代码 (code)
集里怎么说它
- 《Figma Config 2026:当代码、AI 与动效都变成画布上的设计材料》(01:58起):本集从根本观念上把它重新定义为一种用于设计的材料，打破它与设计对立的伪命题，它不应再被工具困在孤岛里，而应像图层一样在画布里被探索。
① 提到它的金句
243 条
因为为智能体构建的代码库实际上看起来和为人类构建的代码库不一样。
指向原始笔记的链接
Because the code base that’s built for agents actually looks different than the code base built for humans.
—— 嘉宾 · [42:06]
我认为 Anthropic 大约 80% 的人员是技术人员，每天都使用 Claude Code。
指向原始笔记的链接
I think about 80% of people at Anthropic that are technical use Claude Code every day.
—— Boris · [27:08]
我认为人们真的忽视了 Claude Code 对非程序员来说有多好用。
指向原始笔记的链接
I think people are truly sleeping on how good Claude Code is for non-coders.
—— Dan Shipper · [07:16]
我认为这样的原因是，我确实认为创建软件的行为将从输入到终端或输入到 Visual Studio 代码转变为操作一台代码生成机器。
指向原始笔记的链接
The reason I think that is I do think the act of creating software is going to transform from typing into a terminal or typing into Visual Studio code to operating a code-generating machine.
—— Bret Taylor · [32:29]
我认为我们需要更多类似的东西，因为如果 AI 正在生成这些代码，根据定义，如果你必须阅读每一行，那将是生产代码的制约因素，或者更糟的是，你只是不会阅读每一行，而且你将向世界发布一堆不安全、未经验证的代码。
指向原始笔记的链接
I think we need more things like that because if an AI is generating this code, by definition, if you have to read every line that is going to be the limiting factor for producing the code or worse, you’re just not going to read every line and you’re going to emit a bunch of unsafe unverified code into the wild.
—— Bret Taylor · [41:16]
但现实是，至少就目前而言，AI 模型吐出的很多代码在根本上并不安全。
指向原始笔记的链接
But the reality is, at least as it sits today, a lot of code that AI models spit out is not fundamentally secure.
—— Vineet Edupuganti · [05:48]
所以我们的目标之一是确保，每当我们收到来自客户的反馈，它都能在 24 小时内进入我们的代码库。
指向原始笔记的链接
So one of our goals is to make sure that, like, every time we cut feedback from customers, that gets to our code base in 24 hours.
—— Evan Reiser · [27:06]
我见过市场销售团队的成员，按各种标准衡量他们都不该会写代码，但这些人正在构建应用程序和工具，去解决他们自己遇到的问题。
指向原始笔记的链接
I’ve had members of the go-to-market team where by all measures they should not be able to write code and these guys are building applications and tools to go and solve the problems that they have.
—— Vineet Edupuganti · [28:40]
如果目标是更多的代码行，我可以提示某物写出有史以来最长的代码并添加大量的注释。
指向原始笔记的链接
If the goal is more lines of code, I can prompt something to write the longest piece of code ever and add tons of comments.
—— Nicole Forsgren · [12:56]
现在这么多时间将花在审查代码而不是编写代码上。
指向原始笔记的链接
So much of the time is now going to be spent reviewing code versus writing code.
—— Lenny · [00:42]
但它现在是什么，好吧，如果你和昨天世界上最优秀的程序员交谈，他们会告诉你，“噢，我的工作是我坐在那里，我正在编排 10 个代码机器人，正在并行运行的编码机器人。“
指向原始笔记的链接
But what it is now is, all right, if you talk to the world’s best programmer of yesterday, what they’ll tell you is, “Oh, my job is I’m sitting there and I’m orchestrating 10 code bots, coding bots that are running in parallel.”
—— Marc Andreessen · [45:29]
如果目标是成为一名平庸的编码员，那就让 AI 去做吧。没问题。AI 将非常擅长生成无限的平庸代码。
指向原始笔记的链接
If the goal is to be a mediocre coder, then just let the AI do it. It’s fine. The AI is going to be perfectly good in generating infinite amounts of mediocre code.
—— Marc Andreessen · [47:37]
虽然英语是第一编程语言，但 Lovable 和所有其他工具仍然用代码进行最佳交流。
指向原始笔记的链接
Even though English is the number one programming language, Lovable and all other tools still communicate in code the best.
—— Lazar Jovanovic · [24:01]
你可以编码很快，但你可以很快编码出垃圾，也可以很快编码出魔法。所花费的时间是一样的。重要的是你和你的输入。
指向原始笔记的链接
You can code fast, but you can code garbage fast as well as magic fast. It’s the same amount of time. It’s you and your input that matters.
