Vibe Coding Hardware

频道: Naval
视频: https://www.youtube.com/watch?v=8R_9W9y8Y78
原文语言: en
统计: 共 15 轮 · 嘉宾 8 · Naval 3 · Blake Scholl 2 · Max Hodak 2

[0:00] 嘉宾

The way that I'm judging you as an engineer, it's like are you producing the factory that will produce multiplicative outputs B through Z? It's not even 10x, it's 100x or thousandx and it always has been.

我评判你作为一个工程师的方式，更像是看：你有没有打造出那座能产出 B 到 Z 这些成倍输出的工厂？这甚至都不止 10 倍，而是 100 倍、1000 倍，而且一向如此。

[0:10] 嘉宾

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

在一个领域里，Claude 或者 ChatGPT 基本上跟你做得一样好。

[0:15] 嘉宾

I would say just waste tokens, save time. Don't look at the tokens either as inputs or outputs. Just look at your time and look at the final output. No matter how expensive these models might seem, they're still way cheaper than a human.

我会说，尽管浪费 token，省时间就行。别去盯着 token 看，不管是输入还是输出。你只看自己的时间，看最终的产出。这些模型再贵，也还是比雇个人便宜太多了。

[0:26] 嘉宾

The models at some point graduated. They used to be junior engineers. Now they're principal engineers

这些模型在某个节点上算是毕业了。它们以前是 junior engineer，现在已经是 principal engineer 了。

[0:30] 嘉宾

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

而现在有了 agent，你基本上再也不会卡住了，这点真的挺神奇的。

[0:34] Naval

Is pure software dead?

纯软件已经死了吗？

[0:36] Naval

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

Blake，你在 Boom Supersonic 是怎么把这些东西用起来的？

[0:40] Blake Scholl

Yeah, what I found is it completely changes the role of software and hardware developers. The thing that we did from day one was uh try to take a lot of traditional engineering workflows and I mean hardware engineering workflows and turn them into software. And so for if you haven't been around hardware engineering let me see if I can make this more clear there uh there's a lot of engineering hardware engineering that happens in Excel spreadsheets on engineers laptops in a silo and you're very complex uh spreadsheets sometimes like VBScript code and all of this is actually software but it's it's treated as if it's not software there's no there's no source control there's no automated testing if you want to 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. And so we we started building these kind of like software frameworks. They can automate and make repeatable hardware engineering flows. The idea we could reduce the cost of iteration. Um but it was it was slowgoing because we could never get enough we could never like afford enough software engineers. And what we've gotten into is this uh mind-blowingly different model where the software engineers actually create the architectures because they understand systems, they understand the algorithms, they they understand, you know, division of concerns. Uh and then the hardware engineers can vibe code their pieces because what they know about hardware engineering and the result is just like mindblowingly different productivity for small teams. I give an example like like if 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 and so you have to design both the aerodynamics and the structural design of the thing to work with it cold shape and this hot shape and so you have to convert between cold and hot and you convert between structures and aerodynamics and this takes like one engineer one day for one blade for one piece of the analysis and there are like a thousand blades in a jet engine and and so you can't do much and we literally now with a combination of software and hardware people created the solution you can change blade geometry you can see in real time the structures and aerodynamics results and so it allows two engineers to design an entire jet engine which is just wildly different. One of the things you mentioned is that you have software engineers creating the tools and architectures for the rest of the engineers. That to me is the biggest u the cataclysm of enterprise software is that there is no like startup that builds hardware collaboration tools that can sell you anything anymore because in internally you're just coding the right things that you need at any given time. Even spreadsheets are kind of cooked, right? Because the reason spreadsheets were successful is that no one could build custom software. So the thing that approximates custom software the most is a spreadsheet with a bunch of EV script functions. I personally have moved almost entirely from uh Excel to Python models where I can actually like get like believable simulations of things.

