氛围编程硬件

我评价一个工程师的标准是，你能不能打造出一个能产生倍增效应的系统，从B一直做到Z？

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?

根本不是十倍的问题。

It's not even 10 x.

是百倍，甚至是千倍的提升。

It's 100 x or thousand x.

一直都是这样。

It always has been.

Claude在某个领域里的表现，几乎和你一样好。

Claude or is basically as good as you are in a domain.

直接浪费点token，省下时间更重要。

I would say just waste tokens, save time.

别去纠结token是输入还是输出。

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

你只看花了多少时间，以及最终产出的结果。

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

不管这些模型看起来多贵，它们还是比人类便宜得多。

No matter how expensive these models might seem, they're still way cheaper than a human.

这些模型现在已经成熟了。

The models at some point graduated.

它们以前只是初级工程师。

They used to be junior engineers.

现在已经是资深工程师了。

Now they're principal engineers.

现在有了智能代理，你再也不会卡住了，这真的很了不起。

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

这完全是纯软件的活了。

It's pure software dead.

布莱克，你在Boom Supersonic是怎么应用这些技术的？

Blake, how are you applying all this stuff at Boom Supersonic?

是的。

Yeah.

我发现这彻底改变了软件和硬件开发人员的角色。

What I found is it completely changes the role of software and hardware developers.

从一开始，我们就试图把很多传统的工程流程——尤其是硬件工程流程——转化成软件。

The thing that we did from day one was try to take a lot of traditional engineering workflows, I mean hardware engineering workflows, and turn them into software.

如果你没接触过硬件工程，让我解释得更清楚一点。

If you haven't been around hardware engineering, let me just make this more clear.

很多硬件工程工作其实是在工程师的电脑上，用Excel表格孤立地完成的。

There there's a lot of engineering, hardware engineering that happens in Excel spreadsheets on engineers' laptops in a silo.

这些表格非常复杂，有时候还嵌入了VB脚本代码。

And it's very complex spreadsheets, sometimes like VB script code.

所有这些东西本质上都是软件，但人们却从来不当它是软件。

And all of this is actually software, but it's treated as if it's not software.

所以没有版本控制，也没有自动化测试。

So there's no source control, there's no automated testing.

如果你想把气动分析的结果交给结构工程师，还得靠手动发邮件传Excel，简直像上世纪90年代一样，太落后了。

If you wanna hand something off from like an aerodynamicist to a structures engineer that's done manually with like a spreadsheet over email, like it's the 1990s, It's terrible.

所以我们开始构建这些软件框架，用来自动化并标准化硬件工程流程，目的就是降低迭代成本。

So we started building these software frameworks that can automate and make repeatable hardware engineering flows with the idea we could reduce the cost of iteration.

但进展很慢，因为我们一直雇不起足够的软件工程师。

But it slow going because we never afford enough software engineers.

后来我们转向了一个完全不同的模式：软件工程师负责设计整体架构，因为他们懂系统、懂算法、懂模块划分。

And what we've gotten into is this mind blowingly different model where the software engineers actually create the architectures because they understand systems, they understand algorithms, they understand division of concerns.

而硬件工程师只需要用他们熟悉的工程知识来编写自己的模块代码。

And then the hardware engineers can Vibe code their pieces because what they know about hardware engineering.

结果就是，小团队的生产力发生了翻天覆地的提升。

And the result is just, like, mind blowingly different productivity for small teams.

我举个例子。

I'll give an example.

比如你在设计一个涡轮叶片，传统做法是，叶片在冷态时是一个形状，运行起来变热后会膨胀，所以你得同时设计它在冷态和热态下的气动和结构性能。

Like like, you're designing a turbine blade, like, classically, so a turbine blade starts like cold, but when it runs, it's hot, so it gets bigger.

因此，你必须让这个部件在冷态和热态两种形态下都能正常工作。

And so you have to design both the aerodynamics and the structural design of the thing to work with this cold shape and this hot shape.

