there's all these different kinds of agents and his take is every agent that will win is gonna be a coding agent that builds the thing it needs versus like it's come, it has like certain number of capabilities, open class, such a good example.

就是说，我会给自己开发一个新技能，现在我就会做这件事了。

It's just like, I will build a skill for myself and now I know how to do this thing.

没错。

Yes.

所有智能体其实都一样，你看那些开源的、大厂做的，包括我们自己的，现在全都越来越像编码型智能体了。

All agents are also like, if you look at all the harnesses, whether it's the open source ones or the ones from the model companies, ours as well, they all resemble a coding agent now.

我回头再聊一下投资回报这块。

I'm gonna come back to the ROI piece.

我觉得这真的挺有意思的。

I think this is really interesting.

就像你说的，咱们先砸钱、砸钱、再砸钱，看看会发生啥，边学边跑，全身心扑进去。

As you said, there's just like, okay, we're gonna spend, spend, spend, just see what happens, learn, accelerate, lean into all this stuff.

你说再过六个月左右，很多公司就会开始认真算这笔账了？

You're saying that in maybe six months, something like that, you think a lot of companies are gonna start really looking at the cost here.

你一般不爱预测，但你觉得接下来会咋样？

What do you, and you said you don't like to predict things, but what do you predict is gonna start happening?

我可能花在这上面的时间太多了，反正我压根左右不了这些事怎么发展。

I probably spend too much time than I should because I have literally zero impact on any of this as sort of how it plays out.

你可以想象这么一个世界：大厂实验室和开源模型之间的差距越拉越大。

But you can imagine a world where the labs, the delta between the labs and open weight models and so on widens.

那种世界我特别不喜欢，因为我最讨厌权力集中。

That is a world that I very much don't like because I hate centralization of power.

但在那种情况下，我觉得这些实验室 basically 就能决定世界长什么样了。

But in that world, I think the labs just kind of get to decide what the world looks like.

如果这个差距没拉大，那就会出现技术扩散，大家都会很自在地跑自己的模型、自己做强化学习，对吧？

I think if that gap doesn't widen, then you will just see a diffusion and people will get very comfortable running their own models, RL ing their own models, right?

你看 Cursor 就是这样，Intercom 也是，Notion 也在慢慢试水。

Like you see this with Cursor, you see this with Intercom, Notion is dabbling in it as well.

现在还只是试试水，但迟早我们会更认真地投入进去。

Use dabbling right now, but obviously at some point we might become more serious about it.

到时候，它可能不是最前沿的，但对很多任务来说，已经完全够用了。

And then you have like, it's not front, it may not be the frontier, but for a lot of tasks it'll be good enough.

所以我觉得，这种情况下，这纯粹就是个成本收益的计算，对吧？

And so I think in that case, that is just an ROI calculation, right?

对我来说，是不是用个更便宜、运行成本更低的小模型，把实验室那层利润省下来，更划算？

That is the, is it cheaper for me to send this task to a smaller model that is cheaper to run where I remove the lab sort of profit margin kind of thing.

我觉得这事有可能发生，但前提是，别突然冒出个情况——哦，差距已经大到没法追了。

I think that may happen, but it only happens if there isn't a fast sort of like, you know, oh yeah, the gap is now so big.

另一个有趣的是，现在我觉得我们正处在一个极其幸运的时间线上——至少在美国，有三家靠谱的实验室在激烈竞争，说不定Meta现在也加入了，那就是四家，未来说不定能有六家。

The other one is that's interesting is right now I think we're actually in one of the luckiest possible timelines, which is we have at least in The US three competent labs that are all sort of duking it out at the, and like who knows maybe meta now, so four, maybe we can make it six at some point.

我真希望美国能有十几家前沿模型团队，而不是总得依赖国外的机构来做这些事。

I think like I would love a world where we have like a dozen sort of frontier models in The US versus having to always rely on other places in the world to do this.

但说实话，这已经挺不错的了。

But like that's sort of pretty good.

要是这种竞争停了，我会有点担心。

If that choose stopped, I would be somewhat worried.

未来会怎样，真的很难预测，对吧？

And then it's hard to predict, right, like what would happen.

但如果这种竞争持续下去，情况可能会像当年的云服务大战——总有一天，底层技术会变成大宗商品，企业不会愿意被绑死在某一家供应商身上。

But if that doesn't, then I think it's gonna look similar to the cloud wars, which is at some point layers commoditize, businesses are not going to wanna lock in into one single provider.

我也不确定。

I don't know.

以前我在Heroku工作的时候，Kubernetes 的成功远超 Heroku。

In the past life I worked at Heroku and like Kubernetes was much more successful than Heroku.

虽然从用户体验上来说，Heroku其实差多了，但它的卖点是：我们能替你搞定运维团队。

Even though I think from a user experience perspective, it was much worse, but the Delta was Heroku was saying, hey, we're going to replace your ops team.

而Kubernetes说的是：我们让你的运维团队变成超级英雄。

And Kubernetes was, we're going to make your ops team superheroes.

而且我们还不把你绑在某个云平台上，你可以自由选择。

And also we're not gonna lock you into a cloud, you can choose.

