浪费代币，节省时间

欢迎收听。

Welcome.

你现在收听的是《海军》播客，你获取前沿知识的权威来源。

You're listening to the Naval podcast, your authoritative source for new knowledge.

今天我们要尝试一些新的东西。

We're trying something new today.

今天我请来了三位前沿领域的创始人，其实都是挺帅的小伙子，还有第四位帅小伙——纳瓦尔。

I have three frontier founders with us, three good looking guys actually, and a fourth good looking guy, Naval.

让我先为大家介绍一下。

And let me just introduce everybody.

吉列尔莫·劳赫，他正在把 Vercel 打造成面向智能代理及未来技术的AI云平台。

Guillermo, the g Rauch, he's building Vercel into an AI cloud for the world of agents and whatever comes after that.

很高兴来到这里。

Good to be here.

布莱克·肖尔正在自己的工厂里研发超音速飞机，连喷气发动机也一起做。

Blake Scholl, he's building supersonic aircraft in his own factory and jet engines as well.

布莱克的公司叫Boom Supersonic。

Blake's company, Boom Supersonic.

还有来自Science的马克斯·霍达克，他正在研发一种生物混合脑接口，让活体神经元在硅片上生长，用来恢复视力等感官功能，未来甚至探索大脑的新区域和全新的感知方式。

And then Max Hodak from Science, he's building a biohybrid brain interface that grows living neurons on silicon to restore sensory functions like sight, but then eventually to explore new parts of the brain and new senses.

这三位创始人都不是在用现成的零件拼凑产品。

All three of these guys are not composing their products with off the shelf parts.

他们都在自建工厂。

They're building their own factories.

我们更关心的不是他们具体在造什么，而是他们在制造过程中学到了什么新东西。

And you know, don't care as much about what they're building exactly as we do about what they're learning about how they're building.

他们正在产生哪些新的认知？

What's the new knowledge they're generating?

他们的核心突破是什么？

What's their alpha?

他们发现了哪些原则，能给其他创业者带来启发？

What principles are they discovering that other founders can learn from?

他们现在正在琢磨什么？

Are they trying to figure out right now?

还有，有哪些前沿或疯狂的想法，他们还没说出来，还在脑子里慢慢成形？

And also, what are the cutting edge or crazy ideas that they haven't even talked about yet and they're still forming in their brains?

纳瓦尔，你对这些有什么看法吗？在聊吉列莫之前，我想听听你的想法。

Naval, do you have any reactions to any of that before I jump into Guillermo?

对。

Yeah.

咱们就图个开心吧。

Let's just have fun.

对。

Yeah.

你们俩别客气，直接说吧。

You guys should just jump in.

对。

Yeah.

我记不清原话了，但我一直被‘软件工厂’这个想法深深影响——工程师的工作就是按时到岗，直接交付成果，公司里衡量一个人的标准就是：他交付B成果的能力有多强？

So I can't remember my exact quote, by the way, but I've been really pilled with this idea of software factories and the job of the engineer being something that you just show up to work, used to used to ship the output directly, and everything inside the company was, you know, how good is person a at shipping output b?

但现在，我评判一个工程师的标准变了：你能不能打造出一个能批量产出B到Z成果的工厂？

And now what's happening is the way that I'm judging you as an engineer is like, are you producing the factory that will produce multiplicative outputs b through z?

对吧？

Right?

这其实是个巨大的转变，因为过去我们一直相信——这个观点其实有点争议——存在十倍效率的工程师。

And that's a that's a pretty significant change because, basically, like, we used to believe and it should be somewhat controversial that there's 10 x engineers.

而现在，显然已经出现了百倍甚至千倍效率的工程师，但整个行业还没跟上这个现实。

Like, now, clearly, there's a 100 x or a thousand x engineers, and the world hasn't fully adjusted to this.

我以前在推特上说‘有十倍效率的工程师’，结果被喷得很惨。

I used to get flamed on Twitter for saying there are 10 x engineers Yeah.

因为这跟现在流行的平等理念相冲突，大家都觉得人人平等。

Because it likely flies in the face of so much, like, equality philosophy that everyone's equal.

但事实是，当你在思想领域、在智力和数字领域工作时，差距根本不是十倍那么简单。

But the reality is when you're operating in idea domains, when you're operating intellectual domains and virtual digital domains, it's not even 10 x.

是百倍甚至千倍的差距。

It's 100 x or thousand x.

一直都是这样。

I mean, always has been.