是这样，我发现这件事彻底改变了软件和硬件工程师的角色。我们从第一天起做的事，就是想把大量传统的工程工作流——我是说硬件工程的工作流——变成软件。如果你没接触过硬件工程，我试着讲清楚一点：很多硬件工程其实是在工程师笔记本上的 Excel 表格里完成的，各自为政、孤岛作业，那些表格非常复杂，有时甚至写着 VBScript 代码。这其实全都是软件，但大家却不当它是软件——没有版本控制，没有自动化测试。如果你想把东西从气动工程师手里交接给结构工程师，那都是手动的，靠一封邮件发个表格，简直像在 90 年代一样，糟透了。所以我们开始构建一些类似软件框架的东西，让硬件工程流程能自动化、可复用，想法是把迭代的成本降下来。但进展很慢，因为我们永远招不到、也养不起足够多的软件工程师。而现在我们进入了一种完全不同、令人震撼的模式：软件工程师负责搭架构，因为他们懂系统、懂算法、懂关注点分离；然后硬件工程师可以拿他们对硬件工程的理解，去 vibe code 自己的那部分。结果就是小团队的生产力变得截然不同。举个例子，比如你要设计一片涡轮叶片：传统上，叶片冷的时候是一个形状，但运转起来会变热、会变大，所以你得让气动设计和结构设计同时适配它的冷态形状和热态形状，你得在冷态和热态之间换算，在结构和气动之间换算。这一项分析，一片叶片就要一个工程师干一整天，而一台喷气发动机里有上千片叶片，所以你根本做不了多少。现在我们靠软件和硬件人才的结合，真的做出了解决方案：你可以改叶片几何形状，实时看到结构和气动的结果。这样两个工程师就能设计出一整台喷气发动机，这跟以前完全是天壤之别。你刚才提到的一点是，你们让软件工程师为其余工程师打造工具和架构。在我看来，这就是企业软件最大的剧变——再也没有哪家做硬件协作工具的初创公司能卖给你任何东西了，因为你在内部随时把你需要的东西直接编出来就行。连电子表格其实也快完了，对吧？电子表格之所以成功，是因为以前没人能做定制软件，而最接近定制软件的东西，就是一张塞满了 VBScript 函数的表格。我个人已经几乎完全从 Excel 转到了 Python 模型，在那里我能真正做出可信的仿真。

[3:44] Blake Scholl

Yeah. I mean the 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. And when it comes for mechanical and electrical engineering, that will be a whole other thing that we haven't seen yet. That'll be very, very cool.

对。我觉得 AI 还没触及、但我认为一年之内、很可能就在 2026 年内会实现的一件事，会非常非常激动人心：现在它能生成软件，但很快它就能生成 step 文件和 PCB 布局了。等它能搞定机械工程和电气工程的时候，那将是我们还没见过的另一番天地，会非常非常酷。

[4:04] Max Hodak

Yeah. 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 cuz they can't make great software and now they're going to be able to make good enough software. Or it may not even software that is a human front end. It might just be completely agentic for an agent to access and you just talk to it through voice and control hardware. And I this is why one of the reasons why I think for example China is big into open- source models, right? They're basically going all in on it because they have hardware superiority. They have these very complex supply chains and component chains and they're basically saying, "Hey, if I can just generate software on demand, then I don't have this disadvantage anymore against Silicon Valley." So that's not the only reason why they're doing open source. I think they're also behind. They're distilling models. They're catching, you know, they're collaborating on resources. 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. And so I think they want to like uh pull all their resources, catch up on AI and use it to give their hardware stuff an advantage. And ironically, they're doing all the open source stuff cuz Open AI is not open. You know, Gro publishes models, but I think they're a model or two behind. Uh Google has some local models, but nothing really that competitive. anthropic to my knowledge. I don't even know of any open source models from them. So all the open source heft is coming from China. It helps all our hardware founders, but it helps their hardware founders and factories and so on that much more. But all the crappy little software that goes with all the little random knickknacks and thingamajigs that you buy off of Amazon for to tinker with a lazy Saturday afternoon, that software is getting a lot better very quickly. I think everyone's had the wakeup call that without great frontier coding models, you don't have self-improvement. And so imagine China as a whole not having the ability to produce frontier everything, right? It's not just producing software is in any piece of this hardware pipeline like Blake was saying like you need to generate software. If you fall behind on your ability to generate software, you fall behind on the ability to generate everything. One thing I'm curious about from you guys is like because everyone loves to talk about Chinese models like do you use Chinese models? Do you know anybody that uses Chinese models?