这不就是企业真正想要的吗？

And obviously that's what businesses want, right?

企业嘛，当然想要选择权。

Like businesses want choice.

所以啊，真不好预测，这事儿太看情况了。

And so, I don't know, it's really hard to predict because it depends.

现在模型进展的差距也太大了。

So it's so asymmetric in terms of model progress.

你说那些赢的产品，往往都是让人觉得自己像超级英雄的，我一听就想到Kathy Sierra。

When you say the products that win often are the ones that make you feel like superheroes, I always think about Kathy Sierra.

你还记得这事吗？

Do you remember that at all as the thing?

有点印象。

It rings a bell.

哦，这样啊。

Okay.

现在回头看这些，简直暴露我年纪了。

It's just like, it's like from from the open days at this point shows how old I am.

她当时那套理念特别打动我，我觉得这影响了很多人对产品的看法——过去大家的思路是，别老吹自己产品多牛，要说‘我们能让你变成超级英雄’。

She was just something that really stuck with me and I think it's informed a lot of how people think about product, at least in the past is just her whole pitch was instead of making talking about your product and how amazing it is, it's about we will make you a superhero.

就像马里奥吃了小花，突然有了超能力，而不是说‘看我们这产品多厉害’。

Like it's like Mario getting the little flower and having superpowers now versus look at our incredible product.

我觉得这其实也是科技公司后来才悟到的：为什么代码审查工具、自动代码审查工具效果总不理想？

I think it's actually a thing that the coding companies had to learn when they tried to move to like, why do code review tools, automatic code review tools not work that well?

这里面有个微妙的点：你把代码推到公开场合——不管是公司内部还是团队里。

I think there's actually a subtle thing, which is you push your code publicly to, or publicly within your organization or your team.

然后有个工具专门挑你代码的刺，让你觉得自己是个垃圾程序员。

And then a thing roasts your code and tells you how terrible of a developer you are.

但你想想 Claude Code 和 Codex 是怎么做的——你写代码，然后发布的是你和 Claude 合作的成果，大家反而夸你是个牛逼开发者。

Versus if you think about what Claude Code and Codex does is you're coding and then you publish the work of you plus Claude and you get bragging rights of how good of a developer you are.

对吧？

Right?

所以我觉得这种‘超级英雄’的思路确实是对的。

And so I think the superhero stuff is definitely true.

说到超级英雄，我本来没打算聊这个，但最近老听人说你们推出的 Notion AI 助手特别火。

Speaking of superhero, I wasn't planning to talk about this, but I've been hearing a lot about how much people love your agent, the Notion AI agent that you all released.

到处都在说：‘这玩意儿也太实用了吧’，而且用的人特别多，各种背景的都有。

Just like, it's just coming up a lot of just like, wow, this is actually really useful and with like a lot of different people.

真想听听你觉得它为啥这么成功。

It'd be interesting to hear what you think made it so successful.

我知道你们这产品憋了挺久才上线的。

I know it was like a long time before you guys launched it.

你觉得是什么让它在实际使用中这么好用、这么成功呢？

Just like, what do you think is helping it be this useful and successful as a product out in the world?

我还希望它能再好一点。

I would like it to be even better.

我其实是自己第一个挑毛病的人，这么说吧。

So I'm like my sort of own first critic, I guess.

我一整天都在想它哪里还不够好，而不是它有多厉害。

I've spent most of my day thinking about where it falls short, not how great it is.

但我同意你的看法，说实话，我挺惊讶它居然这么好——这话听起来可能有点怪，但你懂我的意思吧？

But I agree with you that I'm actually surprised at how, this sounds so weird, I'm surprised how good it is, if that makes sense.

Notion 一直都在 AI 领域走在前面。

Notion has always been fairly at the forefront of AI.

我觉得 Notion 的第一个助手，其实比 ChatGPT 还早推出。

Like I think the first Notion assistant was actually launched before ChatGPT.

所以 Ivan 和 Simon 早就觉得，这玩意儿迟早要改变很多东西。

And so it's not that like, I think both Ivan and Simon had the intuition of hey this is going to change a lot of things.

所以这也是个非常重要的原因。

And so that's a huge sort of reason why.

但现在每个公司都想着要变成AI原生的，不管这到底意味着啥。

But there, every company wants to become AI native now, whatever that means.

就像当年说云原生那样。

It's like cloud native.

你都得特意说出来，那你还真算得上是吗？

I'm like if you have to say it, are you really?

你真有这个机会吗？

Do you have a chance?

但我没想到Notion这么快就做到了。

But I'm surprised how fast that happened for Notion.

这事儿我真没多少功劳。

And I'll take almost no credit in this.

我觉得关键是，智能体得有上下文才能干活。

I think what's good about it is agents need context to operate in.

代理程序可不喜欢那种一堆堆的数据墙，非得钻过一条细得要命的通道才能连上另一个数据源。

Agents don't really like walls of like, oh, I have to go through this narrow orifice to talk to this other data repository.

我觉得这次大家终于明白为什么一个打通的工作空间真的有价值了——因为它太好用了。

And I think for the first time it is kind of obvious to people why a connected workspace is actually valuable because it's great.