像中本聪、Notch、JavaScript的发明者布兰登·艾克，还有约翰·卡马克这些人。

Satoshi, Notch, you know, the guy who invented JavaScript, the Brendan Ikes of the world, John Carmack.

这些人都是千倍效率的程序员。

I mean, these are thousand x programmers.

更别说选对了方向和选错了方向，那简直是天壤之别。

Not to even mention if you choose the right thing to work on versus the wrong thing to work on, that's an infinity difference.

不一定是因为他们编程技术更厉害，而是他们一开始就能判断出该做什么。

And it could just be not necessarily a better programmer, just one who had a better judgment on what to work on in the first place.

现在当然没那么有争议了，毕竟有AI加持。

And now, obviously, it's less controversial because of AI leverage.

真正有争议的是那种代币排行榜。

What's controversial is that the token leaderboards.

对吧？

Right?

人们还是有点糊涂，现在觉得我请了一堆能产出百倍价值的工程师。

Like, people are still getting a little confused because now they think, well, I have a bunch of 100 x engineers.

看看我付的这些代币费用。

Look at all these tokens that I'm paying for.

我想知道你们有没有遇到过类似的情况。

I'm curious if you guys have seen the same.

你们是怎么衡量投资回报的？

Like, how do you measure ROI?

这就跟过去用代码行数来衡量一样。

It's like the old measuring lines of code.

代币消耗量跟代码行数一样，都不是直接有效的衡量标准。

You know, token consumption of lines of code feel like similarly not direct paradigms.

我的观察是，Claude、ChatGPT 或 GPT 在某个领域里的能力，基本就跟你自己一样。

I mean, my observation has been that Claude or chat GPT or GPT is about is basically as good as you are in a domain.

如果你是个能力很强的开发者，这些工具就会特别有用。

And so if you're if you're a really capable developer, then these things are really powerful.

但如果你是个初级开发者，那你用起来可能也就跟个初级开发者差不多。

And if you're a junior developer, then you'll kind of find it to be, like, more of a junior developer.

一方面，这些模型的能力确实非常强。

Like, on the one hand, these models are incredibly capable.

但另一方面，你给它们的反馈虽然断断续续，却极其关键，那些小小的调整往往直接决定了最终的效果。

On the other hand, the feedback that you give them sporadically seems to be incredibly important, and these little updates seem to totally determine the types of performance you get out of them.

我现在多了一种新的支持方式：有人来找我，说模型输出不好，我就教他该怎么改提示词。

There's a new kind of support that I give, which is you come to me and, like, you didn't get good output out of the model, and I tell you what to prompt the model with.

所以，提示词优化的质量——我觉得你也在说这个——真的特别重要。

So, like, the idea of, like, the quality of the reprompting, which I think you're alluding to, is is extremely important.

但，我

But, I

说清楚一点，我觉得随着时间推移，这一点会越来越不重要。

mean and to be clear, I think that this will become less important over time.

比如，随着模型变得越来越聪明，你

Like, as the models get much, much smarter, then you'll

就能

be able

投入更少，收获更多。

to put in less and get more out.

但至少在现阶段，根据我的经验，它确实会反映出用户自身的判断力。

But at least at this stage, it really seems to kind of reflect back the judgment that the user brings in in my experience.

我其实一直不太想去学那些花里胡哨的技巧

I've kind of resisted learning all the ticks and the tricks

还有各种窍门。

and tips.

比如，你知道的，有人说要用拉尔夫·威格姆。

Like, you know, there was a, oh, use Ralph Wiggum.

还有用 Open Claw。

Use Open Claw.

用Hermes。

Use Hermes.

用这个提示引擎。

Use this prompt engine.

用这个框架。

Use this scaffolding.

把这一部分接上去。

Plug in this piece.

你知道的。

You know?

always use plan mode。

Always use plan mode.

我根本没管这些。

I just ignored all of that.

我就觉得，模型进步的速度会比我学会怎么用它快得多。

I just assumed the model is just gonna get better faster than I would figure out how to use it.

它学会怎么利用我，比我学会怎么用它还快。

It would figure out how to use me faster than I would figure out to use it.

所以我对这些模型一直很粗暴，经常气得不行，结果随着时间推移，我越来越懒得输入、懒得干活，反正觉得靠硬堆就能搞定。

And so I've just been completely hamfisted with them, and I get frustrated at them and just sort of I I found myself typing less and less information and doing less and less work as time goes on with the models because I just assume I can brute force my way through it.