所以你得在冷态和热态之间来回转换，还要同时处理结构和气动设计。

And so you have to convert between cold and hot, and you convert pretty structures and aerodynamics.

每分析一个叶片，就得花一个工程师一整天的时间。

And this takes like one engineer one day for one blade, for one piece of the analysis.

一台喷气发动机里有上千片叶片。

And there are like a thousand blades in a jet engine.

这么算下来，根本没法大规模推进。

And so you can't do much.

而我们现在真的靠软硬件团队合作，搞出了这套解决方案。

And we literally now with a combination of software and hardware people created the solution.

你可以随时调整叶片的形状。

You can change blade geometry.

结构和气动的分析结果都能实时看到。

You can see in real time the structures and aerodynamics results.

这样一来，两个工程师就能设计出整台喷气发动机，这变化简直太大了。

And so it allows two engineers to design an entire jet engine, which is just wildly different.

你提到过一点，就是软件工程师在为其他工程师开发工具和架构。

One of the things you mentioned is that you have software engineers creating the tools and architectures for the rest of the engineers.

在我看来，企业软件最大的变革就是，现在已经没有初创公司能做出能卖出去的硬件协作工具了。

That to me is the biggest the cataclysm of enterprise software is that there is no, like, startup that builds hardware collaboration tools that can sell you anything anymore.

因为内部开发时，你只需要随时编写出当下真正需要的程序。

Because in internally, you're just coding the right things that you need at any given time.

就连表格软件也快过时了。

Even spreadsheets are kinda cooked.

对吧？

Right?

以前表格软件能流行，是因为没人能开发定制化的软件。

Because the reason that spreadsheets were successful is that no one could build custom software.

所以最接近定制软件的东西，就是一堆VBA函数撑起来的表格。

So the thing that approximates custom software the most is a spreadsheet with a bunch of Vibe Script functions.

我自己已经基本从Excel转向用Python建模了，这样能做出更靠谱的仿真结果。

I personally have moved almost entirely from Excel to Python models where I can actually, like, get, like, believable simulations of things.

是的。

Yeah.

我的意思是，AI目前还没做到的一点，但我相信明年，大概2026年之前就能实现，那会非常非常令人兴奋——现在它能生成代码，但很快就能生成STEP文件和电路板布局了。

I mean, thing that that AI hasn't come to yet that I think it it will within the next year, like, probably within '26, that will be very, very exciting, is right now it can generate software, but soon it'll be able to generate step files and PCB layouts.

一旦AI能应用于机械和电气工程领域，那将是咱们还没见过的全新局面。

And when it comes for mechanical and electrical engineering, that will be a whole other thing that we haven't seen yet.

那真的会特别酷。

That'll be very, very cool.

对。

Yeah.

在硬件方面，我觉得这对那些小设备公司和零部件公司简直是福音，他们以前做不出好软件，只能凑合写些烂代码。

On the hardware side, I think it's really a boon for, like, all these little gadget companies and part companies that write really bad software because they can't make great software.

现在他们终于能做出够用的软件了。

And now they're gonna be able to make good enough software.

甚至可能根本不需要人去操作界面了。

Or it may not even software that are the human front end.

它可能会完全变成自主代理，直接通过语音就能访问和控制硬件。

It might just be completely agentic for an agent to access, and you just talk to it through voice and control hardware.

这就是为什么我认为中国特别重视开源模型的原因之一。

And this is why one of the reasons why, I think, for example, China is big into open source models.

对吧？

Right?

他们之所以全力投入，是因为在硬件方面有优势。

They're basically going all in on it because they have hardware superiority.

他们拥有非常复杂的供应链和元器件网络，所以他们说：只要能按需生成软件，

They have these very complex supply chains and component chains, And they're basically saying, hey.

我就不再比硅谷落后了。

If I can just generate software on demand, then I don't have this disadvantage anymore against Silicon Valley.

但这并不是他们推动开源的唯一原因。

So that's not the only reason why they're doing open source.

我觉得他们其实也落后了一截。

I think they're also behind.

他们在压缩模型。

They're distilling models.