我可以派代理程序自由穿梭，自己去搞定这些事。

I can have agents roam around and do that.

这其实也牵扯到可塑性软件的概念。

And it touches on malleable software I think.

我更愿意把Notion看作一个操作系统，从这个角度看，它跟编程代理所处的Unix环境其实很像，可能比你想象的还要像。

I think of Notion as an operating system more so and then in that case it resembles the environment that coding agents are in with Unix much more than one might maybe intuitively think.

我觉得这些因素加在一起，有时候就是我们傻得敢去啃硬骨头——比如我们的企业搜索，做了大量自动权限处理之类的事，别人根本没做。

So, I think those all contribute and then sometimes it's just, we're just dumb enough to try hard things And so I think our enterprise search is sort of like this thing where we do a lot of automatic permission handling and so on that others don't.

说白了，你得真在乎才行。

I don't know, you have to care.

我得回头说说那句《圣经》里的话。

I'm going come back to my quote from the Bible.

我觉得这恰恰就是为现在这种时刻量身定做的答案。

I feel like that actually is an answer to this question that it was made for such a time as this.

Notion 把公司里所有东西都整合在一起，这简直是用 AI 辅助工作的完美上下文来源。

The fact that Notion basically has all the things about everything in your company is the perfect source of context for using AI and helping you work.

其实就是熬得久，慢慢就等到了这一天。

So it's just like, just being around long enough for a while.

对，我们本来就是该干这个的。

Okay, this is exactly what we've been meant to be.

干得漂亮。

Nice job.

干得漂亮，各位。

Nice job guys.

这不就是可塑性软件的另一个说法吗？

It's the same as malleable software, right?

现在终于有人开始意识到可塑性软件了，真好，其实它早就存在了。

Like I love that people are waking up to malleable software now, but it's been around for a long time.

以前总觉得这事儿有点难，而且老在想：我干嘛要干这个？

It was just always slightly too hard and slightly too like why would I do this?

所以我觉得，没错。

And so I think, yeah.

我想引用一句圣经里的话。

I'm going to use this quote from the Bible.

谢谢。

Thank you.

就在这儿了。

There it is.

好像更短一点。

It's like shorter.

原文是‘为这样的时机而生’，引申的意思就是：你注定要做这件事。

The original quote is for such a time as this and the interpretation is this you're like destined to do this thing.

每期都特别贴切。

It's very viable every episode.

天啊。

Oh, man.

咱们再聊聊你们团队的运作方式吧，因为我觉得现在很多人都在思考这个问题。

Going back to the way your team operates, because I think this is something that a lot of people are thinking about right now.

现在到处都在谈效率、节奏，像Anthropix这种公司每天都能推出一个大产品，简直了。

There's all this talk of productivity, pace, getting things out like Anthropix launching a massive product every day, basically.

你作为产品负责人，得帮大家持续、稳定、频繁地推出好东西。

Your job is head of product, help people ship consistently, regularly, often ship great stuff.

你们是怎么做到既加快交付速度，又能推出让人自豪、真正好用的产品的？

What has worked in allowing you guys to ship more quickly if you are and stuff that you're proud of, stuff that works?

我觉得这个答案特别依赖公司内部的文化，我职业生涯里就遇到过两次这种情况。

I think this answer is so specific to companies, like internal culture where if you, I've been in this situation sort of twice in my career.

第一次是我加入GitHub的时候，那会儿的产品和市场契合度，简直绝了。

One is when I joined GitHub which is obviously I think, I don't know, insane product market fit.

但我加入的时候，正好赶上他们有点迷茫，不知道下一步该干嘛。

It just so happened that at the time that I had joined there was a little bit of a, I don't know, identity crisis or like, oh, what's our next act?

我们该怎么做？

What do we do?

而且关于该推出什么，大家争论特别多，毕竟前作太惊艳了，后续真不好接。

And like lots of debates about what to ship because it's such a tough act to follow if your first act was just incredible, right?

我觉得Notion也属于这种情况。

And I don't know, I would put Notion in the same bucket.

所以这时候，就是要提醒大家：别想太多，动手做就是了。

And so in this case, it's just like reminding people that, hey, you can just do stuff.

咱们没必要那么小心翼翼。

We don't have to be that precious.

时间一长，人就会变得特别谨慎。

I think there's this preciousness that develops over time.

用户要是不高兴了怎么办？

It's like, what do we do when our users are gonna be upset?

其实，如果我们不持续创新，用户才会更不满；偶尔出点小问题，根本没那么严重。

Well, our users are gonna be upset if we don't innovate more so than if we accidentally break a thing.

这显然得把握个平衡。

So, it's obviously a balance.

但我觉得，关键是要提醒大家：能做出第一版的那群人，十有八九也能做出第二版，但你得敢试。

But I think just reminding people that the same group of people that was able to do the first act is very likely going to be able to do the second one, but you have to try.

我们内部老说‘多射门’，意思就是：好，怎么才能多射门？

Shots on goal is a thing that we say internally a lot, which is like great, how do you increase shots on goal?

回头一看，现在做实验更容易了，这不就是多射门嘛？

Which of course if we go back to, it's easier to experiment now, you're increasing the shots on goal, right?