—— Lazar Jovanovic · [98:47]
有一个团队实际上正在 OpenAI 内部进行一项实验，他们维护着一个 100% 由 Codex 编写的代码库。
指向原始笔记的链接
There’s a team that’s actually doing an experiment right now within OpenAI where they are maintaining a 100% Codex-written code base.
—— Sherwin Wu · [00:35]
而且如果你看那条指数曲线，那个时间点由 Claude 编写的代码百分比，如果你只是沿着线画过去，很明显我们在年底就会超过 100%，即使这完全不符合直觉。
指向原始笔记的链接
If you look at the exponential, the percent of code that was written by Claude at that point, if you just trace the line, it’s pretty obvious we’re going to cross 100% by the end of the year, even if it just does not match intuition at all.
—— Boris Cherny · [14:40]
我的看法是，对于今天正在使用 Claude Code，正在使用智能体写代码的人来说，你仍然必须理解底下那层，但是，是的，在一两年后，这就不重要了。
指向原始笔记的链接
My take is I think for people that are using Claude Code, that are using agents to code today, you still have to understand the layer under, but, yeah, in a year or two, it’s not going to matter.
—— Boris Cherny · [32:27]
我花了一整天的时间来完善那段代码，确保我理解所有东西，撰写一个恰当的 PR 描述，因为我不想失去我的身份
指向原始笔记的链接
I spent a whole lot of day polishing up that code, making sure I understand everything, crafting a proper PR description because I don’t want to lose my identity
—— 嘉宾 · [19:03]
随着编写代码的成本变低得多,变得更有价值的是决定写什么。
指向原始笔记的链接
As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.
—— Kat Wu · [00:50]
我认为巨大的转变在于,2024 年一代的产品是基于聊天的,而云端代码这一代的产品是基于行动的。
指向原始笔记的链接
I think the big shift is that the 2024 generation of products were chat-based and the cloud code generation of products is action-based
—— Kat Wu · [74:38]
如果 Claude Code 失败了,但 Anthropic 成功了,我会非常高兴。
指向原始笔记的链接
if Claude Code failed, but Anthropic succeeded, I would be extremely happy.
—— Kat Wu · [31:47]
我认为把 token 花费作为吹嘘的指标存在真正的危险，这就像人们吹嘘他们一天写了多少行代码一样。
指向原始笔记的链接
I think there’s a real danger in making the token spend the metric to boast about, which is the same as when people boast about how many lines of code they’ve written in a day.
—— Max Schoening · [36:08]
建立交易所的困难部分不是为交易所编写代码。而是说服每个人围绕你的交易所进行协调，而不是别人的。
指向原始笔记的链接
The hard part of setting up an exchange is not writing the code for the exchange. It’s convincing everybody to coordinate around yours rather than somebody else’s.
—— Patrick Collison · [29:24]
过去三十年里，我们被告知永远不要执行一段随机代码。这意味着科技树的整个分支一直未被探索。
指向原始笔记的链接
For the last thirty years, we have been told to never evaluate a random code. Yeah. Which means there’s this entire branch of the tech tree that’s been, like, unexplored.
—— Matt Carey · [07:11]
其中一个是认知债务。特别是当我们习惯了代码生成得如此之快这一事实，我们也可能开始丧失关于事情如何正确完成的肌肉记忆。
指向原始笔记的链接
One of them is cognitive debt. And so especially as we get used to the fact that code is getting generated so quickly right now, we can also start to lose our muscle memory around how things get done right.
—— 嘉宾 · [12:37]
使用这些工具保持高速度的一个很大部分，就是不允许垃圾代码进入代码库。
指向原始笔记的链接
A big part of maintaining high velocity with these tools is just not permitting slop to enter the code base.
—— Ryan Lopopolo · [11:36]
我实际上发现先生成代码会容易得多，让那段代码有点像是一个关于世界可能样子的草案，作为团队与之互动来完善它，然后从中提炼出规范。
指向原始笔记的链接
I’ve actually found it is much easier to produce the code first, to have that code sort of like present a straw man for what the world could look like, engage with that as a team to refine it, and then distill the spec out of that.
—— Ryan Lopopolo · [28:20]
因为每个人都可以写自己的框架，你的不一定比我的更好或更差，因为实际上就像，如果你的整个防御或护城河是你与其他所有东西的集成生态系统，嗯，我也可以 vibe code。所以，你到底有什么用？
指向原始笔记的链接
Because everyone can write their own framework, and yours is not not necessarily as better or worse than mine because it’s actually like, if your your whole defense or your moat is your ecosystem of integrations with everything else, well, I can vibe code that too. So, like, what the hell use are you?