对。在硬件这一侧，我觉得这对那些做小玩意儿、做零部件的小公司是个大利好——它们以前软件写得很烂，因为做不出好软件，而现在它们能做出足够好的软件了。甚至可能根本不需要给人用的前端界面，可能完全是 agentic 的、给 agent 调用的，你只要用语音跟它对话就能控制硬件。这也是我认为中国为什么这么押注开源模型的原因之一，对吧？他们基本上是全力投入，因为他们有硬件上的优势——他们有非常复杂的供应链和元器件链，他们等于在说：嘿，只要我能按需生成软件，那我相对硅谷的这个劣势就不存在了。当然，这不是他们搞开源的唯一原因。我觉得他们也是因为落后——他们在蒸馏模型、在追赶、在共享资源上协作。但中国政府一向有这样的传统：去资助那些能带动整个生态发展的项目，尤其是在有网络效应的行业里。所以我觉得他们想把资源集中起来，在 AI 上追上来，再用 AI 给他们的硬件再添一把优势。讽刺的是，他们之所以全在做开源，是因为 OpenAI 并不 open。你看，Grok 会发布模型，但我觉得他们落后一两代；Google 有一些本地模型，但没什么真正有竞争力的；据我所知 Anthropic 嘛——我甚至都不知道他们有什么开源模型。所以所有开源的分量都来自中国。这帮了我们所有的硬件创业者，但更大程度上帮的是他们的硬件创业者、工厂等等。不过那些配在各种零碎小玩意儿上的烂软件——就是你在 Amazon 上随手买来、想在某个慵懒的周六下午鼓捣鼓捣的那些小东西——那些软件正在很快地变得好很多。我觉得大家都已经被敲醒了：没有顶尖的前沿编码模型，你就没有自我改进的能力。所以想象一下，中国作为一个整体，如果不具备生产前沿一切的能力会怎样？这不只是生产软件的问题，整个硬件流水线里的每一环，就像 Blake 说的，你都需要生成软件。如果你在生成软件的能力上落后了，你就会在生成一切的能力上落后。我对你们俩有一个挺好奇的问题：因为大家都爱聊中国模型——你们用中国模型吗？你们认识任何用中国模型的人吗？

[6:21] 嘉宾

This is an argument I had yesterday actually which is uh one person at the table dinner was claiming that uh you know you'll just use deepseek for 97% of things cuz it's so cheap and if you need more intelligence you'll just run it over and over again the same problem and you'll only use the open AI anthropic etc models for the most advanced tasks. And I was kind of like, I don't know. I think intelligence is an unalloided good. You always want more intelligence. And when these models make a mistake, you don't know it. And it's always cheaper than a real person and real time. So you'll just use the most intelligent model available, which isn't great news necessarily because it means that, you know, you're going to end up creating a monopoly or igopoly kind of situation in AI. But uh I always want the most intelligent programmer. I always want the most correct answer. I always want the best judgment. And given the amount of leverage that I'm going to pour into it through capital and code and people and you know marketing I want to make the right decision every time. And often when between two models, 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. Often I actually don't know which is the correct answer, right? So if I know one model is a little smarter, I'm going to go with that answer and eventually I'm going to stop asking the model that I think is less intelligent. But I don't know. Have you guys found a use for the these, you know, so-called less intelligent models? we see uses so that so we have the AI gateways uh data that basically like every application agent etc goes through and so there's definitely usage of open models but the top is like heavily dominated by the frontier intelligence 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 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 and for a lot of tasks other than co coding actually interestingly enough uh they're the best models they're like the best like industrial production models uh you can throw them at like support tasks or browser automation like I would always put a Gemini model there uh and I would look to Chinese models for those kinds of things but anytime I'm working to push the frontier you need the best possible coding model and that's bas basically now like two or three models and uh and the Chinese are not certainly not in it.

这正好是我昨天还争论过的一件事：当时饭桌上有个人主张，说你 97% 的事都会用 DeepSeek，因为它太便宜了，如果你需要更高的智能，你就把同一个问题反复跑很多遍，只有最高难度的任务才会用 OpenAI、Anthropic 这些模型。我当时的感觉是——我不确定。我觉得智能是一种纯粹的好东西，你永远想要更多智能。而且这些模型出错的时候，你是察觉不到的，但它又总是比真人更便宜、更实时。所以你就会一直用现有最聪明的那个模型。这其实不一定是好消息，因为它意味着 AI 最终会演变成一种垄断或寡头垄断的局面。但我永远想要最聪明的程序员，永远想要最正确的答案，永远想要最好的判断力。考虑到我要往里灌注的杠杆量——资本、代码、人力、还有市场营销——我希望每一次都做出正确的决策。而且通常在两个模型之间，比方说我知道其中一个比另一个稍微聪明一点，它俩都给了我答案，很多时候我其实并不知道哪个才是对的，对吧？所以如果我知道一个模型聪明一点，我就会采用它的答案，最终我会干脆不再去问那个我认为比较笨的模型了。但我不知道你们怎么看，你们有没有给这些所谓没那么聪明的模型找到用武之地？我们确实看到一些用法——我们有 AI 网关的数据，基本上每个应用、每个 agent 等等都会经过它，所以开放模型肯定是有使用量的，但顶端被前沿智能严重占据。这里有个子类别、或者说有个附带说明：在合理的成本和性能下的前沿智能，在规模化场景里真的很能打。比如大家对 Gemini 不怎么兴奋，但他们推出的那些模型在合适的性能成本组合下其实超级聪明，而且对很多任务——除了 coding 之外，挺有意思——它们是最好的模型，是最好的工业级量产模型，你可以把它们丢去做客服任务或浏览器自动化，那种场景我永远会放一个 Gemini 模型，这类活我也会考虑中国模型。但任何时候只要我是要去推进前沿，你就需要最好的那个 coding 模型，而那现在基本上就是两三个模型的事，中国模型肯定不在其列。

