人工智能的悖论：更多自动化，更多人力，更多工作 | 丹·希珀

我觉得 Cursor 的云平台实现比 OpenAI 或 Anthropic 的都要更优秀、更先进。

Like, I think that Cursor's cloud implementation is better than either it or OpenAI's or Anthropix and is more advanced.

而且到目前为止，Cursor 明显走了一条更清晰的路。

And I think that Cursor has, at least so far, more distinctly chosen a lane.

他们更明确地把自己定位为程序员的工具，这可能会限制他们的扩展空间。

Like, they're more distinctly choosing to be a four programmers, and that may limit how far they get in here.

我觉得‘程序员’这个定义正在扩大，他们的市场空间会很大，但我不确定他们会不会往其他方向拓展，比如——

Like, I think the definition programmer is expanding enough that they'll have a big market, but I don't know that they're gonna jump into, like, okay.

用它来做演示文稿之类的任务。

Use this to make a slide deck or whatever.

但很明显，所有大模型公司都在意识到，必须有个好用的框架，才能把模型的能力充分发挥出来。

But it is really clear that every model company is starting to realize how important it is to have a harness to get the most out of the the model.

所以，所有平台的发展方向，都是要超越单纯的‘输入提示、得到回复’这种模式。

And so where the where all the platforms are moving is to a world where you're not just doing prompt and response.

当你在 OpenAI 或 Anthropic 的平台上调用模型时，他们实际上是在自己的云端服务器上运行模型，再把结果返回给你。

When you call the model on the OpenAI platform or the Anthropic platform, they're literally running the model on a computer that is in the cloud that they run and then giving you the result out of it.

他们明白，要想让模型发挥最佳效果，就必须提供这样的工具。

And they know that they, in order to get the best results on the model, they need to offer that.

所以你看，Anthropic 已经推出了云端托管的智能代理。

And so you see, Anthropic's got cloud managed agents.

OpenAI 还没回应，但我猜这事迟早会发生。

OpenAI does not have a response yet, but I assume that that's going to happen.

现在，Cursor 实际上已经被 SpaceX 收购了。

And now Cursor was just essentially acquired by SpaceX.

不是完全意义上的收购，但已经非常接近了。

It's not like a full acquisition, but it's close.

所以我觉得，大家开始意识到，光靠模型本身是不够的。

So I think people are starting to realize, like, I can't just do the model part of it.

你必须在上面加一层配套的系统。

I have to have this harness above it.

而这个系统的终极形态，就是能处理任何类型的知识型工作。

And I think the ultimate form of that harness is, like, I can do any kind of knowledge work.

Cursor 本身让人觉得，他们得好好权衡一下，到底是只保留开发者用户，还是扩大范围。

Cursor itself is feels like one of the things that it's gonna be a hard decision for them whether to stay just for coders or not.

所以，如果你们不是 OpenAI 或 Anthropic 这样的公司，但这个趋势成立的话，预测就是，你们的产品迟早会被整合进这些智能代理里使用。

So if people building products that aren't OpenAI or Anthropic, if this proves to be true, the prediction here is they're gonna be using your product over time inside of one of these agents.

如果你们是这类公司，会做些什么来提前做好准备？

Is there something you would do if you're one of those companies to prepare for that future?

我就会直接为这个未来做准备。

I I would just prepare for that.

比如，像 Slack、Word 文档、PowerPoint 这类传统的生产力工具，本质上都是给人用的。

So, like, you know, for for example, every more more classic piece of productivity software, whether it's Slack or Word docs or PowerPoints or whatever, it's really mostly meant for a human to use.

但现在很多人开始做命令行接口了，这些工具是专门设计给智能代理独立使用的。

And now people are doing CLIs, so it's, like, meant for an agent to use independently of a human.

我觉得我们正进入一个新阶段：人和智能代理在同一项任务上协作，双方都在行动，所以我得清楚代理在做什么。

And we're moving into this new paradigm, I think, where the human and the agent are on the same piece of work together, and they're both doing things, and you need to have I need to have visibility into what the agent is doing.

代理也得清楚我在做什么。

The agent has to have visibility into what I'm doing.

我们得在这种无缝的方式来回配合。

We have to go back and forth in this sort of like seamless way.

为此类场景设计的软件，会跟以前完全不同。

And the kind of software that you make for that is gonna be very different.

比如，Proof 目前还有很多功能没做。

So for example, like there's a lot of stuff that Proof doesn't have.

我不用去管那些文档的格式、分页、表格之类的，因为代理自己就能搞定。

I don't have to have a lot of the like word document kind of like formatting or page breaks or like, you know, making tables or whatever because the agent just does it.

这些我完全不用操心。

I don't need to worry about that.

它能替我把所有格式都处理好。

It can do all the formatting for me.

所以产品可以做得更简单，开发起来也比传统软件快得多。

So you can make the products a lot simpler and faster to start than the legacy products are.

另外，还得加入一些新功能，因为代理和软件的交互方式跟人类完全不同。

And then there's all these other affordances that you need to start to have because the way agents interact with software is very different.

比如，智能代理可以同时处理很多任务。

So for example, agents can do a lot at once.

它们能一次性对你的文档、幻灯片、代码库之类的东西做上百万种改动。

They can just do like a billion different things to your document or your slide deck or your code base or whatever.

而如何向用户展示这些操作，跟展示一个人在文档里同时做事情的方式完全不同。

And how you display that to the user is gonna be very different than the way you might display a human being concurrent in your document and doing stuff.

你需要一个确认机制，需要一个类似收件箱的功能，汇总所有即将发生或已经发生的操作。

You need you need like approval, you need a sort of inbox that sort of summarizes, here's all the stuff that's going to happen or has happened.

你还需要操作日志，以及能快速回滚的功能。

You need you need logs and the ability to roll it back real quick.

这些因素都会改变产品的设计，连底层的用户体验和基础设施也不一样了，因为智能代理可能在三秒内发出上百万次请求。

So there's all those kinds of considerations that that change the actual product and then the underlying UX of it or the underlying infrastructure you need is different too because, you know, agents can make a billion requests in like three seconds.

那你要怎么应对这种情况呢？

So how are you gonna deal with that?

对吧？

Right?

这正是为什么 GitHub 现在遇到麻烦的原因——用 GitHub 的人数量正在呈指数级增长，而这些用户背后其实都是各种自动化代理在操作。

This is exactly why, you know, GitHub is having problems right now because because the the number of people using GitHub has is skyrocketing exponentially and it's really just people's agencies in GitHub.

我觉得这完全是一个崭新的世界，才刚刚露出一点苗头，但其中有很多令人兴奋的地方。

So I I think it's a this whole new world that is just starting you're just starting to see like a little peak of it, but there's so many cool things about it.

比如在 Proof，还有我们其他一些产品里，当用户遇到问题时，他们不会发邮件给客服。

So for example, in Proof, and some of our other products too, when someone has a problem, they don't email support.

他们的代理会自动提交一份错误报告。

Their agent sends a bug report.

而代理提交的错误报告，远比人工提交的要好得多。

And an agent bug report is way better than a human bug report.

它会清楚地告诉你：我具体做了什么操作。

It has, like, here's exactly what I did.

还有完整的复现步骤，一清二楚。

Here's the exact repro steps.

另外，Proof 是开源的。

Here's, like, proof is open source.

我觉得代码库里的问题其实是这样的。

So here's what I think is going on in the code base.

然后我们直接把它变成一个 GitHub 问题，再派个代理去解决就行。

And then we just get that, it becomes a GitHub issue, and then we can just like send off an agent to fix it.

不是所有情况都适用，但这种方式好太多了。

And you can't do that with everything, but it's so much better.