—— Swyx · [16:04]
作为 CTO,我不想登上头版,就像我安装了某个奇怪的 NPM 包并泄露了所有的代码。
指向原始笔记的链接
as the CTO, I don’t want to end up on the front page as like I installed some weird NPM package and leaked all the code.
—— Matei Zaharia · [21:47]
多年来设计界无休止地讨论设计与代码这个话题,我现在就直说了,这是一场虚假的辩论。
指向原始笔记的链接
For years the design industry has endlessly discussed this topic of design versus code and I will just say it flat out right now this is a false debate.
—— Dylan Field · [09:29]
虽然智能体让写数千行代码变得容易得多,但围绕代码的工作流根本不是为共同探索而构建的。
指向原始笔记的链接
while agents have made it so much easier to write thousands of lines of code, the workflows around code are not built for exploring together at all.
—— Nico · [12:20]
你今天看到的一切,代码、动画、你自己的工具、你自己的素材,它们都共存于画布之中。
指向原始笔记的链接
Everything you saw today, code, motion, your own tools, your own materials, it all lives together in the canvas.
—— Dylan Field · [69:43]
只是你不是把它应用在代码上，你是把它应用在制造代码的机器上。
指向原始笔记的链接
It’s just you’re not applying it to the code, you’re applying it to the machine that makes the code.
—— Daniel Jones · [27:31]
我现在称之为软件工程的垃圾桶法,在那里你实际上可以丢弃所有代码,就像开始一个 slash V2 分支并从头开始重建它。
指向原始笔记的链接
I call this the trashcan method of software engineering right now, where you can actually trash all the code, start like a slash V2 branch and rebuild it from scratch.
—— Claire Vo · [00:57]
如何编写干净代码的旧价值观等等都不重要了
指向原始笔记的链接
the old values of how you write clean code and blah, blah, blah don’t matter
—— Kitsa · [13:31]
因为想法是容易的。想法是显而易见地容易。我们每年有一千个想法,其中很多甚至拿不到代号。想法不是难的部分。我们的 alpha 来自我们能多高效地发现哪些想法有价值。
指向原始笔记的链接
Because the ideas are easy. Ideas are trivially easy. We have a thousand ideas a year, that many of them just don’t even get code names. Ideas are not the hard part. Our alpha comes from how efficiently we can discover which ideas are valuable.
—— Astro Teller · [13:20]
无论你多厉害，你都不可能用 vibe coding 做出 Shopify。
指向原始笔记的链接
You’re not going to vibe code Shopify, no matter how good you are.
—— Avishai Abrahami · [00:00]
当你坐下来使用 Cloud Code 或 Codex 时，你不是在编写软件，你是在雇佣、培训和管理一支由 Markdown 组成的劳动力。
指向原始笔记的链接
When you sit down with Cloud Code or Codex, you’re not writing software, you’re hiring, training, and managing a workforce made of Markdown.
—— Garry Tan · [05:58]
所以,可能是一个智能体编写了代码,或者我帮助进行了分析,而这并不是我的背景,但这并不意味着人们不需要对他们所创造的东西带来的责任负责。
指向原始笔记的链接
So, it can be that an agent wrote the code or I helped to do an analysis when that’s not really my background, but it doesn’t make people not have the responsibility that comes with what they’ve created.
—— Elizabeth Stone · [06:02]
但我认为如果说设计和深刻的设计专业知识及思维因此而被挤出,那将是一个错误,仅仅是因为我们可以更快地编写代码,我们可以更快地进行数据分析。
指向原始笔记的链接
But I think it would be a mistake to say design and deep design expertise and thinking gets squeezed out, just because we can write code faster, we can do data analysis faster.
—— Elizabeth Stone · [21:32]
换句话说,如果明天所有人都病倒了,很多 Claude Code 的使用量会直接归零。
指向原始笔记的链接
In other words, like if everyone woke up sick tomorrow, like a lot of Claude code usage would be zero.
—— Matan Grinberg · [46:27]
在对冲基金,我们要试图交易数十亿美元,所以我们不能仅仅让 vibe coding 成为这些分析的基石。
指向原始笔记的链接
At a hedge fund, we’re trying to trade billions of dollars, and so we can’t have just vibe code be the underpinning of how these analysis go.
—— 嘉宾 · [17:16]
对代码的理解将会变少，因为没有人再编写代码了。
指向原始笔记的链接
So much less understanding of code is going to be present because no one wrote the code anymore.