我觉得这招真挺管用的。

So I think that has worked really well.

光堆功能没用，我们过去半年左右确实发了不少新功能，但说到底，功能数量跟代码行数、token消耗量一样，都是瞎扯的指标。

Just shipping feature after feature doesn't, we have been a little bit on a roll in terms of shipping new functionality maybe in the last, like, I don't know, six months or so, but at the end of the day, feature count is the same silly metric as lines of code or tokens consumed or whatever.

我宁可要少而精的功能，每个都做得特别好，还能组合起来干遍所有事儿。

I would rather have fewer features that are really, really good and where the combinatorics let you do everything.

我现在还在苦苦琢磨的一件事，就是软件质量。

And so I think something that I'm still very much struggling with is software quality.

我还要说一句，实验室那边也没能例外。

And I will also say, I don't think the labs are exempt from this.

他们的工具确实不错。

Like, love their tools.

挺棒的。

It's great.

我超爱用命令行，但每隔两周就冒出个回归问题，明明三周前刚修好的。

Like, I love, I live in the CLIs, but a regression like every two weeks of like a thing that was fixed like three weeks before.

结果连个双向渲染都跑不流畅，帧率根本没法看。

And they still can't render a two way at, I don't know, a frame rate that's reasonable.

所以我觉得，咱们缺的是那种苹果式精密一体铝制机身般的极致做工。

And so I think, yeah, quality is a thing that's missing like this Apple esque machined unibody aluminum kind of engineering.

真希望整个行业能重新找回这种水准。

I would like us to figure out how to get back to that as an industry.

你有没有做过什么来改善这个问题？

Is there something you've done to help improve that?

代码质量是一回事，软件本身的品质是另一回事。

So there's code quality and then there's actual software quality.

你要是不断发版、不断试错，总得在‘赶紧上线’和‘等它做到极致’之间找平衡。

If you're shipping, you know, shots on goal, there's always this balance of, okay, but wait for it to be awesome.

我知道这问题特别难回答，但你平时怎么跟团队说清楚：咱们到底怎么拿捏这个平衡点？

I know this is just like very hard question to answer, but just how do you, what's your kind of communication to the team of here's how we're gonna, here's where we're gonna find that balance.

这事儿真挺让人抓狂的，可惜我没法给你看——我现在用的是笔记本，但我们团队有张特别有名的贴纸，上面就一句话：只做明显优秀的东西。

And this is a very frustrating thing for people, but I actually, I can't show you because I'm using my laptop, but we have obviously good stickers, which is let's just only make obviously good stuff.

一开始大家都会问：等等，什么叫‘明显优秀’？

The origin, which is like, okay, wait, what does that mean?

但后来你就懂了，好东西，一眼就能认出来。

And then like, ah, you know it when you see it.

没人会否认第一代iPhone出来时，那玩意儿明显就是好到没话说。

Like I don't think anyone argued when they saw the first iPhone that it's obviously good.

Chat刚出来的时候，也没人会质疑它是不是真的出色。

I don't think anyone argued that when Chat first came out that it's obviously good.

所以我觉得这就是标准——就做明显优秀的东西。

And so I think that's the bar, like just make obviously good stuff.

我觉得很多公司犯的错恰恰是太追求完美了。

I think the mistake that maybe a lot of companies then make is great.

我们宁愿躲在一个角落里，非得憋出个绝对惊艳的成果才肯出来。

We're going to be in this cave in isolation until we have it sort of be obviously good.

我核心的理念之一是‘渐进式正确’——不断迭代，把迭代这件事做到极致。

One of my core values is incremental correctness, which is sort of iterate, get really, really good at iterating.

所以，我觉得这其实是两者的结合：多试、多出成果。

And so, I don't know, it's probably a union of, okay, increase shots on goal.

举个特别好的例子。

Like here's a great example.

我们天天被客户吐槽，但我特别喜欢这一点——Notion 里其实有六个自动化基础功能，你知道吧？

We get roasted from our customers all the time, which I love about, we have like six automation primitives inside of Notion, right?

要是算上所有智能代理之类的，就更多了。

Like if you include all the agents and so on.

我觉得吧，我们就让各种不同的想法自由发展。

I'm like, yep, we let like a bunch of sort of different ideas sort of grow.

我们看看它们怎么表现，但之后你必须下苦功，把它们浓缩回那个最纯粹、最核心的机器人骨架。

We look at how they work, But then you do have to do the hard work at consolidating it back into like the naked robotic core of that idea.

这事儿可不容易，对吧？

And that's hard, right?

因为你得接受，为了整合，下一个版本可能得稍微推迟一点发布。

Because you have to sort of be okay with perhaps then shipping the next thing slightly delayed as you reconcile.

我觉得Notion这儿还有很多工作要做，整个行业也一样。

I don't know, think we have work to do there at Notion, but as an industry too.

有人开玩笑说，为什么Claude的桌面版有三个标签页都是协同编辑代码，而聊天啥的却不知道在哪。

Like somebody was joking like why does Claude the desktop app have three tabs of co work code and I don't know what the for the chat or whatever.

我们为啥非得搞六个自动化基础功能？

Why do we have six automation primitives?