他们在整合资源，同步进展。

They're catching you know, they're collaborating resources.

但我觉得中国政府一向支持那些能带动整个生态发展的项目，尤其是网络效应明显的领域。

But I think the Chinese government has a history of funding efforts that then sort of help their entire ecosystem along, especially in network effect businesses.

所以他们想把所有资源集中起来，赶超人工智能领域，用它来给硬件产品创造优势。

And so I think they wanna, like, pool all their resources, catch up on AI, and use it to give their hardware stuff an advantage.

讽刺的是，他们做开源，正是因为OpenAI不开放。

And ironically, they're doing all the open source stuff because OpenAI is not open.

Grok确实发布了模型，但我觉得他们大概落后了一两代。

You know, Grox publishes models, but I think they're a model or two behind.

谷歌也有一些本地模型，但没什么真正有竞争力的。

Google has some local models, but nothing really that competitive.

至于Anthropic，据我所知，还没有公开的进展。

Anthropic to my knowledge.

我甚至都不知道他们有什么开源模型。

I don't even know of any open source models from them.

所以现在所有的开源力量都来自中国。

So all the open source heft is coming from China.

这对我们这边的硬件创业者有帮助，但对中国那边的硬件厂商、工厂等等的帮助要大得多。

It helps all our hardware founders, but it helps their hardware founders and factories and so on that much more.

还有那些你从亚马逊买来，周末懒洋洋地捣鼓的小玩意儿配套软件，现在也进步得飞快。

But all the crappy little software that goes with all the little random knickknacks and thingamajigs that you buy off Amazon to tinker with a lazy Saturday afternoon, that software is getting a lot better very quickly.

我觉得大家都已经醒悟了，如果没有顶尖的代码生成模型，就谈不上自我进化。

I think everyone's had the wake up call that without great frontier coding models, you don't have self improvement.

那想象一下，整个中国如果没法产出前沿技术，会怎么样？

And so imagine China as a whole not having the ability to produce frontier everything, right?

这不只是软件的问题，就像布莱克说的，整个硬件链条里每一步都需要生成软件。

It's not just producing software is in any piece of this hardware pipeline, like Blake was saying, like, you need to generate software.

如果你在生成软件的能力上落后了，那你所有方面都会落后。

If you fall behind in your ability to generate software, you fall behind in the ability to generate everything.

我很好奇你们的看法，因为大家都爱聊中国模型。

One thing I'm curious about from you guys is, like because everyone loves to talk about Chinese models.

你们自己会用中国模型吗？

Like, do you use Chinese models?

你们认识谁在用中国模型吗？

Do you know anybody that uses Chinese models?

其实昨天我刚和人吵过一架，饭桌上有人坚持说，97%的情况直接用DeepSea就够了，毕竟便宜。

This is an argument I had yesterday, actually, which is one person at the table dinner was claiming that, you know, you'll just use DeepSea for 97% of things because it's so cheap.

真需要高智商的时候，就反复跑同一个问题，只有最复杂的任务才用OpenAI、Anthropic这些模型。

And if you need more intelligence, you'll just run it over and over again the same problem, and you'll only use the OpenAI, Anthropic, etcetera, models for the most advanced tasks.

我当时就有点犹豫，不太确定是不是这么回事。

And I was kinda like, I don't know.

我觉得智商这东西，越多越好。

I think intelligence is an unalloyed good.

永远都该追求更高的智能。

You always want more intelligence.

没错。

And Yeah.

这些模型出错的时候，你根本察觉不到。

When these models make a mistake, you don't know it.

而且它比真人实时服务便宜多了。

And it's always cheaper than a real person and real time.

所以你自然会选用最智能的模型，但这未必是好事，因为最终可能导致AI领域出现垄断或寡头局面。

So you'll just use the most intelligent model available, which isn't great news necessarily because it means that, you know, you're gonna end up creating a monopoly or oligopoly kind of situation in AI.

但我始终想要最聪明的程序员。

But, I always want the most intelligent programmer.