你能看到一种非常快速的闭环正在浮现：我遇到个问题，是个小瑕疵、一个小功能需求，或者个小 bug，我的代理就自动去找公司的代理，然后对方立马去修复。

And you can see the like, the glimmers of this this very fast, like, closed loop between I ran into something, a paper cut, a little feature I want, a little bug, and my agent just goes off and talks to the company agent, and then the company agent just goes and fixes it.

我觉得这简直太酷了。

That I think is incredibly cool.

现在很多人都在转向命令行，想在终端里干活，这也是趋势的一部分。

So as a part of this that you a lot of people are moving to CLI and trying to work from the terminal.

但你有没有预测，未来人们会逐渐离开这种方式，转而回归到有代理在旁协助的真正图形界面？

Is the part of this prediction that people shift away from that and back to actual UX with agents kind of running alongside them?

命令行已经过时了。

CLIs are over.

我们匆匆走过了命令行的时代。

We we speed ran the CLI era.

那段日子还不错，但现在很明显了。

It was nice while it lasted, but I think it's pretty it's pretty clear.

我不是说命令行不好。

It's not that CLI sorry.

我不是说命令行会彻底消失。

It's not that CLIs are going to completely go away.

毕竟，它们已经存在了三四十年，甚至更久。

Obviously, they've been around for the last, like, thirty years or forty years or fifty years or whatever.

它们还会继续存在。

They will continue to be around.

我觉得，当Claude Code刚火起来的时候，很多人以为成功的关键是命令行界面，但其实不是这样。

And I think there is this moment when Claude Code was, like, so popular or or when when Claude Code was really starting to gain in popularity that people were like, the the thing that's working is the fact that it's the CLI and I don't think that's what it is.

当你真正用上图形界面时，才会明白：我们当初开发图形界面是有原因的，用起来就是更舒服。

And when you move into an actual UI for this, you start to realize we we made GUIs for a reason And it's just nicer to be in a GUI.

在图形界面里，你同样能获得所有这些好处，尤其是对于非程序员的工作来说。

And you can get all the same benefits inside inside of GUI, especially for non programmer work.

我觉得Every公司里绝大多数技术人员现在早就不再把命令行当作主要工作界面了。

But I would I would estimate that definitely the majority of the technical people inside of Every are not using CLIs anymore as their main work surface.

很多程序员偶尔还是会切换回去用一下，但基本上他们现在用的都是Codex、Claude Code、Cursor这类工具。

I think a lot of programmers are still flipping into it every once in a while, but it's more or less they're using Codex, Claude Code, Cursor, that kind of thing.

太棒了。

Awesome.

明白了。

Okay.

我确实想把这一点说清楚。

I I I would I definitely wanted to make that part clear.

回到我们刚才谈的整体预测，你预计会存在两种主要的工作模式。

So coming back to kind of the the big picture of the prediction here, there's kind of these two modes of work that you're anticipating.

一种是公司内部的超级智能代理，你大概通过Slack和它聊天，它能自行去完成任务、回答问题。

One is this kind of super agent within a company that you chat with through Slack, most likely, that can go off and do work and answer questions.

然后在你的电脑上运行着 Codex 或 Cloud Code。

And then there's on your computer running Codex or Cloud Code.

在这些工具里，你平时在电脑上做的所有工作，现在都会转移到 Codex 或 Cloud Code 里，也可能未来会出现其他我们还没想到的第三方工具。

And within that, all the work that you'd normally do kind of on your computer is now gonna be living within Codex or Cloud Code, or maybe some third party that emerges that we're not even aware of yet.

对。

Yes.

你还会在这些工具内置的浏览器里使用各种应用。

And you're going to use apps inside of the internal browser of those of those tools.

真厉害。

Wow.

明白了。

Okay.

听你这么一说，感觉好像没那么震撼，但实际上这会彻底改变我们的工作方式。

Like, listening to you talk about it, it's like, it may not feel as profound as it is, Cause this is a big change to how we work.

我们现在还没有一个能经常在 Slack 里聊天的 AI。

We don't currently have an AI that we talk to regularly in Slack.

而且我们现在也主要不是在用 Codex 或 Clarke Code 工作。

And we also don't work currently mostly in Codex or Clarke Code.

所以这实际上是一个相当大的转变。

So this is actually a pretty massive shift.

我也这么觉得。

I think so.

在进入下一个预测之前，还有没有其他类似的事情要聊？

Is there anything else along these lines before we get into our next prediction?

嗯，还有几件事。

Well, a few things.

我绝对不是那种认为所有事情都该由智能体来做的极端支持者。

I am definitely not an agent maximalist.

我觉得我们将来会用到各种各样的智能体。

Like, I really think we're gonna have a lot of different agents that we use.

这一点对我来说很清楚。

Seems pretty clear to me.

我真心觉得，两个智能体比一个好。

And I really do think that two agents are better than one.

那有什么好例子呢？

So what's a good example?

当我让 Codex 和其他智能体互动时，它能提供比我手动输入多得多的关于我以及我需求的背景信息。

When I have Codex interact with another agent, it can give so much more context about me and what I want than I would be able to type.

而且它们之间来回沟通，能节省大量时间。如果假设用户是通过 Codex、Claude Code 或 Cowork 来访问你的应用，那么这种效率提升的效果会非常明显。

And it can go back and forth talking about things that would take a long time for me to express directly to an agent that you get this like speed up effect when you assume that your users are are using Codex or Claude Code or Cowork as their as their basic way they access your app.

举个特别简单的例子，

And a a really simple example,

我们有

we have

一个托管版的 Open Claw 产品，之前是放在候补名单上的。

this hosted Open Claw product, which we we had it we had on waitlist.

我们后来不得不暂停了，因为开始对接候补名单里的人，而 Open Claw 这个智能体框架真的很难让它跑起来。

We actually had to pause it because we started taking to all the waitlist, and Open Claw is just a very hard agent harness to to make work.

简直快得不可思议。

It's like, it's moving so incredibly fast.

如果你是这个平台，一旦出问题，根本来不及修。

And if you're like a platform for it, it just it's like, when things break, you can't fix it.

这太难了。

It's very hard.

但在这一过程中，我们学到一点：如果你在开发一个智能代理产品，或者任何全新的软件体验。

But one of the things that we learned in that process is if you're let's say you're building an agent product or or a new any new software experience.

你可能会觉得，要设置一个代理，就得做个网页界面，或者搞个Slack流程，问用户：你是谁？打算用它来做什么？

What you would assume, let's say to set up an agent is you need to build like a little like web interface or a little Slack workflow that asks people about, okay, like, who are you and what are you gonna use this for?

还有，你理想中的最终效果是什么？那些通常会写在用户引导清单里的内容。

And, like, what's your what's your ideal, you know, dream outcome or what whatever the things you are that you would put on an onboarding checklist.

但如果你干脆定下一条规矩：我们只服务使用Codex或Cowork的用户。

If instead, you you just you just make a hard line of, we are only going to service users who use Codex or Cowork.

那会发生什么？你只需要把一段提示词直接粘贴到Codex或Cowork里就行。

What happens is you just paste something into you just paste a prompt into Codex or Cowork.

它会去跟应用沟通，这个应用可以是个普通的服务器，也可以是个独立的智能体。

It goes and talks to the app and the app can be either just a regular server or or it can be its own agent.

Codex 对你的了解太深了，它能直接告诉我：这是我跟丹一起做的所有事情。

And Codex has so much information about you that it can just give it here's all the stuff I've been working on with Dan.

这是他可能想用这个应用的几种方式，然后把结果反馈给我。

Here's all the ways that, you know, he might he might wanna use this app and then bring it back to me.

整个体验非常个性化。

And it's this very custom experience.

而且对于像智能体这样的技术产品，一旦出问题，我只要对 Codex 说：去修一下。

And also for a technical product like an agent, when something goes wrong, I can just tell Codex, go fix it.

Codex 就会去跟应用沟通，帮我找出问题出在哪。

And Codex will go talk to the app and figure out what's going on for me.