总得有人坐下来，把它们理清楚，找出那个真正核心、简单、能超越其他分支的原始想法。

Well, someone has to sit down and reconcile them and like figure out what's actually the core simple thing that should outlive the other sort of evolutionary branches of that same idea.

知道什么时候东西明显好，这其实靠的是品味。现在品味这个概念越来越重要了，因为AI负责把东西做出来，我们的任务就是挑好的。

This idea of knowing when things are obviously good, there's an element of having taste, and this is where taste that comes up a lot now, and this is like what we will need more and more because AI is building the thing now, our job is taste.

这东西很棒吗？

Is this great?

这东西够好吗？

Is this good?

我觉得你特别有品味。

I feel like you're someone that has a really great taste.

大家总问一个问题：怎么才能培养出品味？

A question people always ask, how do I build taste?

我有没有品味？

Do I have taste?

如果你也想提升自己的品味，有什么建议吗？

Do you have any advice for someone that's like, I want to develop my taste?

首先，说实话，我也不知道。

First of I don't know

如果我有很好的品味的话。

if I have great taste.

比如我会看别人是怎么展现他们的品味的。

Like I look at others and I look at how they exercise tastes.

我觉得最核心的一点就是不断迭代加反馈。

And I think that the common thing I think is iterations with feedback.

要在某个领域培养出好品味，真的得花很长时间。

So it takes a really long time to build up taste in a specific domain.

等你有了，就很容易把这种品味迁移到其他领域。

Then you maybe often can extrapolate into other domains with that taste.

但要我说品味到底是什么，这么说可能有点技术宅——就是你脑子里能跑个虚拟机，一有个点子，就能预判某个圈子里的人会不会喜欢。

But if I had to describe what taste actually means, it's you're able to run, this is such a nerdy way of describing it, you're able to run a virtual machine in your head where given an idea you can predict for a certain in group whether they're going to like it or not.

对吧？

Right?

极端情况是，全世界就你一个人觉得这东西好，那它算好吗？

The extremes are, is you are the only person on the planet that thinks something is good, is it good?

不是。

No.

但你也不一定非得为八十亿人做产品，我从来没做过消费类软件，估计也做不好。

But maybe you also don't need to build a product for 8,000,000,000 people, I've never built consumer software, would probably be terrible at it.

关键是你要先明确你的目标群体，然后看你能多准地模拟他们对东西的反应。

But you decide what your in group is and then how good do you get at emulating how they will react to it.

要做到这点，就得不断试错、反复练习。

And to do that you just have to do reps.

这简直就像训练模型，所以我也不是很认同那种说法——‘剩下的唯一东西就是品味了’。

It's almost like training a model, which is also why I'm not super, you know, the whole, oh, the one thing that we have left is taste.

我觉得这事儿没那么简单。

I'm not so sure.

你想想这个循环：输入一个想法，观察人们怎么反应。

Like if you think about the loop, it's input idea, how do people react?

这不就是反向传播吗？

That seems very back propagation.

我也不太确定，但感觉这跟我们训练模型的方式真的很像。

I don't know, like it seems very much how we train models too.

我特别喜欢的一点是，你就是靠不停地做、拿反馈、再迭代，慢慢培养出品味。

So what I love is basically you build taste by just doing the thing, getting feedback, iterating.

你看日本的手艺人，对吧？

Look at Japan, like Japanese crafts people, right?

他们就是一直、我不知道，反复画同一个碗，时间久了自然就成高手了。

They've just been, I don't know, painting the bowl for however long and it just takes a while.

所以我觉得，练得越多，练习的频率就越高。

And so I think the more reps, they increase the frequency of reps.

这就是我的看法。

That's what I would say.

有意思的是，这跟智能体学习成长的方式完全一样，就像你刚才说的，模型也是靠不断做、看这次做得好不好、对不对。

It's so funny that's exactly how, you know, agents learn and develop as you said, models learn just like doing the thing, seeing was this good, was this correct?

不对，行，那下次改。

No, okay, learn.

所以其实就是埋头做、不断学、拿反馈，这事儿真没法抄近道。

So it's just, yeah, it's just doing the thing, learning, getting feedback, and there's no way to speed run this.

这就是为什么最有品位的人，往往都是干这行干得最久的。

This is why often people with the best taste have been doing this for a long time.

我注意到一点，特别是对设计师来说，那些我觉得品位很高的——至少在软件设计领域——往往是那些既有个人项目、能从头到尾自己搞定的人。

The one thing I will say that I've noticed is specifically for designers, the designers that I think have, at least in software design, high taste are the ones that both have side projects that they build where they're responsible of the full thing in stand.

而且他们还总在捣鼓各种新应用。

And they're also always tinkering with some new app.

就是那种烦人的类型，老跑来问：‘咱们团队要不要试试这个？’

Like they're the annoying person that is like, hey, what if we tried this in our team?

我都想说：拜托，这都第49回了，你又推荐新工具？

I'm like, really, this is the forty ninth time that you suggested a new tool.

真有必要吗？

Do we really need this?

其实就是多接触别人的想法。

It's exposure to other people's ideas.