[8:41] Naval

Hey Max, you're pushing pretty hard into vertical integration and extreme urgency. Do you want to talk about that?

嘿 Max，你在往极致的垂直整合和极端紧迫感方向猛推，要不要聊聊这个？

[8:47] Max Hodak

Yeah, I mean for many things we um you can't buy it so you got to make it somehow. Our preference would always be to buy something um like if there's a vendor that offers a service at a great price. Like for example like PCBs like we don't make PCBs like those are they're basically free. You can buy them in unlimited quantity from Asia. But the the closer that our products get to being like a single block of coalently bonded matter, the better they'll be lower power, smaller, higher performance, last longer. And um there's just like there like the components aren't available. 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. I guess you have to like learn it to do it yourself. And that shows up as vertical integration. So, we own a captive MEMS foundry on the East Coast, which we bought because there was really no other way to do the type of packaging and assembly stuff that we wanted to do. And I think that all of this is going to be affected heavily by AI over the next few years. It's not quite there yet. In fact, ironically, one of the biggest impacts that we've seen of AI inside the companies and regulatory interactions because if we can do things like generate documentation or if we can ask like we want to change, we want to evolve this product like there's thousands of ISO standards that might apply. Which ones do we have to comply with? and like trace this through. 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 kind of knows. Um, but when I think about stuff like the the surgical program or the MEMS FAB, I think ultimately the software still needs hands. Like it's going to be smarter than us, but if it can't make things, then like those are real real boundaries. And so we've instrumented our foundry as well as many other parts of the company in in ways where um as these models get better uh that should show up pretty immediately in in things like the the cell engineering that we're doing and the material science that we're that we're developing.

好。很多东西，你买不到，那就只能想办法自己造。我们其实总是更愿意去买——比如有哪个供应商能以很好的价格提供某项服务的话。举个例子，PCB 我们就不自己做，那些基本上是白菜价，你可以从亚洲不限量地买。但我们的产品越接近成为一整块共价键合的物质，它就越好——功耗更低、体积更小、性能更高、寿命更长。而问题在于，有些元器件根本买不到。要做到那种程度的整合，要真正能在「把买来的现成件拼在一起」之外去创新——而光靠拼现成件其实非常非常受限——我想你就得自己去把它学会、自己去做。这体现出来就是垂直整合。所以我们在东海岸自有一家专属的 MEMS foundry（微机电代工厂），是我们买下来的，因为实在没有别的办法去做我们想做的那种封装和组装。我觉得未来几年这一切都会被 AI 深度影响。现在还没完全到那一步。事实上，讽刺的是，我们在公司内部看到 AI 最大的影响之一，是在和监管机构打交道这块——因为我们可以做一些事，比如生成文档，或者比如我们想改、想迭代某个产品，可能有成千上万条 ISO 标准适用，到底我们必须遵守哪些？然后把这条线索一路追下去。这在以前，你得跟着一整个监管和质量团队忙好几个月，让他们去把这事捋清楚，而现在 AI 基本上就直接知道了。不过当我想到像外科手术项目、或者 MEMS FAB（晶圆厂）这类东西时，我觉得软件归根结底还是需要「手」。它会比我们更聪明，但如果它造不出实物，那就是实实在在的边界。所以我们给我们的代工厂、以及公司很多其他部门都做了仪器化和数据化，这样随着这些模型越来越强，它的效果应该会很快直接体现在我们正在做的细胞工程、正在开发的材料科学这类事情上。