所以我觉得，当大家都默认每个人都有一个智能体，而且这些智能体以一种奇妙而关键的方式彼此交流时，整个模式就开始变了。

And so I think the whole paradigm starts to change when you assume that everyone's got an agent and those agents are talking to other agents in this like really magical and important way.

在正式开始之前，我还想再聊几个点，因为能谈的内容实在太多了。

There's a couple of more things I wanna touch on before we get started because there's, like, so much to talk about.

一点是，你提到过，用 SaaS 工具的时候，其实可以直接用模型公司提供的令牌。

One is you made this point about SaaS tools not using like, you can use tokens from the model companies, basically, when using a SaaS tool.

再详细说说这个，因为这可能会彻底改变未来 SaaS 公司的商业模式。

Talk a bit more about that because that may change the business model for SaaS companies in the future.

这感觉是个挺重要的变化。

That feels like a big deal.

我觉得这反而可能帮他们保住利润，因为现在所有公司都在急着往产品里加智能代理，觉得代理会成为用户主要的交互方式。

Well, I think it actually may save their margins because right now all the all these companies are rushing to like add a agent to their offering and thinking, oh, the agent is gonna be the main way that I that people interact with me.

但你知道，运行代理是要消耗令牌的，这成本不低。

And I think that and that cost tokens, obviously.

但我个人觉得，等我有了 Codex 或 Cowork 作为主要工作平台，我依然会继续用各种 SaaS 工具。

And I actually think once I have once I have Codex or Cowork as my main work surface, I still wanna use SaaS.

所以这其实是个很不错的预判。

So this is another good prediction.

我现在就会买 SaaS 相关的股票。

I would buy SaaS stocks right now.

我觉得所谓的SaaS末日论根本站不住脚，未来几年SaaS股票肯定会大幅上涨。

I would I think the SaaS apocalypse is dumb, and SaaS stocks will be up majorly in the next couple years.

这不是投资建议，但说实话，我会买SaaS股票。

Not not investment advice, but, you know, I would buy SaaS stocks.

所以这其实能帮你保住利润，因为你现在不需要再想‘我得把AI集成进去’了。

So so so I think it saves your margin because now what you're what the way that you're thinking then is not, I have to build AI into this.

你应该想的是，怎么做出一款让人和AI都能顺畅协作的软件。

It's it's more like, I have to make a piece of software that humans and AI wanna collaborate on together.

这确实有难度，但一旦做成了，成本远比假设每个用户都在消耗token要低得多。

And that's hard, but it's once you build it, it's a lot cheaper than assuming everyone's spending tokens.

我觉得这是个不错的商业模式。

And think it's a good business.

我之所以这么看好SaaS，原因之一是，就像我说的，我们公司内部人人都在用各种AI代理、Codex之类的工具，但我们依然在大量购买SaaS服务，而且SaaS支出每年都还在增长。

And part of the reason I'm so bullish on SaaS is A, everybody internally here is, like I said, we have all got agents and we're all using Codex and whatever, and we still pay for a ton of SaaS and our SaaS spend is up year over year.

我们可不是每件小事都靠‘灵感编程’来解决的。

And we're not like vibe coding every single, like, little thing.

你知道吧。

You know?

我觉得智能代理的作用是增加SaaS的用户数量，而不是取代它。

And I think that what agents do is increase the number of users of SaaS, not get rid of it.

所以我认为，SaaS公司会迎来需求的爆炸式增长，因为会有大量智能代理以极高的频率使用这些产品。

And so I think SaaS companies are going to see an insane spike in the amount of demand that they have because there's going to be tons of agents using these products at a very high volume.

就像我说的，这会带来巨大的基础设施挑战。

And like I said, that's a huge infrastructure challenge.

定价方面也有很多有意思的问题，但这些都让我对SaaS非常看好。

There's a there's a lot of, like, interesting pricing challenges, but it it it makes me very bullish on SaaS.

这说法太棒了。

I love that.

如果从这次对话里还能总结出别的，那丹·希珀，SaaS就是AI的未来。

If anything else comes out of this conversation, Dan Shipper, SaaS is the future of AI.

企业级SaaS。

B2B SaaS.

发个推文吧。

Hashtag send tweet.

我就是特别喜欢这一点。

I love just yeah.

这观点还挺反主流的。

This is quite contrarian.

另一个有意思的地方是，你们在过去一年里员工人数翻了一倍，这跟大家对一家如此AI导向公司的预期完全相反。

And the other interesting piece is that the fact that you guys are hiring that you doubled in people in the past year, which is not what people would have expected from a company that is so AI forward.

说说你们的实际经历吧——原来还是得靠人。

Talk about what your experience there of just, okay, we still actually need humans.

自动化是个假象。

Automation is a lie.

每当你自动化一样东西，为了确保它运行良好，都得有人在后面盯着，检查它是不是真在正常工作。

In the sense that every time you automate something, in order to make sure the automation is working well, you need a human on top of it, like making sure that it's working well.

几年前我写过一篇文章，叫《分配经济》。

And so, you know, I wrote this piece a couple years ago called the allocation about the allocation economy.

人类和AI协作的方式，其实就像当个管理者一样。

Like the idea that the way that humans are gonna work with AI is gonna is gonna be, like, like being a manager.

你要记住，管理者其实花很多时间在干活儿。

And the thing that you have to remember about managers is, like, managers actually spend a lot of time working.

大多数管理者可不是躺在沙滩上度假。

Most managers are not like on the beach.

他们一直在跟下属沟通，琢磨怎么把事情做得更好。

They're like checking in with their employees all the time and and and and trying to figure out, okay, how do we make this work good?

怎么才能做得更优？

How do we make it better?

现在进展怎么样？

How's it doing?

这个人最近状态如何？

How's this person doing?

诸如此类的事情。

All that kind of stuff.

我觉得人类经理和模型经理之间确实有些不同，但归根结底，都需要投入大量时间和精力。

And I think there's there's some differences between it being a human manager and being a model manager, but fundamentally, it still requires a lot of time and attention.

我觉得我们在讨论模型时，恰恰忽略了这一点。

And I think that we kind of missed that in the model discourse.

其中一个原因就是，各种评估基准让AI看起来比实际情况更自主。

And one of the reasons is benchmarks make it look like AI is more autonomous than it is.

我所说的‘自主’有特定含义，我来试着说清楚。

And by autonomy I mean something specific by autonomy, and I'm gonna try to express this.

这其实有点难表达，但我自己慢慢体会到了这个矛盾。

It's, like, a little hard to express, but I learned this for myself because I've feeling this paradox a little bit.

我感觉到，虽然我们有这么多自动化和AI，但自己反而干得更多了。

I've been feeling the, like, we have so much automation, so much AI, and I also work way more.

当我自己搞了一个评估标准后，这个矛盾才开始有点想通了。

And I think part of the paradox part of the paradox started to, resolve for me a little bit when I made my own benchmark.

所以我做了一个叫‘高级工程师评估标准’的东西。

So I made this senior it's called the senior engineer benchmark.

那AI和人类工程师比起来，到底有多厉害呢？

And it's like, how good is AI versus a human engineer?

我是这么做的，还是用了一个实际的项目来验证。

And the way that I built it is, again, have this app proof.

我只是在空闲时间随手写出来的，一边跑着Every的其他功能一边搞的。

I just vibe coded it on the side and like while running the rest of Every.

上线之后，因为完全是凭感觉写的，结果立马出问题了，我根本修不好。

And when we launched it, because it was completely vibe coded, it just started going down and I couldn't fix it.

那场面真是挺丢人的。

And it was very embarrassing.

我简直颜面尽失。

I had a lot of egg on my face.

但说实话，产品本身是能用的。

And like the product worked.

我们内部已经测试过了。

We we tested it internally.