我觉得特别重要的是，要让自己身边都是有品位的东西。

I think that is the, it's also really important to surround yourself with tasteful things.

这样你才会觉得，自己做的东西还差得远。

So that you feel like the thing you're making is lacking.

比如我们在Notion，所有会议室都用著名物品命名，比如第一台打字机、麦金塔、保时捷911之类的。

Like one of the things we do at Notion is all of our conference rooms are named after famous objects like the first typewriter, the Macintosh, a Porsche nine eleven and so on.

所以每次我坐在某个会议室里，一抬头看到名字，就会觉得：我做的这点东西，根本没法跟这些比。

And so inevitably when I'm sitting in one of the rooms and I pay attention to the room, like nothing I'm doing amounts to this.

我得做得更好才行。

Like I gotta do better.

你已经做出了这么多成功、出色又受欢迎的产品。

You've built so many successful, great, loved products.

如果非让你总结一下，你觉得做一款成功产品，最关键的是什么？

What do you think matters in the end to building a successful product if you had to just kinda boil it down.

好吧，我这有个绝招，下周就开课卖。

Yes, here's the one trick that I'll sell a course next week.

对你来说，下周就开课了。

For you they'll say next week.

首先，我觉得我得说清楚，我只是参与了一些很棒的产品，但不是我亲手打造的。早年我根本不这么想，但随着干得越久，我越觉得，真正重要的是谁在做这个产品。

First of all, I think I would actually say that I have contributed to some really great products, not built them because I did not, I think I did not used to believe this early on in my career but like the longer I'm in this the more I care about what's the team that's building the thing.

以前我觉得这事儿没啥大不了的，现在我才明白，这根本就是唯一重要的事。

I used to think that was such like a, I don't know, not important thing and now I'm like oh it's the only thing.

我没法从我参与过的项目里找出一条清晰的主线来。

I don't think that there is a through line out of the things that I've contributed to where I can pinpoint it.

我觉得不能说设计最好的产品就一定能赢。

I think that you can't say that the best design always wins.

有很多产品，设计根本无关紧要。

I think there's many products where just design doesn't matter.

作为设计师，你可能会因此陷入身份危机：我到底在干嘛？

And like, I think then as a designer you can have this identity crisis of like, why am I doing this?

你甚至不能说，技术最牛的产品就一定能胜出。

I think you can't even say that the way it's built always like the best engineering always wins.

我觉得最大的坑之一，就是总想着：再加一个功能，产品就完美了。

I think one of the biggest pitfalls is if you get into the loop of, if I just add one more thing to the product, it'll be finally great.

你仔细看那些真正牛的产品，它们都有一个极小的核心，但做得特别极致。

Like if I really look at the truly great products, they all have one tiny core that is so exceptionally good.

这其实是运气撞上了，再加上市场也认了。

And that is both a combination of you stumbled upon it by luck and then the market agreed.

但关键问题是，那个核心到底是什么？

But I think it's the, what's the tiny core?

比如说，手机上的多点触控。

I don't know, multitouch on the phone.

GitHub大概就是那个‘拉取请求’吧？

GitHub is probably the poll request, right?

就是那种任何人都能给你提建议，你还能看到的机制。

Like this idea that anyone can suggest something to you and sort of you see it.

我觉得Notion的核心是‘模块’和‘斜杠命令’，而Figma呢，是实时协作和非实时体验之间那种无缝的融合。

I do think that at Notion it's the blocks and like the slash commands like Figma, it's sort of the seamless blend between real time collaboration and not.

所有厉害的产品都有一个能成为核心优势的小亮点，这和那种说「我们凑了一堆功能，再多加一个就能用了」的想法完全不一样。

Like all the great products have something tiny that is a superpower, like that's sort of like a versus, oh yeah, if we have this suite of things and like we add one more thing, it'll finally be useful.

那套永远行不通。

That never works.

说到GitHub，有意思的是它的核心就是PR（拉取请求）。

And for GitHub, interesting, it was the PR.

那你之前待过的其他公司里，有没有类似的例子？

Are there examples of that at other places you worked?

你说的这点太有意思了，就是那个能撑住所有其他功能的核心亮点到底是什么

Because this is really interesting, just like what's the tiny core that makes everything

才能让其他一切都运转起来？

else work?

Heroku肯定就是个典型例子，他们当年的核心就是一句git push heroku master就能部署应用，那时候部署应用本来麻烦得要死

At Heroku for sure, I think it was the Git push Heroku master of like at the time it was really hard to deploy apps.

对吧？

Right?

说实话，挺让人难过的，现在都没人记得Heroku了。

Like, this is like nobody it's sad because people don't remember Heroku.

但他们根本不知道，我得跟人解释，它才是真正的第一个PaaS平台。

They but it like, I have to explain it as it's the Versal it's the first Versal.

它后来被Salesforce收购了吗？

Did it get bought by Salesforce?

对，被收购了。

Yes.

是啊。

Yeah.

我们公司就是被Salesforce收购的。

We got bought by Salesforce.

明白了。

Okay.

没错。

Yeah.

Git push Workmaster 就是这么一句话，从我电脑上一敲，事儿就完了。

Git push Workmaster was just like this very simple one liner that went from the thing on my computer.