我会把Codex、Claude和Gemini反复扔到同一个问题上，浪费大量token，就为了省时间。

And I'll throw Codex, Claude, and Gemini at the same problem over and over and just waste tokens to save time.

我觉得不管这些模型看起来多贵，还是比请一个人便宜多了。

And I think no matter how expensive these models might seem, they're still way cheaper than a human.

所以我的建议就是，别心疼token，省时间最重要。

So I would say just waste tokens, save time.

别去管什么输入输出的token数量。

Don't look at the tokens either as inputs or outputs.

你只看自己的时间花了多少，还有最终结果怎么样。

Just look at your time and look at the final output.

就算它们写出来的代码质量不高——我知道很多时候确实如此，也不是能直接上线或能扩展的代码。

And even if they're writing low quality code, which I know in many cases they are, it's not necessarily production quality or scalable code.

等到真要上线的时候，我就再多扔点token进去。

When the time comes and I wanna ship it to production, I'll just throw more tokens at it.

我会说，好吧。

I'll say, okay.

现在你去仔细看看，重写一遍，每一代都会越来越好了。

Now go through, look at it, rewrite it, and they're just gonna get better every generation.

所以，说实话，我觉得这事儿根本停不下来。

So, yeah, I don't I don't see where this necessarily stops.

只要我们有明确的领域，能解决问题，它们就能把这些问题搞定。

As long as we have verifiable domains and solve problems, they're gonna resolve those problems.

真正需要小心协作、深度配合模型的，是那些还没解决的难题，比如你得当个顶尖网球选手，或者站在创意的最前沿。

That's in the unsolved problems domain where maybe you're a tennis tie or you're the cutting edge of creativity that you need to be, you know, working very collaboratively and carefully and closely with the model.

但我没到那个层次，我的软件工程还没到那一步。

But I'm not in that I'm not at that level of software engineering.

至于你，吉列尔莫，你大概是团队里最极端的软件工程师了。

And then, Guillermo, you're probably the most extreme software engineer in the team.

对吧？

Right?

在这群人里，你大概是唯一一个真正从软件背景一路硬核打拼上来的。

Like, out of this set, you're probably the one who most hardcore came up from a software background.

你觉得这些模型在接近极限时表现怎么样？

Like, how are you finding these models at the edge of their capability?

最近发生了一件事，跟你刚才说的特别吻合：以前你给模型一个提示，它就按经典方式做下一个词预测，然后自己跑偏了。

Well, there's one thing that's happened recently that what you're saying resonates strongly with, which is it used to be that you would give a prompt to the model and it kind of does it like classic, like, next token prediction thing, and it, like, runs away with your idea.

但现在模型会自动进入一种直觉性的规划模式，就像你说的，根本不需要你主动去规划，它自己会回来告诉你：你看。

And models now have been doing this, like, intuitive planning mode without, to your point, not even having to plan, where it it comes back to you and says, look.

你让我做的这件事，其实有三条路可以走。

What you're asking me for, there's these three routes we can take.

每条路都有不同的权衡，我们会选其中一条。

There's this set of trade offs that we're gonna go down.

这时候，就真的像大家常说的那样，整个流程都交给它了。

That's a moment where, like, you know, people do the whole thing on x.

就好像突然间，我们有了一个博士级别的工程师模型。

It's like, oh, now we have a PhD level engineer model.

很明显，这些模型在某个时刻已经毕业了。

Like, that's very clear that the models at some point graduated.

以前它们还只是初级工程师。

They used to be junior engineers.

现在它们成了资深工程师，因为会主动回来跟你讨论各种权衡方案。

Now they're principal engineers because they come back to you with a set of trade offs.

当然，有时候它们也会瞎扯，那可太好笑了。

And, obviously, sometimes they bullshit, which is hilarious.

它跟你说，这个方案要三周，那个方案完全能搞定——结果却做出一堆糟糕的预测。

It tells you, this one is gonna take three weeks, and this one can totally it tries to make really bad predictions.

但很明显，我现在已经把它们当成平等的伙伴，真正和它们进行智力上的来回碰撞了。

But, clearly, it's now this like, I respect the models a lot more as a as a peer, like, that I'm going back and forth intellectually with.

不过，它们还是有很多不足之处。

But there there are a lot of gaps still.

所以，如果你是个特别厉害的工程师或者架构师，我觉得你还是能挖出更多价值。

So, like, if you're a really, really proficient engineer or architect, you I think you're still extracting more juice.

马克斯之前提过一个问题：如果你是个初级工程师，模型会不会也给你个初级回复？

So the question sort of that Max was positing of, like, if you're junior, do you get junior back?