我们有很多内测用户，但上线第二天，每隔十分钟服务器就崩一次，大家都盯着我，我只能干着急，说我不清楚到底出了什么问题。

We had a lot of beta testers, but like the day after launch, it was like just every like ten minutes, the servers would go down and people were looking at me and I'd be like, I don't know what's going on.

就像让Codex帮我修一下。

Like Codex, fix it.

结果Codex也说，我不清楚问题出在哪。

And Codex is like, I don't know what's going on.

其实Codex明明知道问题出在哪。

Or really Codex was like, I do know what's what's going on.

它说它修好了。

I fixed it.

可一修复，又冒出四个新错误，就这样来回打转，我根本睡不着觉。

And then it it would cause four other errors and then you're just going around in a circle and I wasn't sleeping.

我真是拼了命地靠直觉写代码。

And I I I vibe coded so hard.

结果肘部都得了肌腱炎。

I got bursitis on my elbow.

这其实是个生活教训。

So that's a there's a life lesson in there.

vibe 编程导致的肘部劳损。

Vibe coder elbow.

是啊。

Yeah.

所以后来，我请了两位资深工程师，各自独立地修复了这个问题。

So anyway, I got a I got actually two different senior engineers to fix it independently.

于是我现在有了两套完全不同的代码重写版本，能看出他们各自的解决思路。

So I have two different rewrites of the code base that tells me how they did it.

对吧？

Right?

所以现在，每当我们引入新模型时，我只需要给新模型一个提示。

And so what I get to do is when we get new models, I just give the new model a prompt.

我会说：这个是 vibe 编程的产物。

I say, like, this is vibe coded slap.

如果你要从头开始重写，你会怎么改？

If you wanted to rewrite it from first principles, how would you rewrite it?

去试试看吧。

Go do it.

直到GPT-5.5之前，所有模型的得分大概都在100分里拿30分左右。

And all the models until GPT 5.5 got, like, a 30 out of a 100.

而资深人类工程师的得分通常在八十几到九十几分之间。

And senior like, human senior engineer gets, like, high eighties, low nineties out of a 100.

所以还有很大的提升空间。

So there's a lot to go.

后来我试了GPT-5.5，它拿到了62分。

And then I tried GPT 5.5, and it got, like, a 62.

顺便说一下，这个62分是GPT-5.5用了OPUS 4.7方案得来的。

And mind you, the 60 the 60 score was GPT 5.5 using an OPUS 4.7 plan.

OPUS方案里有4%的版本非常出色。

Opus 4% of the plans are very good.

不过，GPT 5.5 是唯一一个有主见和自信，直接把旧代码全删了、从头重写的模型。

G p 5.5 is the only model though that has the sense of agency and confidence to just, like, rip out old code and just, like, actually rewrite for first principles.

其他代码模型呢，总是小心翼翼，想修修补补，结果就在边上打补丁，还说：这活儿太大了。

Other coding models, they kinda, like, try they, like, end up papering over the edges or around the edges, and they're like, oh, this is a big job.

我还是改一点算了。

Like, I'll just do a little patch.

你可别这样。

And you're like, no.

我明明特意跟你说过别这么干。

I, like, specifically told you not to.

所以 GPT 5.5 的分数一下子提升了 30 分。

So GPT 5.5, there's, like, a 30 bump in the score.

六十分，满分一百。

60 out of a 100.

很明显，再过一年，甚至更短时间，它就能达到资深工程师的水平了。

It's, like, very it's very clear that in a year or less, it's gonna be senior engineer level.

对吧？

Right?

这会让你在心里形成某种画面，尤其是考虑到我给这个基准测试起的名字——我觉得很多基准测试都是这样做的。

And that gives you a certain picture in your mind, especially based on how I named the benchmark, which I think a lot of benchmarks do.

我可以告诉你，当我们达到那个阶段时，我会很容易把当前模型的得分直接调成零。

And I can tell you that when we get to that point, I will be very it will be very easy for me to change the benchmark to zero out the current model.

所以它的得分就是零分。

So that gets a zero out of a 100.

看起来好像提示里完全没用什么技巧或思考，就是那种‘ vibe 能搞定，直接从第一性原理重写’的风格。

And so, for example, it seems like there's no skill or no thought into the prompt, which is this is vibe coulda slap, like, fix it from first principles.

但实际上，我花了挺长时间才设计出一个既不泄露答案、又能逼出模型真实能力的提示。

But actually, it took me a a while to get to a prompt that didn't give away the answer, but but got the model to reveal what it's capable of.

我最初给它的提示，其实是我在生产系统出问题、系统即将崩溃时，用来尝试修复的那个原始提示——那天早上我醒来，心想：好吧。

And the original prompt I gave it was the original prompt that I gave it when when I was trying to fix the issue and production was going down, which is, like, I'd woke waking up I'd I'd woken up in the morning, and I was like, okay.

昨天我们收到了四五个故障报告。

We had four or five reported issues yesterday.

我希望你把所有问题都过一遍，然后制定一个解决所有问题的计划，并且去执行。

I want you to go through all the issues and then come to, like, a make a plan for how to resolve all of them and go do it.

对吧？

Right?

目前市场上所有的代码模型，我敢打赌这是一条预测。

And every coding model on the market and I I am I'm pretty sure this here's a prediction.

我敢肯定，一年后市场上所有的代码模型还是会这样。

I'm pretty sure every coding model on the market will still do this in a year.

市场上所有的代码模型都会认真对待这条指令。

Every coding model on the market will take that instruction seriously.

如果我跟它说：这里有好些问题。

And if I tell it, here's a bunch of issues.

去修一下。

Go fix it.

它们就会直接去尝试修复这些问题。

They will just go try to fix the issues.

真正的资深工程师会去查看代码库，然后说：这代码简直一团糟。

What a actual human senior engineer does is they go look at the code base and they're like, this is a piece of shit.

这人根本不知道自己在干嘛。

This guy doesn't know what he's doing.

接着他们会说：我们得把大部分代码重写一遍，这会很难，而且有风险。

And then then they say, we're gonna have to, like, actually rewrite a lot of this, and it's gonna be hard and risky.

我知道你不想听这个，但我们真的得这么做。

I know you don't wanna hear that, but, like, we're gonna have to do that.

如果你去问模型：嘿，

And if you ask the model, hey.

我们该这么做吗？

Like, should we do that?

它可能会慢慢想到这一点，但不会主动去干。

It'll it'll probably it'll probably get there, but it's not gonna do it on its own.

而且有很多因素在阻碍它这么做。

And it and there's a lot of incentives pushing against it doing that.

即使它做到了，我们依然有更高的目标要去追求。

And even if it does that, there's there's always a higher frame for us to go.

所以我觉得，当我们讨论基准进展的时候，必须从这个角度去看——基准提升的前提，是我们能明确界定、清晰表达、并且可以量化的问题。

And so I think it's it's really important when when we think about benchmark progress to think about it from that perspective, which is benchmarks rise on problems that we've framed, that we can articulate, that we can score.

有很多工作本质上是人的工作，不写下来就无法衡量，但那种思考、构思、或者把它写出来的过程，本身是无法量化的。正因如此，就算基准饱和了，也不等于我们就能完全取代资深工程师。

And there's a lot of work that's human work that it it can't be scored until you write it down, but the act of thinking to prompt it or write it down is is something that you can't measure, but, like, kinda means that even if the benchmarks get saturated, it doesn't mean the same thing as we you totally replace all senior engineers.

所以我觉得，尽管模型在自动化方面越来越强，我依然在招聘工程师。

And it's and I think it's why even though the models are getting better at automation, I still hire engineers.

我特别想跟大家聊聊本季的赞助商——Vanta。

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

Vanta 帮助了超过一万五千家企业，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.