然后立马就有了个网址，这种感觉太上头了，后面所有东西都顺理成章了。

Now I have a URL and that's so intoxicating that everything else sort of flows from there.

Dropbox 就是个绝佳的例子，对吧？

Dropbox is a great one, right?

Dropbox 真的特别有意思，那个菜单栏的小图标同步做得太绝了，大家甚至拿它当判断有没有网的标志。

Like I think Dropbox is like such an interesting study where it was the little menu bar icon that was so good at syncing that you could even use it as a symbol for do I have internet or not?

它比你 Mac 本身还懂你有没有联网。

Because it was better at figuring out whether you had an internet connection than your Mac itself.

这事儿就该这么干。

And it was, it just, that's the job.

别瞎折腾，就让我文件一直都在那儿就行。

Get out of the way and just all my files are always there.

后来好多年，他们一直在想怎么扩大功能边界。

And then for years they tried to increase the surface area.

我一直在想，不，不，别再加了。

And I kept thinking, no, no, push it back.

我不需要更多了。

I don't want more.

我就希望你干好这一件事。

Like, this is the only job I want from you.

对吧？

Right?

我觉得，真正厉害的产品，核心就在这儿——越小越好。

And so I think the tiny core like is the thing that makes great products.

比如 Snapchat，那个照片自动消失的设计，真的很妙。

And Snapchat, obviously just like the disappearing photo concept.

这想法太有意思了。

That's so interesting.

我听说你也提到过，其实第一个做出来也没那么重要。

I hear, I heard you've also talked about just like being first doesn't matter that much either.

关键是你得

You have

选对方向，做对的事

to be right.

而不是做第一个吃螃蟹的人

Not first.

我是这么觉得的

I don't know.

要说的话，其实有些情况确实有先发优势——比如涉及网络效应，或者现在的模型训练领域，早点起步好像确实有点用，但我觉得先发优势被过分夸大了。

Like, I think, I mean, are probably, there are elements of like, if you're talking about network effects and like perhaps now with like training models, does make sense if you have sort of a head start, but I think it's overrated.

我举个最喜欢的例子吧，以前的蓝牙耳机都很难用，直到苹果出了AirPods，不管是配对还是连接体验都特别好。

I don't know, like my favorite example is like Bluetooth headphones were kind of crappy and then you have the AirPods and like, oh they connect and so on.

苹果从来都不是什么第一个吃螃蟹的人：MP3播放器不是苹果先做的，蓝牙耳机也不是，关键是要把产品做对、做好。

And they weren't the first, like I don't know, they weren't the first mp3 player, they weren't the first, like you just gotta do it right.

我觉得抢第一个入场根本没那么大用处。

I don't think being first is all that useful.

我觉得现在大家因为太难留住注意力了，总想着怎么才能火、怎么才能爆红。

I think we're currently, because it's so hard to keep people's attention, we try to say like, we're like, oh, how do I become, how do I go viral?

对吧？

Right?

怎么才能做出像Cluely那样的爆款？

How do I do the Cluely thing?

但我真觉得，关键是东西要耐用，对吧？

And I'm like, yep, I don't, durability matters, right?

你想啊，宜家是怎么做成百年企业的？它根本不在乎今天Twitter上啥话题火。

Like think of like, how would you build IKEA like a generational company that is not concerning itself with whatever is trending on Twitter today?

说到模型，有个好例子就是Anthropic，他们起步比OpenAI晚得多，融资也少，现在却直接杀出重围， dominate 了。

I think speaking of models, a good example is Anthropic, which was way behind, started after OpenAI, got less funding and now is just killing it and dominating.

我觉得最厉害的是，

The thing that

要说最让我佩服的，我不知道该归功于谁，但像迪特·拉姆斯这样的人，他可不是运气好了一次就完事了。

I find the most impressive about, I don't know who to give credit, but like obviously you give the CEO a bunch of credit, but like Dieter Rams is that he wasn't, oh, he wasn't just lucky once at OpenAI.

他两次都做了同样的事，而且两次都成功了。

He did the same thing twice and it was successful twice.

我觉得这其实特别厉害。

Like, I think that's like, that's actually really cool.

我知道你也信奉‘用户要完成的任务’这个产品思维框架，这话题在我们播客里吵了好久了，主要是因为Sriram一直反对这个观点。

I know you're also a believer in jobs-to-be-done as a way of thinking about product, which is kind of this, it's been a long time controversial topic on this podcast, mostly cause of Sriram who's there anti jobs- to be done.

你平时是怎么用这个框架来思考产品的？

What's your kind of framing of how you find this framework useful in thinking about product?

我猜如果我重新读一遍克莱顿·克里斯坦森的书，我也不会特别认同他的说法。

I bet that if I reread all of the Clayton Christensen stuff, I would also not identify super strongly with it.

我更多是用它来问：你有没有全面想过，用户到底想让你的产品帮他完成什么？

I use it mostly as have you thought holistically about what the user wants to hire your product for?

你有没有诚实地面对用户真正想要的，而不是你希望用户想要的？

And are you honest about what the user wants versus what you want the user to want?