[10:40] 嘉宾

It sort of makes me realize that like it's been a while since I've generated a basic legal document using a lawyer, right? 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 because you know there's the old joke that law is like spaghetti code you know they have this very complicated code that they try to put in English and it contradicts this code over here and has to fit into that code over here and there no real APIs for it. Um 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 and so the same way I think in a way the downside is you can look at law and say you know parallegals just got fired or you could say parallegals just got promoted to senior lawyers and now they can spend their time thinking about the law. 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. You just trust them. Like, hey lawyer, can you look at this document? Can you tell me if it's legit? Can you do red lines? Whatever. Like, at the end of the day, you're what you're valuing in the relationship with a lawyer is that is that they're a trusted authority. They went to law school and they're putting their reputation on the line. I think there's a par parallel with like the biggest problem in software engineering today is these mountains of slob that end up as a PR and then people are say like there's all these memes on Twitter like way back in the day we used to read every line of code of a PR well in my world infrastructure I want engineers to be able to say I understand doesn't necessarily mean that you've read every line of the of the PR 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, the proofs, the type checkers, etc. to be able to say even without reading this, I have confidence I can sign off on it's going to be safe in production. And so it's it's kind of interesting because there's a world in which we embrace that everything is going to be spaghetti code and that we don't fully understand it but we write the basically evaluators that give us confidence and then we rely on like people uh like the infrastructure production engineers to say okay I'm fine uh sending this into prod like someone is going to get paged if your systems go down. I think another thing that people are underestimating is that creating software is really easy 0 to one. But think about a thousand days from now. What is what does your software look like? Is it secure? Is it tested? Is it production grade? Uh is it performant? And are you still motivated to invest all of those tokens in maintaining it in prod?

这让我意识到，我已经有挺久没找律师起草一份基础的法律文件了，对吧？我不再找律师要 NDA、要这个协议、签那个、查这个了——所有这些基础法律活儿也都没了。因为有个老笑话说，法律就像意大利面条式代码（spaghetti code）：他们有一套非常复杂的「代码」，硬要用英文写出来，这边这段「代码」跟那边那段相互矛盾，又得塞进另一处的某段「代码」里，而且根本没有真正的 API。但就 junior engineer、junior engineering 而言——我应该说，junior engineer 基本上被提拔成了 senior engineer，而 junior engineering 这类活则被 agent 接管了。所以同样地，从某种意义上说，坏的一面是你可以看着法律行业说「初级律师助理（paralegal）刚被裁了」；但你也可以说「paralegal 刚被提拔成了 senior lawyer，现在他们可以把时间花在真正思考法律上」。把软件工程的演变方式和律师做个类比其实挺有意思的，因为律师嘛——你永远不知道他们往这些文件里塞了什么，你就是信任他们。比如「嘿律师，你能看看这份文件吗？能告诉我它靠不靠谱吗？能帮我做下 red line 修订吗？」诸如此类。说到底，你在和律师的关系里看重的，就是他们是一个可信赖的权威：他们上过法学院，把自己的声誉押在了上面。我觉得这跟软件工程有个类比：今天软件工程最大的问题，是堆积如山的「垃圾代码」最后变成一个 PR。Twitter 上有各种 meme 说，想当年我们会逐行读完一个 PR 的每一行代码。而在我的世界——基础设施——里，我希望工程师能够说「我理解它」，但这不一定意味着你读了 PR 的每一行；你需要能够说「我对理解这个 PR 的后果负责签字」，或者「我写了测试框架、仿真、证明、类型检查器等等，所以即便不逐行读，我也有信心签字担保它上生产环境会是安全的」。所以这挺有意思的：存在一种可能的世界，我们坦然接受「一切都会是 spaghetti code、我们并不完全理解它」，但我们去写各种评估器（evaluator）来给我们信心，然后我们依靠像基础设施生产工程师这样的人来说「好，我没问题，可以推到 prod」——因为如果你的系统宕了，总会有人被呼叫到（被 page）。我觉得人们还低估了另一件事：从 0 到 1 创造软件其实非常容易，但想想 1000 天之后呢？你的软件会变成什么样？它安全吗？经过测试了吗？是生产级的吗？性能好吗？而你还有没有动力继续把所有这些 token 都投进去、维护它在 prod 上的运行？

[13:30] 嘉宾

I mean, humans are becoming verifiers, right? And and that's kind of how we train these models with good verification data. And now we need human verifiers. 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, this is roughly correct and I I'll roughly stand behind it and I'll support you if it goes wrong.

我是说，人正在变成验证者（verifier），对吧？我们训练这些模型靠的就是优质的验证数据，而现在我们需要人来做验证者。所以是的，我觉得人的很多旧职能——律师、工程师、运营人员——都转向了去验证整个技术栈，去说「嗯，这个大致是对的，我大致愿意背书，如果出了问题我会支持你」。