团队借助 AI 正以前所未有的速度开发和发布产品，但随之而来的，是产品和业务中引入的风险也达到了历史最高点。

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，都感受到越来越大的压力——要保护公司、业务，更别提客户的数据了。

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.

因为事情变得太快，他们总是在被动应对，不得不猜测优先级，还得凑合用过时的方案。

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

Vanta 通过超过 35 个安全与隐私框架，比如 SOC 2、ISO 27001 和 HIPAA，自动完成合规与风险管理。

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

这能让企业快速达到合规，并持续保持合规状态。

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企业的成败。

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，您还能获得 1000 美元的折扣。

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

就是 vanta.com/lenny。

That's vanta.com/lenny.

我最近在播客里提过一件事，听说有人在收购 2021 年、2022 年之前写的人类代码，那时候 AI 还没兴起，这些代码现在成了很值钱的数据。

One thing I mentioned recently on the podcast, I heard that speaking of the code that you have of, like, humans writing code, data labeling companies are buying code that was written before 2021, 2022, before AI became a thing, is, like, very valuable data.

古老的代码。

Archezonal human code.

没错。

Yeah.

正是如此。

Ex exactly.

说得对。

That's exactly right.

有意思的是，正是这类代码被用来训练这个模型。

And it's so interesting that that's exactly the kind of code used to build this model.

我想澄清一下，这其实挺有意思的。

Well, what's interesting so I wanna I wanna clarify there.

我并没有让人类逐行手写代码，因为我觉得那样做有点荒谬。

So I did not have a human write the code all by hand because I actually think that that's sort of it feels silly to me.

说实话，我不太在意，因为如果一个工程师不用AI，我根本不会和他合作。

Like, I don't really care because I know if if an engineer is not using AI, like, I'm not gonna work with them.

我真不在乎。

I don't really care.

这就像拿人和汽车赛跑一样。

It's like it's sort of like, am I gonna race a human against a car?

我肯定不会这么干。

Like, I probably wouldn't do that.

但我肯定会让人开车和另一个人开车比一比，看谁更强。

But I would race a human in a car versus another human in a car and say, which one's better?

而这个评测的设置方式是，那些工程师确实用了AI，但他们用的方式我没法学，因为我根本没时间去搞懂，说实话，我也不想花精力去啃那个代码库。

And in this case, what the the way the benchmark is structured is, yeah, like, these human engineers use AI but they use it in a way that I could not because I didn't understand it and I didn't have time and I didn't really wanna like go in and try to understand the code base to be honest.

我们在做评测时，这点特别重要：AI是一种人人都能用的通用技术。

And I think that's a really important thing when we think about benchmarks is AI is a broadly distributed technology that any human can use.

当我们拿AI和人类做对比时，其实真正比较的是：一个用AI的人，和另一个用AI的人。

And when we are benchmarking against humans, AI against humans, we're actually really always talking about one human using AI versus another human using AI.

因为AI自己不会用自己。

Because AI doesn't use itself.

在某种程度上，它可能确实能以这种有点自我指涉的方式运作，但在任何实际应用场景里，总会有个人在旁边盯着，确保它正常运行。

It it may be able to in this, like, slightly somewhat recursive way, but there's in any real use case, there's always a human, like, pretty close to it making sure that it's working.

明白了。

Okay.

我想先总结一下我们这一部分的内容。

I wanna try to wrap up our first bucket.

能聊的话题实在太多了。

There's so much to talk about.

我根据你的预测，列了个清单，说说大家该怎么做才能成功。

I made a little list of things that I think people should do based on your predictions to be successful.

这个我们待会儿也会细聊，不过先说几个重点。

We'll talk about this at the end too, but just a few things.

第一，多用 Codex 或 Claude Code 来做你的工作，尤其是浏览器里的那些工具，一定要用起来。

One is start using Codex or Claude Code more and more for the work you're doing, and especially the browser, use tools inside of it.

第二，让你的智能代理能够接入并使用你的产品。

Two is allow your a allow agents to be to use your products.

如果你在开发一款 SaaS 工具，就让代理能轻松成为你的用户。

If you're building a SaaS tool, make it easy for agents to be a a a user, essentially.

第三，开始想想能不能做个 Slack 机器人，拿来试用一些工具。

Three is start thinking about some Slack bot that you can work with, like try out tools.

比如 Slack 自己就有个很棒的机器人，我还没怎么用过，但很多人都说特别好用。

Like, know Slack has their own Slack bot that I think is really good too, and I haven't played with it, but people really like it.

所以找找看，有没有哪个工具能变成你们公司内部的 AI 代理。

So look for, I guess, a tool that could become the AI agent within your company.

赶紧买点 SaaS 股票，这不是投资建议。

Buy SaaS stock ASAP, not investment advice.

我觉得完全对。

I think that's totally right.

我稍微补充一点：当你为代理设计软件时，现在的做法是做个命令行工具让代理用，也就是你把任务交给他，他去调命令行。

I will like, my slight tweak is when you're thinking about building your software for agents, the current model is I'm building a CLI that an agent uses, but they're you're using it in a sort of, like, I'd they're debt being I delegated a task to the agent and the agent's using the CLI.

但我觉得未来会是：你和代理一起用这个应用。

And what we where I think it's going is you and the agent are using the app together.

代理可能在用命令行工具，而你用的是网页界面，两者必须保持同步。

The agent's probably using the CLI, but you're using the web interface and they're they both need to be in sync.

这确实是个全新的挑战，特别有意思。

And that is, I think, a new challenge that's really interesting.

太棒了。

Awesome.

在进入下一个话题之前，还有别的要说吗？

Anything else before we get to our next category?

关于SaaS。

By SaaS.

这标题就这么定了。

That's the title.

天哪。

Oh, man.

好的。

Okay.

所以第二个预测方向是，我们未来要做的工作形态会发生变化。

So the second category of predictions is around just the shape of the work that we're gonna be doing is gonna change.

你预测会怎样？

What do you predict?

关于工作形态，其实有很多有意思的现象。

There's all this interesting stuff in terms of in terms of the shape of work.

比如，一旦你进入一个环境，能派发任务给异步代理，再配上像 Codex、Claude Code 这样的工作界面，整个流程就开始变了。

Like, once you're in this land where you've got, you know, these you've got async async agents off that you delegate work to, then you've got your like Codex, Claude Code, like work surface that that starts to happen.

我们内部经常看到，大模型公司也一样，拉取请求的数量简直暴增。

So one thing that we see a lot internally, and you also see this in the big model companies, is the number of pull requests that you get is, like, skyrockets.

有些做咨询、运维，或者干脆就是编辑的人，也开始频繁地提交拉取请求了。

You know, we have people, you know, in consulting or in ops roles or whatever who are or or edit editors just, like, making pull requests.

这确实挺棒的。

And, a, that's really cool.

这是一种完全不同的工作形态——你可以预期，公司里或用户中会有更高比例的人，去做过去只有技术人员才做的事。

And it's a very different shape of work where you should you can expect that a higher percentage of your company or your users are gonna be doing things that previously only technical users can do.

这给那些得处理大量新代码的人带来了巨大压力，他们得想办法应对这些变化。

And what that does is it creates all this pressure on the other end for the people who have to deal with all of the new code for how to deal with that.

所以我觉得这里面有很多值得探讨的地方。

And so I think there's a lot of, there's a lot of interesting things that happen with that.

比如 OpenClaw 就是个例子。

Like, so for example, like Open Claw.

我之前提到过。

I mentioned that earlier.

Pete 每天能收到成千上万的合并请求，然后他直接启动五万个 Codex 实例来筛选，最后合并其中大约一千个。

Pete gets, like, thousands of pull requests a day on Open Claw, and then he has, like and then he just spins up, like, 50,000 Codex instances and then sorts through them and then merges, like, a thousand of them.

这简直太疯狂了。

It's really crazy.

我觉得这种情况会变得越来越普遍。

I actually think that that's going to be more and more common.