我始终想要最准确的答案。

I always want the most correct answer.

我始终想要最好的判断。

I always want the best judgment.

既然我要在资本、代码、人力和营销上投入这么多资源，那就必须每次都能做出正确的决定。

And given the amount of leverage that I'm gonna pour into it through capital and code and people and, you know, marketing, I wanna make the right decision every time.

而且很多时候，两个模型之间——比如说，我有一个模型，我知道它比另一个稍微聪明一点，它们都给了我答案。

And often when between two models let let's say, like, I have one model that I know is a little smarter than the next one, and they both give me answers.

但很多时候，我其实分不清哪个答案才是对的。

Often, I actually don't know which is the correct answer.

对吧？

Right?

所以如果我知道哪个模型更聪明一点，我就会选它的答案。

So if I know one model's a little smarter, I'm gonna go with that answer.

久而久之，我就不再问我觉得没那么聪明的那个模型了。

And eventually, I'm gonna stop asking the model that I think is less intelligent.

但我也不确定。

But I don't know.

你们有没有发现那些所谓的没那么聪明的模型有什么用处？

Have you guys found a use for the these, you know, so called less intelligent models?

我们确实用得上。

We see uses.

所以我们有AI网关的数据，基本上所有应用代理都会经过这里。

The so we have the AI gateways data that basically like every application agents that are goes through.

开源模型确实有使用，但最顶尖的还是那些前沿模型，占据绝对主导。

And so there's definitely usage of open models, but the top is like heavily dominated by the frontier intelligence.

不过这里有个细节，就是前沿模型如果能在合理成本和性能下规模化，那就太有优势了。

And there's a subcategory or there's like a caveat to that, which is that frontier intelligence at reasonable cost and performance, like slaps at scale.

大家对Gemini可能没那么兴奋，但有些模型在性能和成本的平衡上做得特别好，特别聪明。

So like people don't get really excited about Gemini, but they put out these models that are like super smart at the right performance cost combination.

有趣的是，在很多任务上，除了编程之外，它们反而是最好的模型。

And for a lot of tasks, other than coding actually, interestingly enough, they're the best models.

它们才是真正适合工业级生产环境的模型。

They're like the best, like, industrial production models.

你可以拿它们去处理客服任务，或者浏览器自动化这类工作。

You can throw them at, support tasks or browser automation.

我一般都会直接用Gemini模型来做这些事。

Like I would always put a Gemini model there.

至于这类任务，我会考虑用中国产的模型。

And I would look to Chinese models for those kinds of things.

但每当我需要突破前沿时，就必须用最顶尖的编程模型。

But anytime I'm working to push the frontier, you need the best possible coding model.

目前能做到这一点的，也就两三个模型而已。

And that's basically now like two or three models.

而中国厂商显然还没进入这个圈子。

And the Chinese are not certainly not in it.

嘿，马克斯。

Hey, Max.

你最近在大力推动垂直整合和极致紧迫感啊。

You're pushing pretty hard into vertical integration and extreme urgency.

要不要聊聊这个？

Do you wanna talk about that?

对。

Yeah.

很多东西，你根本买不到，只能自己想办法做出来。

Mean, for many things, we maybe you can't buy it, so you gotta make it somehow.

我们当然更倾向于直接采购。

Our preference would always be to buy something.

比如有供应商能以极低的价格提供某种服务，像电路板就是个例子。

Like, if there's a vendor that offers a service at a great price, it's like, for example, like PCBs.

我们从来不自己生产电路板。

Like, we don't make PCBs.

那些东西几乎不值钱。

Like, those are they're basically free.

你可以从亚洲随便买，要多少有多少。

You can buy them in unlimited quantity from Asia.

但我们的产品越接近一个整体的、原子紧密相连的实体，表现就越好。

But the the closer that our products get to being, a single block of covalently bonded matter, the better they'll be.

功耗更低，体积更小，性能更强，寿命更长。

Lower power, smaller, higher performance, last longer.

有些零部件根本买不到。

And there's just, like they're like, the components aren't available.