—— Anish · [02:07]
每个问题的答案都是一个模型。你很想为它写一大堆代码？不。
指向原始笔记的链接
The answer to every problem is a model. You’re tempted to write a bunch of code for it? No.
—— 嘉宾 · [00:00]
我们的目标是把更多东西放进学习到的模型里，而不是放在代码里。
指向原始笔记的链接
our goal is to push more things inside learned models rather than in code.
—— 嘉宾 · [24:58]
对于那些不是在构建智能体产品、但在使用 Claude Code 的人:每六个月,删掉你的 Claude MD。
指向原始笔记的链接
And for people that aren’t building agentic products, but you’re using Claude Code, every six months, delete your Claude MD.
—— Boris Cherny · [06:56]
它运行了 11 天,重写了整个代码库。
指向原始笔记的链接
And it ran for 11 days. And it rewrote the entire code base.
—— Boris Cherny · [17:40]
我上次看的时候，Dark Factory 代码库中只有大约 5% 曾经由人类查看过 PR 并批准它。
指向原始笔记的链接
Last time I looked, I think about only 5% of the Dark Factory code base had ever had a human look at the PR and approve it.
—— Rob Willoughby · [00:11]
我认为人们实际上不再去查看代码了，因为他们都在使用智能体。
指向原始笔记的链接
I don’t think people actually go look at code anymore because they’re all just using agents.
—— Rob Willoughby · [00:18]
而且我不认为人们还会去查看代码了，因为他们都只是在用智能体。
指向原始笔记的链接
And I’m not, I don’t think people actually go look at code anymore because they’re all just using agents.
—— Rob Willoughby · [32:01]
它运行了11天。它重写了整个代码库。
指向原始笔记的链接
And it ran for 11 days. And it rewrote the entire code base.
—— Boris Cherny · [18:01]
写代码的问题在于你可能有 10,000 行你希望是对的代码,但真正去阅读并理解它们几乎是不可能的;理解代码可能比写代码花的时间还多。
指向原始笔记的链接
The problem with writing code is that you might have 10,000 lines you hope are right, but actually reading and understanding them is almost impossible; it can take more time to understand the code than to write it.
—— Carlos González de Villaumbrosia · [24:10]
他们审查的不是 10,000 行代码;他们审查的是一个他们能轻松理解并逐步走过的工作流,所以他们确切知道它会做什么。
指向原始笔记的链接
And they’re not reviewing 10,000 lines of code; they’re reviewing a workflow they can easily understand and step through, so they know exactly what it will do.
—— Carlos González de Villaumbrosia · [35:08]
劣质代码就是任何你不阅读的代码。
指向原始笔记的链接
Slop is just any code you don’t read.
—— Vaibhav Gupta · [01:30]
无论你们承不承认,这都是你们的代码库曾经拥有的最少量的劣质代码。
指向原始笔记的链接
Whether any of you admit it or not, this is the least amount of slop that your code base will ever have.
—— Vaibhav Gupta · [01:32]
代码可以是劣质的,写作不能。
指向原始笔记的链接
Code can be slopped, writing cannot.
—— Vaibhav Gupta · [02:46]
代码是信任的问题。
指向原始笔记的链接
Code is a matter of trust.
—— Vaibhav Gupta · [17:15]
我们不盲目使用 LLM 代码的原因是我们还不信任它,因为它们底层的系统没有足够的刚性。
指向原始笔记的链接
The reason that we don’t use LLM code blindly is because we don’t trust it yet, because the systems underneath them don’t have enough rigidity.
—— Vaibhav Gupta · [17:17]
AI 审批的 PR 以及总体上 AI 编写的代码甚至可以比仅有人类在流程中所做的更安全、质量更高。
指向原始笔记的链接
AI approved PRs and AI written code in general can be even safer and even higher quality than what you’re doing with just a human in the loop.
—— Claire · [03:47]
我的核心观点是个人 AI 代码生成打破了传统的云基础设施。
指向原始笔记的链接
My key point is personal AI code gen breaks traditional cloud infrastructure.
—— Kenton Varda · [00:37]
我认为仅靠代码技能正变得过时。
指向原始笔记的链接
I think code skills alone are becoming passe.
—— Oji Udezue · [58:31]
我们发现的核心张力在于，这些智能体喜欢看起来像代码的东西。
指向原始笔记的链接
The central tension that we found is that these agents love things that feel like code.
—— 嘉宾 · [06:03]
如果你作为工程师让智能体做出关键决策，你就是在放弃你的代码的控制权。
指向原始笔记的链接
If you as an engineer are letting an agent make a critical decision, you are ceding control of your code.