这引发了很多有趣的问题，比如到底该合并哪些合并请求。

There's, like it brings up a lot of really interesting questions around which pull request should you merge.

当你在流程的某个环节增加能力时，往往就会打破平衡。

And, you know, when you whenever you add capacity in one part of your process, like, it breaks things.

以前做东西特别难，现在却变得很简单了。

It used to be really hard to build things and now it's very easy.

所以关键不是我们能不能做出来。

So the the point is not, can we build it?

而是它和我们已有的系统搭配起来是否合理。

It's like, would it make sense with the rest of what we've built?

我们该怎么保持整个系统的连贯性和一致性？

And how do we keep a, like, sense of a coherent whole?

还有，我们该删掉哪些东西？

And also, what do we delete?

我觉得Anthropic在这方面做得特别好。

I think Anthropic does this really well.

他们经常从Claude Code里删掉很多功能，避免系统变得臃肿。

Like, they they delete a lot of stuff from Claude Code to make sure that it's not bloated.

所以我觉得，这方面会有很多这样的情况发生。

So I I think there's a there's a lot of that gonna happen on one side.

很多技术人员能做技术工作，同时也要负责确保这些工作能以连贯、统一的方式融入产品或流程中。

There's a lot of technical people can do technical work, and then technical people are in charge of making sure that that work gets into a product or into a process in a cohesive, coherent way.

而且他们的产品人员也会做同样的事。

And also their product people are gonna be doing that too.

我觉得这其实挺棒的。

And I think that's that's quite cool.

我最近听不少人说，现在每个人都能做所有事——工程师能做设计，产品经理能写代码，市场人员也能上线功能。

Something I'm hearing from people is that now that everyone can do everything, like engineers can design, PMs can code, marketing people can ship stuff.

结果大家反而搞不清楚，自己到底该干啥了。

There's just this, like, confusion about what the hell is my job anymore.

是啊。

Yeah.

那我到底该对什么负责呢？

What am I responsible for exactly?

对啊。

Like Yeah.

我是不是该负责把东西推出来？

Am supposed to be shipping stuff?

我还算是一名市场人员吗？

Am I still a marketing person?

这在业界制造了大量困惑和不确定性。

And it's just creating a lot of confusion and uncertainty in the world.

就像是

Just like

我觉得

I think

确实如此

that's for

真的。

real.

我觉得Every特别的地方在于，每个人都是通才，特别喜欢插手各种不同的事情，不管怎么形容吧。

And one of the things that I think is special about Every is everyone is sort of a generalist and really loves like having their fingers in a lot of different pots or whatever the metaphor is.

我觉得过一阵子这种情况会慢慢稳定下来，就会显得更正常了。

I think that'll probably settle down at some point and it'll feel more normal.

就算市场人员开始碰网站，他们该做的市场工作还是得做。

Like, marketing people are still gonna do marketing even if they're touching the website.

现在这本来就是市场工作的一部分。

Like, that's just part of marketing now.

但我还觉得，现在当个通才能走得更远，这点特别棒，尤其是对于小公司来说。

But I also think that you can get a lot further being a generalist now and that's like really cool especially for for smaller companies.

另外我觉得有意思的是，确实出现了一些全新的职位。

The other thing that I think is interesting is there are definitely some new job roles that are a thing.

越来越明显的是，‘前线工程师’这个概念真的站得住脚。

And the thing that is becoming really clear is the whole forward deployed engineer concept, I think, is for real.

这其实源于每个智能代理都需要有人来配合。

And it comes out of every agent needs a human.

你去那些大模型公司，他们会内部部署一些智能代理。

You like, you go to the big model companies, they have they they have these agents that run internally.

他们还专门有团队负责维护这些代理，我觉得这些团队不会消失。

They have, like, teams of people that run these agents, you know, and I I don't think those teams are going away.

模型会变得越来越强大。

The models are gonna get more powerful.

代理也会越来越强，数量也会增加，但始终需要人来管理它们。

The agents are gonna get more powerful, and the number of agents is gonna grow, but people are still gonna manage them.

这就需要一种非常特定类型的人才。

And so that looks like a very specific kind of person.

我们公司内部就有这么几个人，专门负责确保代理正常运行、做正确的事。

And, you know, we have a couple of those people internally here and it's like the the people who are in charge of making sure your agents are working and doing the right thing.

我们还做咨询业务。

We also do consulting.

我们会把这种服务提供给客户，我觉得这正是很多人需要的，也再次说明了：自动化本以为会取代工作，结果却创造出了新的岗位。

So, we we we lend that out to people and and I think that's a big that's a big thing that that people want and it's another one of those places where you're like, automation was supposed to take away jobs, but it looks like it just created one or many.

你明白吧？

You know?

还有一种特别的工程师，特别喜欢做这种事，比如我们这儿的尼特什，他就特别符合这种类型。

And there's a specific type of engineer that really loves, you know, Nitesh, who's one of our who who fits this.

他是个AI工程师。

He's an AI engineer.

他属于那种前线部署型的人才，现在就在我们团队里。

He he fits sort of forward deployed category and he's on our team.

他大部分时间其实都在Slack上跟我们的一个智能代理聊天。

He spends most of his time actually talking to one of our agents in Slack.

我们内部有个叫克劳迪娅的代理，负责管理整个咨询业务。

We have an agent internally called Claudia, which runs our whole consulting practice.

他经常在Slack上跟它互动。

And he spent a lot of time in Slack.

他当然也会写代码，用Claude Code这类工具，但更多时候就是跟它聊天，问它：‘你为啥干了这么蠢的事？’

There is code and he is using Claude Code and other things like that, but a lot of it is just talking to it and being like, why did you do this dumb thing?

咱们得把这个改掉。

Let's fix that.

你知道的。

You know?

有些工程师特别喜欢这种工作方式，喜欢亲手操作最新工具，也喜欢在工作环境中直接参与这些正在开发的东西。

And so there are certain kinds of engineers that I think love that and love having their hands on the latest thing and also love making this, like, being that's, like, in in the works in a workspace.

这跟传统软件开发的方式看起来不太一样。

And it looks a bit different than more traditional building more traditional software.

你的意思是，我们离不需要人类的代理还远得很，对吧？

And your sense there is we're not gonna we're not near a place where these agents don't need a human.

你已经说过好多次了，代理离不开人，既要 setup，还得长期维护。

You've said that so many times now that agents need a human, and there's kind of like the setup part, and then there's the maintaining it forever part.

我听出来你觉得这两部分都挺重要的，也就是说，这活儿短期内不会消失。

And it feels like both are important Is what I'm hearing, like, this is gonna be a job for a long time.

AI 还没聪明到能完全自己搞定一切，至少短期内不会吧？

AI is not gonna get smart enough to just automate its be fully automated for a while?

是的。

Yes.

我一方面非常相信AI的能力，另一方面也极度看好人类在确保AI正常运行中的作用。

I'm simultaneously extremely AI pilled, extremely, and very bullish on humans and the role of humans in making sure that AI is working

没错。

well.

有意思。

Interesting.

明白了。

Okay.

你刚才提到的这两类情况，我理解的是，软件交付的速度越来越快，随之而来的是需要审查大量粗糙的输出结果，工作量也大幅增加。

So the two kind of buckets here that you're talking about, one is, like, the way I think I hear what you described earlier is just the pace of shipping software and everything is just increasing, which also means there's so much more work reviewing all this sloppy output.

我刚和一个数据科学领域的同事聊过，他说他们团队以前的工作就是做分析、回答问题，判断某个实验结果是否积极。

I I was just talking to a data science friend and he was saying how his team is just his data science team is just their job used to be do analysis, answer questions, see if this experiment was a good was a was positive.

现在呢，人人都在做这些事，大家互相分享结果，但结果却并不理想。

Now it's just everyone's doing that, and they're sharing their results, and they're and they're like, no.