显然不会，因为初级工程师能学到自己根本写不出来的更高级的代码。

Well, clearly not because a junior gets more advanced knowledge in code that they would have never been able to write by themselves.

但资深架构师能获得十倍提升，而初级工程师可能只有两倍，对吧？

But doesn't an experienced architect get 10 x, whereas a junior engineer gets two x?

这个我还在琢磨清楚。

That's what I'm kinda trying to figure out still.

是的。

Yeah.

对。

Yeah.

但我觉得，这中间涉及到一些架构上的决策。

But, I mean, I think there's there's architectural decisions.

我现在在观察我们一些初级软件工程师的发展，他们在职业进阶上下一步该做什么呢？

So when you think about the development I'm seeing this now with some of our the junior software engineers and the team of, like, what is the next step in their career progression?

从只是实现某个功能，转向自己做技术选型，比如选 PostgreSQL 还是其他数据库，或者选 ZMQ 还是别的消息队列系统。

It's going from, like, writing implementation for a feature to picking technologies, like choosing between Postgres versus some other database or picking between ZMQ versus some other message queue or, like, some other queuing system.

这些模型倒是能提建议，但真正关键的是，你一看就知道不对劲。

And those I mean, the models can suggest them, but that's the thing where you'll see it, and you'll be like, no.

不对。

No.

不对。

No.

想用别的东西？这种细微的反馈才真正重要，也是现在你能看到的最有价值的输出。

Wanna use this other That's the type of little feedback that I'm saying really matters and the types of output that you seem to get at this point.

品味和判断力。

Taste taste and judgment.

对。

Yeah.

对吧？

Right?

品味和判断力。

Taste and judgment.

你可以问他们：我该选哪个？为什么？

That said, you can ask them, which one should I use and why?

他们什么都知道。

And they know everything.

他们会给你分析得很透彻的权衡。

They'll give you really good trade offs.

这就是最近发生的变化，之前你会说：

That's the change that was saying has happened recently where you would say, hey.

去把这种超高基数的遥测数据放进 Postgres 里。

Go and put this super high cardinality telemetry data into Postgres.

结果他们却说：不行。

And it's like, no.

不行。

No.

绝对不行。

No.

老兄，我们不会把这种数据放进 Postgres 里。

Bro, we don't put that kind of data into Postgres.

你应该考虑用 ClickHouse、Athena 之类的工具。

You should consider Clickhouse or Athena or whatever.

这种事我遇到过好多次，真的挺让人惊讶的。

Like, that's happened to me a lot, which is really impressive.

但我现在还是有点纠结，毕竟人类依然在完成模型的工作。

But I the the thing I'm still, like, kinda struggling with is clearly the human is still completing the model.

也就是说，会不会反过来呢？

Like, at one point, is it the other way about?

人类反而在接收指令，比如：去帮我拿一下这个 API 密钥，因为只有你能做到；或者帮我争取下一笔投资资金，你看着办就行。

Like, the the human is the one sort of getting the instructions back on, like, go get me this API key because it's something that only you can do, or get me this amount of capital for my next set of investments that I need to make, you just watch.

很明显，我们离那一步还差得远。

Like, that clearly, we're still not there yet.

这只是一个暂时的异常。

That's a temporary aberration.

用不了多久，每个优秀的SaaS公司或者托管服务商都会提供命令行和API接口，让模型可以直接调用。

Pretty soon, every good SaaS company or hosting provider will have a CLI and API interface that the models can be made directly.

其实连API都不一定非得有。

They don't even necessarily need an API.

只要它是基于文本、符合Unix风格的，代理就能自己搞出一套接口来。

Like, as long as it's, like, text based, Unix based, the agent can hack its own API.

至于花钱的部分，你只要接入加密货币代币，比如比特币或者其他，模型就能自己去支付它需要的开销。

And then the money part, you insert crypto tokens, you know, put in Bitcoin, put in whatever, and the model goes and just pays for whatever it needs.

我觉得啊

And I think, like

对。

Yeah.

你知道，已经有人在做这方面的研究了。

You know, there are people working on this.

但我现在一直在想，纯软件是不是已经过时了？

But the thing I am now thinking through is is pure software dead?

也就是说，纯软件工程是不是成了一个被淘汰的行当？

Like, is pure software engineering, like, an obsolete thing?

这就像是说英语一样。

It's like saying speaking English.

对吧？

Right?

现在的模型已经能说英语了。

The models now speak English.