要想做到这种程度的整合，真正突破只能买现成零件的局限，真的非常受限。

And in order to do that type of integration, be able to actually innovate beyond things, just piecing together things that you can buy off the shelf, which really is is very very limiting.

你只能自己学着去做，这就体现为垂直整合。

I guess you have to, like, learn to do it yourself, and that shows up as vertical integration.

我们在东海岸拥有一家专属的MEMS代工厂，因为除了这样，我们实在找不到别的办法来实现想要的封装和组装。

So we own a captive MEMS foundry on the East Coast, which we thought there was really no other way to do the type of packaging and assembly stuff that we wanted to do.

我觉得未来几年，AI会极大地影响这一切。

And I think that all of this is gonna be affected heavily by AI over the next few years.

但现在还达不到那个程度。

It's not quite there yet.

讽刺的是，AI对我们公司和监管沟通影响最大的地方，反而体现在这些方面。

In fact, ironically, of the biggest impacts that we've seen of AI inside the companies and regulatory interactions.

比如，我们可以用AI生成文档，或者当我们想修改、升级产品时，能快速提出需求。

Because if we can do things like generate documentation or if we can ask if we wanna change, we wanna evolve this product.

有成千上万项ISO标准可能都适用。

Like, there's thousands of ISO standards that might apply.

我们到底得遵守哪些？

Which ones do we have to comply with?

还得把这条线给理清楚。

And, like, trace this through.

以前得让整个合规和质控团队跟几个月，一点一点地追踪，现在AI自己就懂了。

This used to be like you're like you're following a whole regulatory and quality team for several months as they trace this, and now the AI just kinda knows.

但一想到手术项目或者MEMS晶圆厂，我还是觉得，软件终究得靠人来动手。

But when I think about stuff like the the surgical program or the MEMS fab, I think ultimately the software still needs hands.

它可能会比我们聪明得多。

Like, it's gonna be smarter than us.

但如果它造不出东西，那才是真正不可逾越的界限。

But if it can't make things, then, like, those are real real boundaries.

所以我们已经把晶圆厂，还有公司里很多其他环节都做了数字化监控，等这些模型越来越强，很快就能在细胞工程和材料科学这些方面看到明显效果。

And so we've instrumented our foundry as well as many other parts of the company in in ways where as these models get better, that should show up pretty immediately in in things like the cell engineering that we're doing and the material science

我们正在开发的那些。

that we're that we're developing.

这让我意识到，已经很久没找律师起草过一份基础法律文件了。

It sorta makes me realize that, like, it's been a while since I've generated a basic legal document using a lawyer.

对吧？

Right?

我不再找律师弄保密协议、签这个签那个、做那些法律调研了，所有这些基础法律事务都没了。

I stopped asking lawyers for NDAs and, you know, agreement for this and sign that and research this and, like, all the basic legal tasks are gone too.

因为以前总有个笑话，说法律就像 spaghetti 代码。

Because, you know, there's the old joke that law is like spaghetti code.

他们把一套特别复杂的规则硬塞进英语里，结果这儿和那儿还互相矛盾。

You know, they have this very complicated code that they try to put in English, and it contradicts this code over here.

还得跟别的条款对上，可根本就没有清晰的接口。

It has to fit into that code over here, and there are no real APIs for it.

对初级工程师来说，他们其实直接升职成了高级工程师，而初级工程师的工作已经被AI代理接手了。

But for just like junior engineers and junior engineering, I should say, junior engineers basically got a promotion to senior engineers, and junior engineering got taken over by agents.

所以我觉得，法律行业也是一样的。从某个角度看，坏处是你可以说 paralegals（法律助理）被解雇了；但换个角度说，他们其实是升职成了资深律师，现在有更多时间去思考法律问题了。

And so the same way, I think, in a way, the downside is you can look at law and say, you know, paralegals just got fired, or you could say paralegals just got promoted to senior lawyers, and now they can spend their time thinking about the law.