另外，我在大公司里经常看到，人们在评估自己产品的时候，一上来就忘了把自己当用户，根本不去想：这体验真的好吗？

And then the other thing that I find happens very frequently in larger organizations is that people sort of turn off the brain when they're reviewing their own products from a I'm a user, is this a good experience?

他们更觉得自己是公司员工，做了个东西就完事了。

And they're more like I'm a employee of this company and I made a thing.

所以我觉得‘待办任务’这个框架能让人跳出来，别光顾着埋头做产品。

And so I think jobs to be done might encourage people to zoom out and sort of not get lost in the sauce of like making the thing.

这就是我喜欢这个框架的原因。

That's why I like the framework.

它能提醒你：别，别，别搞错了。

It's a good reminder of like, no, no, no.

用户是请你来解决某个问题的。

The user hires you for a thing.

你先把自己当个用户试试。

Be that user for a second.

你自己做的这玩意儿，你会买吗？

Would you even buy the thing that you just made?

答案往往是：哎，我还真没这么想过。

And the answer often is like, oh, I hadn't thought about that.

对吧？

Right?

所以这就是我用这个方法的方式。

Like, and so that's, that's how I use it.

除了那个奶昔的例子，你能举个你参与过的产品里的实际例子吗？让大伙儿更明白点。

Is there an example of this from some of the products you worked on just to make this real for people other than like milkshake example, obviously?

最近倒真有个例子，跟沟通有关。

There's a very recent one, which is more about communication.

我们马上要上线一个新功能，正在做这个功能的落地页来介绍它。

We're launching a new feature soon, and we're working on this landing page to describe the feature.

我发现，很多人做落地页时，第一件事就是文笔瞬间崩了——总想装高深，满嘴都是营销套路。

And I found that when people make landing pages, first of all, their writing skills just like deteriorate immediately because they wanna sound clever and like marketing speak comes out of their mouth.

我就想，等等，你平时跟朋友解释这事，会这么说话吗？

And I'm like, wait, that's not how you would explain it to a friend.

如果我要向你介绍这个产品，你就当自己正站在白板前。

And then if I'm communicating this product to you, just pretend you're standing in front of a whiteboard.

你画在白板上那些疯狂的图，跟刚才你设计的那个东西，真能是一回事吗？

What's the manic thing that you're drawing on the whiteboard to communicate this versus, okay, now go back to the thing you just designed, look at it.

你跟我说这两个是一回事？

Are you telling me that those are the same thing?

你真觉得你搞清楚了这东西是干嘛的，还能跳出来看？

Are you telling me that you understand what this thing does and like that zoom out?

所以我也说不准。

So I don't know.

对啊，我也说不准。

Yeah, I don't know.

不过我也不想专门挑最近几个例子来挑刺。

I don't wanna pick on individual recent things though.

好吧。

Okay.

在咱们结束之前，我想问问你，你对全民基本收入有个特别犀利的观点。

As we close out this conversation, there's something I wanna get your, You have this hot take on universal basic income.

这事儿来得真突然，但我觉得挺有意思的。

It's completely out of the blue, but I think it's interesting to hear.

现在有种说法是，随着AI兴起，我们可能根本不用工作了。

There's this idea that, you know, with AI emerging, we may not need to work.

大家直接领点基本收入，好好享受生活就行。

We'll all just get some UBI and enjoy our life.

而你却说，也许我们早就有了基本收入。

And you have this, I'll take that maybe we already have universal basic income.

这又是怎么回事？

What's going on?

是啊，你得给我点宽容啊，我这话一半是开玩笑，一半又真这么想——看你从哪个层面看人性了。

Yeah, so please extend me some grace here because I both mean it as a joke and maybe somewhat real, like just depends on which altitude of human nature you look at.

我的观点是，我们其实早就有了基本收入，只不过叫‘知识工作’。

My take is that we already have universal basic income, it's called knowledge work.

我当然也包括我自己在内，但你仔细想想，人真正活得舒服、有保障，需要的东西其实没那么多。

And I don't exclude my job from it, but if you really look at what do we actually need to live and like to be content, is a lot less.

我们搞出了一整套等级制度，还有这么多工作，好像每一样都必不可少。

And we've built this hierarchy and this sort of all these jobs and all these things that are absolutely necessary.

在我看来，其实UBI早就有了——那就是知识型工作。而且我们人类这么重要，是宇宙里最特别的物种，自然会想方设法把自己塞进智能体的对话里。

And to me it's like, we already have it at UBI and we'll come up with other ways in which we as humans, because we're the most important species in the universe, insert ourselves into the conversation around agents.

那样子会跟现在一样吗？

Will it look the same?

我不确定。

I don't know.

但说实话，我也说不准。

But yeah, I don't know.

我们人类太会折腾了，总能想出各种理由，非得把自己绑在那个循环里不可。

We are so inventive and we come up with new reasons of why we absolutely must be in that loop.

所以这就是我的硬核观点。

And so I think that's my my hard take.

人们老开玩笑说，我们每天就坐在电脑前，敲点字儿，居然还能拿这么多钱。

People have always joked, like, we get paid so much just to sit in front of a computer and put the right sorts of words and letters into the into this thing, and we get paid a lot of money to do it.