以前我们得学编程才能跟模型沟通。

We had to learn code to communicate with the models.

现在模型自己就能说英语了。

Now the models speak English.

而且它们说的英语像人一样模糊又随意，却能真正理解东西。

And they speak fuzzy, sloppy English like a human, and they understand things.

那对于创业者来说，护城河到底在哪呢？

So where's the moat, like, for a founder?

硬件反而成了优势。

Hardware, it's a boon.

你知道吧？

You know?

比如现在，如果要搞硬件的话，

Like, now if Yeah.

以前你要是得自己做硬件，再同时做软件公司，那可太难了。

You had to build hardware, it was hard to build software company alongside.

就像帕特里克·科利森说的，软件是门艺术，找艺术家可不容易。

Like Patrick Collison says, software is art, and it's hard to hire artists.

所以现在作为硬件领域的创业者，那就太好了。

So now as a hardware founder, great.

现在可以很快地开发出非常出色的软件。

You can have really good software developed fairly quickly.

如果你在做模型，那可能就是新的软件工程了——训练模型、调整模型、训练后的优化和微调。

If you're creating models, maybe that's the new software engineering, training models and tweaking models and post training and fine tuning models.

但传统的软件工程已经过时了。

But classic software engineering is that dead.

纯软件项目还值得投资吗？

Is pure software investible?

纯软件还能不能组建公司、组建团队，然后找到突破口？

Is pure software something you can organize a company, a team around, and try to get some leverage?

你们有没有看到Mitchell Hashimoto在X上发的一篇文章，叫什么‘区块经济’或者‘积木式经济’之类的？

Did you guys see the there is an article on X by Mitchell Hashimoto called the block economy or the building block economy, something like that.

他的观点是，目前对智能体来说最有用的，是强大且可复用的模块组件。

Like, his argument is that the most useful thing for agents to have now is really powerful reusable building blocks.

就像Max举的例子，你不会指望你的机器人每次发邮件都要重新发明一套队列系统吧。

Because to Max's example, you wouldn't expect your clanker to reinvent a queue infrastructure system every time it needs to send an email.

它需要调用一个正好匹配你任务需求的构建模块，然后说，好吧，这个任务就用 BOLMQ。

It needs to bring in the right building block that's right sized for the task that you're asking for, and then say, well, okay, for this one, it's BOLMQ.

我不同意那种认为智能体应该从零开始、完全重新发明一切，而且还要跟整个社会和文明脱节的想法。

I challenge the notion that I would want the agent to reinvent the entire universe and first principles in a way that's incompatible with the rest of society and civilization.

这简直就像为了你自己，重新设计高速公路、法律和政策一样。

Like, it's almost like reinventing highways, laws, policies, etcetera, just for you.

即使你从中能获得一些额外的优化和性能提升，但大规模协作的价值依然不可忽视——比如我们都依赖着 Postgres 13.2。

Even if there is a potential for extra optimization and extra juice that you can get out of it, there's still a sort of, like, cooperation at large scale value of saying, we're both depending on Postgres 13.2.

这种协作带来的价值，真的非常、非常、非常大。

And so that's still really, really, really valuable.

我认为，智能体将来会用到的那些基础设施和构建模块，显然属于构建这个世界的核心部分。

I would say, like, the category of infrastructure software and building blocks that these agents are gonna use, obviously, and bias is this world building.

这显然极具价值，而且我并不认为智能体很快就能独立完成这一切。

It seems extremely valuable, and I I don't see the agent anytime soon.

顺便说一句，我常打的一个比方是：任何已经被创造出来、模型可以直接复用的东西，都像是一个缓存的令牌。

And by the way, you could even another metaphor that I've been using is, like, any of it's already been created that the models can reuse is like a token cache.

因为你没必要花上万亿个标记去重复那些已经存在的东西。

Because you don't you don't want to churn through a trillion tokens to reproduce what's already existing.

所以模型总得有个起点来展开，但这个起点会彻底改变很多事情。

And so there's always a starting point that the model can fork off from, but it's gonna change things quite profoundly.

所以这些就是模型用的库和依赖项吗？

So these are, like, libraries and dependencies, for models?

是的。

Yes.

特别是针对智能体而言。

For agents, specifically.

不过说到 nav.al 的问题，我小时候刚开始学编程的时候就入了迷。

To nav.al's question, though, I mean, I learned a program when I was really little.