其实想想软件工程和律师行业演变的相似之处还挺有意思的。因为律师到底在文件里写了什么，你根本无从得知。

It's actually kind of interesting to think about the parallels of how software engineering is evolving with lawyers because lawyers, you never know what they put into these documents exactly.

你只能信任他们。

You just trust them.

比如，嘿，律师，你能帮我看一下这份文件吗？

Like, Hey, lawyer, can you look at this document?

你能帮我看看这靠不靠谱吗？

Can you tell me if it's legit?

你能帮忙做修订标记吗？

Can you do red lines?

随便啦。

Whatever.

说到底，你跟律师打交道时最看重的，就是他们作为值得信赖的专业人士的身份。

Like, at the end of the day, what you're valuing in the relationship with a lawyer is that they're a trusted authority.

他们上了法学院，把自己的声誉押了上去。

They went to law school, and they're putting their reputation on the line.

我觉得这跟软件工程现在最大的问题很像——一堆乱七八糟的代码最后成了一个合并请求。

I think there's a parallel parallel with, like, the biggest problem in software engineering today is these mountains of slob that end up as a PR.

然后大家在推特上到处刷各种段子。

And then people are saying there's all these memes on Twitter.

以前啊，每个人都会逐行审阅合并请求里的代码。

Like, way back in the day, used to read every line of code of a PR.

但在我的领域，基础设施这边，我希望工程师能说：我理解了。

Well, in my world, infrastructure, I want engineers to be able to say, I understand.

这不意味着你真的读了合并请求里的每一行代码。

It doesn't necessarily mean that you've read every line of the PR.

你得能自信地说：我认可了这个合并请求可能带来的后果，或者我写了测试框架、模拟器、证明逻辑、类型检查器这些工具，所以即使没看代码，我也能确信它上线是安全的。

You need to be able to say, I am signing off on understanding the consequences of this PR, or I wrote the test harness, the simulations, proofs, the type checkers, etcetera, to be able to say, even without reading this, I have confidence I can sign off on it's gonna be safe in production.

所以这挺有意思的，因为确实存在一种情况，我们接受一切都会变成乱麻代码，而且我们根本不可能完全搞懂它。

And so it's it's kinda interesting because there's a world in which we embrace that everything is gonna be a spaghetti code and that we don't fully understand it.

但我们编写了这些评估工具，让我们有信心。

But we write the basically evaluators that give us confidence.

然后我们依赖基础设施和生产工程师说：好吧，我觉得这个可以上线了。

And then we rely on, like, people, like the infrastructure production engineers to say, okay, I'm fine sending this into prod.

说到底，要是系统出问题，总有人得被叫醒处理。

You know, at the end of the day, like, someone's gonna get paged if your systems go down.

我觉得很多人低估了一点：写软件其实很容易，从零到一很简单。

And I think another thing that people are underestimating is that creating software is really easy, zero to one.

但你想想一千天之后呢？

But think about a thousand days from now.

你的软件那时候会是什么样子？

What does what does your software look like?

它安全吗？

Is it secure?

它经过充分测试了吗？

Is it tested?

它达到生产环境的标准了吗？

Is it production grade?

是的。

Yeah.

它的性能怎么样？

Is it performant?

你还有动力持续投入这么多资源来维护它在生产环境中的运行吗？

And are you still motivated to invest all of those tokens in maintaining it in prod?

人类正逐渐变成验证者。

I mean, humans are becoming verifiers.

对吧？

Right?

我们正是用高质量的验证数据来训练这些模型的，而现在我们正需要人类来担任验证者。

And and that's kinda how we train these models with good verification data, and now we need human verifiers.

所以，我觉得过去很多由人承担的职责——律师、工程师、运维人员——现在都转向了验证整个系统，并确认：没错。

So, yeah, I think a lot of the a lot of the old function of people, lawyers, engineers, operations people moved to verifying the stack and saying, yeah.

这个大致没错，我愿意为它背书，出了问题我也扛着。

This is roughly correct, and I I'll roughly stand behind it, and I'll support you if it goes wrong.