这不对。

This is not correct.

他们现在大部分工作都是在审查糟糕的数据分析结果。

And most of their job is now reviewing bad data science work.

这正是问题所在，工程师那边也是一样，你确实需要更多的工程师和数据科学家来应对这种情况。

Which is the problem, and it means that and the same thing are is happening with engineers, and it means that you need more like, you actually need that engineers for this, and you need data scientists.

这意味着你还没有建立起合适的系统或代理来帮你处理这些问题。

And it means that you haven't set up the appropriate systems or agents to help you with this.

比如在大型模型公司，至少有一家公司专门设了一个数据科学机器人，公司里每个人都能向它提问，它连着数据仓库，清楚谁是谁，知道每个人在数据层面有什么访问权限。

So like, the way that it works inside of the big model companies, for example, like, at least one of them has literally a data science bot that every single person in the org can query that is hooked up to their data warehouse that knows who's who so that it knows at the warehouse level, like, who has permission to access what.

因为有个团队专门维护这个机器人，所有那些基础问题，就算偶尔答错，他们也会不断修正，确保它越来越准。

And so all of the basic questions, because there there's a team that sets up this bot, all of the basic questions that people might wanna ask that it sometimes gets that might get wrong, that they're constantly making sure it's getting it right.

这样一来，数据科学家就不用再应付一堆琐碎无聊的问题了，因为有另一个团队在专门打造一个能高效处理这些问题的智能代理。

And so the data science team doesn't have to answer all the, like, bullshit questions because there's another team building an agent that that that is set up to do that really well.

但如果没这个团队，数据科学家真得被逼疯了。

But if the team didn't exist, the data scientists would hate their lives.

是的。

Yeah.

不过这确实让工作没那么有趣了，因为你整天就在那儿帮别人收拾烂摊子。

It does, though, make the job maybe less fun because you're just sitting there, you know, gardening people's sloppy work.

但我觉得这反而让工作更好了，因为数据科学家不用再应付那些琐碎的请求了。

Well, that's what I think is, like, it it it can actually make the job better because for the data scientists, you are now not dealing with all the silly requests.

你现在处理的是更深层的问题，那些基础性的问题已经交给专门负责搭建代理系统的团队去应对了。

You're dealing with the deep the deeper questions that are harder for the the team who's dealing with all the basic requests and building agent to do that.

这就像是把杂事都过滤掉了，让你能专心做真正重要的事。

It's it's like filtering all that stuff out so you can focus.

我最近一直在想一个问题。

Here's a question I've been thinking about.

我本来没打算谈这个，但确实一直萦绕在我心里。

I was not planning to talk about this, but it's something that I've been thinking about.

现在哪个技术岗位的变化最小？

The question is which product tech role is the least changed now?

工程师现在写代码，100%都靠AI了。

So, like, engineers, 100% of code AI now.

这简直成了完全不同的工作。

It's like a completely different job.

产品经理呢，很多PRD都不用写那么多了。

Product management, a lot of the, you know, PRDs are you don't have to write as much.

你自己就能直接上线代码。

You can ship code.

不用再等别人了。

You don't have to wait for people.

整个设计流程，已经没人花了。

Design the whole design process, dead.

根据最近几位嘉宾说，根本没时间走完整的设计流程了。

According to recent guests, just like there's no time to do the whole design process.

这角色跟以前完全不同了。

Very different role.

数据科学现在的工作方式也完全不同了。

Data science, very different work now.

还有市场营销。

There's marketing.

还有销售。

There's sales.

我有个问题。

Here's the question.

你觉得到目前为止，哪个角色变化最小？

What do you think is the least fundamentally changed role so far?

有意思的是，我不确定算不算，但像CEO和投资人，他们用不用这些工具似乎还是可有可无的。

Well, one interesting thing is, you know, I don't know if this counts, but, like, CEOs and investors, it seems still very, very optional whether or not they use this stuff.

对，确实如此。

Mhmm.

看起来就是这样。

It seems that way.

我觉得恰恰相反。

I I I think the opposite is actually true.

以我的经验来看，我们经常和高管团队一起用这些工具。

Like, my experience, we do a lot of this with senior executives and senior leadership teams.

我觉得，公司能走多远，完全取决于CEO对AI的投入程度，这事儿没法交给别人。

My experience is that your company is only going to go as far as your CEO goes in AI and it's not something you can delegate.

你必须亲自参与，不然就缺乏直觉。但长期以来，大家总觉得这是执行层的事，我不用管。

You have to have your hands in it because you don't, otherwise, you don't have an intuition for it but for a long time, it has seemed like, yeah, that's something that the people who are doing the work have to do, but, like, I don't have to do that.

我就告诉他们该怎么做就行了。

Like, I'll just tell them what to do.

所以，作为CEO，你完全可以继续过着原来的日子。

And and so I think if you're a CEO, you kinda can get away with your day looking very similar.

但我觉得，迟早有一天会突然意识到：糟了。

I I think that will change rapidly at some point where it'll be like, oh, no.

我已经落后太多了。

I'm, like, way behind.

但目前来看，甚至中层管理者这类人，情况也差不多。

But for now, because or maybe even middle managers, like, those kinds of people, I think, are are it's fairly similar.

我觉得销售可能是个例外，因为那本来就是面对面的工作。

I think, like, maybe sales because That's exactly what in person.

没错，就是这样。

That that's yeah.

我同意这个看法。

That's my vote.

你明白吧？

You know?

像BDR这类岗位，已经开始慢慢受到影响了。

There it's sort of creeping up in the kind of BDR.

我们可以轻松处理很多BDR类型的咨询问题。

Like, we can deal with a lot of, you know, BDR type type queries.

你接触的都是真正有需求的人。

You're only talking to, like, people who actually want it.

在销售这行，做调研真的特别有用。

And you can do for sales, it's like re it's so useful to to, like, do research.

我有个特别喜欢的Codex使用案例，是我们正在招聘学习与发展负责人。

Like, my favorite Codex like, one of my favorite Codex experiences is we're hiring a head of L and D.

我们当然会发招聘启事，但我突然想到，纽约有家叫General Assembly的公司。

And I you know, we always put out a job post, whatever, but I was like, I feel like there's this company called General Assembly in New York.

他们做科技教育已经很多年了，口碑一直很好。

And they do like, they've done really good technology education for a long time.

我觉得，如果有人曾在General Assembly工作过，现在又对AI感兴趣，那肯定特别合适。

And so I was like, I feel like someone who is into who who who worked at General Assembly and is now into AI would be really good.

我就直接在Codex里输入了这个想法，然后去忙别的事，过了一会儿回来一看。

And I just, like, literally typed it into Codex and then, like, went off and was doing something else, and I came back.

结果真找到了一个完美人选。

And it found, like, this the perfect guy.

他之前在General Assembly当讲师，特别懂AI，还关注了我的Twitter。

It was, like, worked at general assembly, was an instructor, like, is super AI pilled and follows me on Twitter.

所以我直接给他发了私信，然后约他吃了顿饭。

So I just DM'd him, and then I had dinner with him.

这简直太不可思议了。

And it's like, that's crazy.

你懂的。

You know?

以前做这些事得花上好久，现在对销售、招聘这些工作来说，价值太大了。

That would have taken so long before and super valuable for sales, for recruiting, all that kind of stuff.

是的。

Yeah.

我第一反应就是销售。

Sales is where my mind went.

在销售的前端，AI在挖掘和筛选潜在客户方面帮了大忙。

Like the top of funnel AI is helping a lot with sourcing and qualifying things like that.

感觉销售的工作本质并没有发生根本变化。

It feels like the the work of a salesperson is not fundamentally different.

是的

Yep.

客户支持也彻底变了样。

And and customer support's fundamentally changed.