从青少年到二十多岁，我一直被它吸引，一写代码就是二十个小时。

And I like, that was the thing that through all of, like, being a teenager and in my twenties, like, I get, like, sucked into it and just, like, code for, like, twenty hours.

那真的特别有趣，我对各种编程语言都了如指掌。

And it was super fun, and I knew all this about programming languages.

我已经很久没写过一行代码了。

I haven't written a single line of code in quite a while now.

这 partly 是因为我的工作内容变了。

And, I mean, partly, that's because my job is different.

但自从十二月以来，我已经开发了很多软件，现在每天都用。

But also, since December, I've built a huge amount of software that I now use every day.

那些我几年前就一直幻想的项目，现在都成了我手里的实际工具。

There's all these projects that I've kind of fantasized about for years that now I'm, like, using, that I've actually built.

但我根本没亲手写过一行代码。

And I didn't write any of that.

我实在想象不回去手动写代码的日子了。

And I just can't imagine going back to, like, actually writing code by hand anytime.

就算不考虑别的，我也觉得写代码不太可能成为我未来的常态。

Like, I mean, I I'm unlikely to do that anyway, but just, like, in general, I see that I have a hard time seeing that as part of the future.

是的。

Yeah.

特别酷的一点是，你能看懂各个部分是怎么配合在一起的。

There's something really cool is that you understand how the pieces click together.

比如，只要理解什么是 API，数据是怎么流动的，输入输出和性能，你就得围绕这些设定预期——这个操作到底该达到什么水平。

Like, I feel like anyone that understands what an API is and how data flows, inputs and outputs, performance, because you kind of you have to orient the model around, like, this is a certain level of expectation that I have out of this operation.

这比亲手写代码有用多了，一直都是这样。

Like, that's that's always been infinitely more useful than than writing code.

我觉得真正厉害的工程领导者，其实是在 Slack 或者一对一交流里，用‘氛围编程’带动团队，传递你的意图、经验和方向，然后让别人去执行。

Like, I feel like a really good a proficient engineering leader has been, quote, unquote, like, vibe coding through people on Slack or one on ones, because you're transmitting your will, your intent, your experience, and you're letting others run with it.

现在只不过是我们用智能代理来做同样的事。

It's just that now we do the same but with agents.

所以我觉得你之所以能成功，原因就在这，但我不确定别人能不能有同样的效果。

And so I think that's why you've been successful with it, but I don't know that everyone sees the same level of success.

我二十年没写过代码了，但现在全靠智能代理，天天都在写，还做了不少软件。

I mean, I went from not having written code in twenty years to I'm coding all the time now, but through agents, and I'm building tons of software.

结果发现，只要懂点软件工程和算法的基本原理，就能走得很远。

And it turns out that just understanding the basic principles of software engineering and algorithms actually gets you a long ways.

因为我不再写代码的原因，是我没时间再去搞懂那些最新的编程语言和架构基础设施该怎么对接。

Because the reason I stopped coding was because I didn't have time to figure out the latest language, latest architecture infrastructure pieces to plug into.

我知道 Vercel 确实让事情简单了不少，但即便如此，刚开始上手还是很费劲。

And I know Vercel makes it a lot easier, but even then, just getting started was a bear.

就像把各个模块拼在一起，搭建基础设施，真的让人头疼。

Like, just plugging pieces together, assembling infrastructure was just so annoying.

真正带来改变的是，以前你确实能做出不少东西，很多环节都很直接，但总会突然碰到一些意想不到的麻烦。

The thing that really changed is I mean, it used to be that you could build a lot, like, you like, there's a lot that was straightforward, but then you would hit some random thing.

然后你可以花……对。

And then you could spend Yeah.

在一段不确定的时间里，去调试某个很具体的小问题。

Kind of some indefinite period of time debugging some narrow thing.

而现在有了智能体，你不会再卡住了，这真的很棒。

And now with the agents, what happens is you just don't get stuck anymore, which is pretty amazing.

或者是它们（智能体）卡住了。

Or they get stuck.

已经消失了。

It's removed.

嗯，也不是。

Well, no.

我的意思是，他们能很快找到正确的做法。

I mean, like, relatively quickly, they can find, like, the right way to do things.

我记得以前朋友学编程的时候，都会说：算了。

And it used to be that, like, I remember when their friends learned to program, they'd be like, nope.

这事儿本身就特别让人挫败。

It's just, like, intrinsically frustrating.

以前觉得，这部分就是学习过程的一部分。

Like, if, like, that part of the deal, that's how you learn.

但现在完全不是这样了。

And that just isn't true anymore.