这挺有意思的。

So that's interesting.

销售方面。

Sales.

到目前为止，这些人的情况都还不错。

So far so good for for those folks.

是的

Yep.

好的

Okay.

咱们来总结一下这一类预测，主要是关于工作形态会怎么变化。

So maybe just summarizing some of the predictions in this bucket of just like the shape of the work, how it's gonna change.

到目前为止，我听到的是，未来工作中会更多地涉及审阅他人的成果。

What I'm hearing so far is there's gonna be a lot more reviewing of other people's output as a part of the work.

其次，还得花大量时间去引导AI代理，确保它们按你的要求运行，部署之后还要持续监控，保证它们一直正常工作。

And then two, there's gonna be a lot of, like, almost babysitting of AI agents to make them do the thing you want them to do for deploying and then just gardening them along the way, make sure they continue to do their work.

在进入第三个类别之前，还有没有其他补充？

Anything else before we get into our third bucket?

我觉得与其说是盯着AI代理，不如说前线团队是在构建一套系统，让知识较少的人也能安全使用，而不会出错。

I would sort of split it into less babysitting agents and more your forward deployed team is trying to build a whole system that makes it so that people who have less knowledge can use that system without, like, doing something dumb.

这其实是个非常有意思的工程难题。

And that's like a really interesting engineering challenge.

我觉得‘盯着’这个词听起来好像……是的，就是等着它出错再修。

I think babysitting kinda makes it feel like it's yeah.

你可能就是等着它出问题，然后去补救，诸如此类。

You're just kinda like, you know, waiting for it to fuck up and then fixing it or whatever.

这种情况确实存在，但我认为更多的情况是，这是一个极具挑战性的工程任务：构建一套系统，让组织里每个人都能完成过去只有技术人员才能做的工作。

And you can you can that can be the case, but I think a lot of it is this extremely interesting engineering challenge of building a system for to enable everybody else in your organization to do what used to be a technical job.

如果你不是那种人，比如你是数据科学家之类的，那你就能更深入地利用AI去探索一些真正重要的问题，这些问题最终可能会反馈到前线工程团队的工作中，但它们更具开创性、更前沿，你也得应对更复杂的问题。

And then if you're not one of those people, like you're the data scientist or whatever, you can go a lot deeper with AI into like really important questions that eventually probably filter into the work that the, you know, the forward deployed engineering team is doing, but is like more generative and more new and and and you're you're dealing with harder questions.

还有最后一点我觉得特别有意思：我们会越来越多地阅读AI生成的文档和邮件，而且我们会很喜欢。

One other one last thing that I think is really interesting is I think that we will be reading way more AI generated writing in documents and emails, and we will like it.

其实我们在编程领域已经这样做了，比如我们会读AI生成的计划文档。

And I think we're, we will, we are already doing this in coding where we read plan documents.

我不希望工程师亲手写计划文档。

Like, I don't want an engineer to handwrite a plan document.

那太荒谬了。

That would be very silly.

这明显很荒谬。

It would be, it would be obviously silly.

我觉得同样的道理也适用。

And I think the same is true.

比如我们在2025年底做季度规划时，就完全用Notion代理来完成的。

You know, when we did our our quarterly planning for Every at the end of twenty twenty five, we did it all with Notion agents.

我们当时用了一堆Notion代理，其实主要就是一个，再配上公司整体的战略方向，然后让全公司每个人去跟这个代理对话，问他们去年发生了什么、进展如何、目标是什么、今年想做什么、关键指标有哪些，代理还会提出质疑，接着再问这些内容跟公司整体方向有什么关联——就是这类问题。

And we just had a bunch of Notion agents and or we had really one Notion agent, and then we had a top level company strategy and then we had everybody in the company just talked to an agent and it asked them about what happened last year, how did it go, what were your goals, what what do you want to do this year, what are your metrics, it pushed back, and then it was like, how does it, how does this relate to the overall company idea, like, all that kind of stuff.

然后我就收到了一份份特别出色的AI生成报告，每支团队的季度计划都写得非常到位。

And then I got, I got these like incredibly good AI generated, like, strategy reports or or plan like, quarterly plans for for each part of each team.

接着我就可以进去看看，嗯，好吧。

And then I could go in and be like, okay.

谁需要跟谁沟通？

Who needs to who's, like, who needs to talk to each other?

哪些团队其实彼此需要协作，但他们自己都没意识到？

Like, which teams need to talk to each other that, like, don't know they need to talk to each other?

哪些计划质量不高，哪些又特别出色？

And, you know, who's which one of these is, like, at like, actually low quality or which one of these is high quality?

这类事情让整个处理过程变得轻松多了。

Like, all that kind of stuff makes it it makes it a lot easier to process.

我现在经常看到这种情况。

And I see that all the time now.

我经常收到AI生成的内容，但AI写的文档有好坏之分，这点很明显。

Like, I I I consistently get AI generated stuff, and there is a difference between an AI generated document that's slop and not.

差的那份，写的人花的时间还没我读完它多，而且他们对每句话都不负责。

And the slop one is it took them less time to make it than it takes me to read it, And they don't stand behind every line.

所以我的要求是，如果你给我一份AI生成的文档，我觉得没问题。

So my expectation is, if you send me an AI generated document, I think that's great.

但如果我们讨论的时候，你明显根本不知道里面写了什么，那绝对不行。

And if we talk about it and it's clear you have no idea what's in it, like, big no no.

这种事不被允许。

Not allowed to do that.

我觉得现在大家对AI生成内容的抵触迟早会消失，因为当我的团队成员好好引导时，GPT-5.5能写出的战略文档，比他们自己瞎敲键盘强太多了。

And I I think we have this this aversion to AI generated stuff that will go away because the kind of strategy document that g p t 5.5 can write when it's directed well by someone on my team is way better than, like, them just, like, dinking and dunking, like, like, their fingers on the keyboard.

没错。

Right.

其实大多数人写东西都写得不怎么样。

Like, most people are really bad at writing their documents.

所以要求其实很低。

So the bar is low.

对。

Yeah.

邮件也是同样的情况。

And and same thing with email.

我现在大部分邮件都是由 GPT 5.5 和 Codex 代笔的。

Like, I most of my email is written by g p t five point five and Codex right now.

我反而希望邮件能直接注明是 GPT 5.5 写的，我甚至可能主动改成这样。

And I wouldn't I honestly would prefer it to say that it's coming from g p t five point five and I may change it to do that.

前几天我遇到一件事，得给一位投资人发邮件，就让 Codex 去处理。

But I had this I had this experience the other day where I had this I had to send an email to to one of our investors and asked Codex to go do it.

Codex 通常都会先问我确认，但这次它没问，直接发了。

And use like, Codex knows to ask me and it usually does, but this time it didn't.

我压根没看那封邮件，就让它发出去了。

And it just sent the email, and I didn't look at it at all.

我当时就想，糟了。

And I was like, fuck.

于是我打开已发送邮件看了看，发现这简直跟我自己写的没两样。

And so I went to my sent and looked at it, and I was like, oh, this is exactly what I would have sent.

所以很多时候，它写出来的内容都挺接近的。

And so it's like, it's pretty close to to that a lot of the time.

有时候它语气太正式了，还有一些小问题，但你仔细想想，其实大部分邮件内容都是套路化的。

It can be like a little over formal and there's a couple of things that that it's just when you really think about it, most of your email is kinda it's not it's kind of rote.

就是挺平淡的。

It's kind of prosaic.

我当然希望由我自己来决定邮件该写什么内容，但具体怎么措辞，我通常并不太在意。

It's kind of I I definitely wanna be the one to think about what it should say, like, what what it should say, but the actual sentences don't matter that much to me, usually.

当然，有时候我也会花很多心思去改。

Sometimes I do a lot.

这话可是从一个写作者嘴里说出来的。

And this is coming from a writer.