—— Matt Dailey · [03:27]
你正在从代码速度转向想法速度。
指向原始笔记的链接
You’re shifting from code velocity to idea velocity.
—— Matt Dailey · [15:13]
我热爱写代码、喜欢构建产品，能写多少就写多少，但最近我在有意识地让自己戒掉这件事，因为公司的瓶颈不在这。
指向原始笔记的链接
I love writing code and like building the product so as much as I can, but recently I’ve been like consciously weeding myself off of it because the bottlenecks of the company.
—— Ali Haghani · [07:47]
就像现在的工程师不再写代码，而是 vibe code 并引导这些编码模型一样，我们也开始做同样的事情。
指向原始笔记的链接
the same way that engineers now don’t write code, they vibe code and guide these coding models, we’re starting to do the same thing.
—— Gabe Pereyra · [03:48]
代码不再能代表你的能力了。
指向原始笔记的链接
The code doesn’t speak for your ability anymore.
—— Wes Bos · [01:17]
但是无代码的时代，那些人类可以在没有工程资源的情况下做的事情，它正在慢慢结束。
指向原始笔记的链接
But the era of no code, of things that humans can do without engineering resources, it’s slowly winding down.
—— Jason Lamkin · [13:04]
Basil 可能亲自写了公司大概 60% 的代码。
指向原始笔记的链接
Basel probably wrote, I don’t know, 60% of the code of the company himself.
—— Adam · [06:53]
但即便如此，智能体在代码审查方面也比以前的人类做得更好。
指向原始笔记的链接
But even then, agents are doing code review even better than humans did before.
—— Will Gabrick · [26:52]
之所以难,是因为我觉得我们只是在两个世界之间搭一座桥,而这并不是最优的,你仍然需要知道该用什么词汇去问,用引擎的词汇、用代码的词汇。
指向原始笔记的链接
It’s hard because I think we’re just building a bridge between two worlds, and it’s not optimal, and you still need to know kind of like what to ask in the vocabulary of an engine, in the vocabulary of code.
—— Arturo Nunez · [07:04]
现在如果你正在使用 Claude Code 或 Codex，很有可能已经启动了 20 或 30 个子智能体。
指向原始笔记的链接
And now if you’re using Claude Code or Codex, it’s very possible that 20 or 30 sub-agents have fired.
—— Addy Osmani · [63:04]
但如果你把整个系统硬编码，说我们要抛弃这整个智能体流程，你就限制了自己，或者你的代码会膨胀到完全无法管理。
指向原始笔记的链接
But if you hard code your whole system, you say we’re going to throw out this whole agentic process, you limit yourself or your code can explode to be just unmanageable.
—— Vasant Kearney · [10:54]
把你还没读过的代码交给其他审阅者去看，这个诱惑太大了。不要这么做。
指向原始笔记的链接
It’s so tempting to put other reviewers on code you haven’t read yet. Don’t do it.
—— Hursh Agrawal · [17:07]
我想先说一句：今年我还没写过一行代码。
指向原始笔记的链接
I want to start with saying I haven’t written a single line of code this year.
—— Kieran Klaassen · [00:22]
重要的是记录思考，而不是代码。
指向原始笔记的链接
It’s important to document the thinking, not the code.
—— Kieran Klaassen · [17:37]
但一定要抬高门槛，因为我们不发布垃圾代码。
指向原始笔记的链接
But make sure to raise the bar because we’re not shipping shitty code.
—— Kieran Klaassen · [19:31]
正如我提到的，那才是这里真正的金矿，要让代码审查自动化。
指向原始笔记的链接
As I mentioned, that’s the real gold mine here to get the code review automated.
—— Itamar Friedman · [09:43]
我实际上在说的是，软件开发，至少代码治理，将从审查你的 pull request 转变为从一个图抽象层面审查你的整个软件开发。
指向原始笔记的链接
What I’m actually saying is that software development, at least code governance, is going to change from reviewing your pull request to actually reviewing your entire software development from a graph obstruction.
—— Itamar Friedman · [14:11]
如果你已经在以快于人类能审查的速度发布 AI 生成的代码，我要说的是，你正身处问题之中。
指向原始笔记的链接
If you are already shipping AI-generated code faster than your human can review, I’m actually saying that you are in the problem.
—— Itamar Friedman · [15:47]
如果你搭建了基础设施，如果你收集了上下文，如果你开始积累代码审查如何为你自动化的经验，那才是你能获得 10 倍速度的地方，也就是你向你的 CEO 或你自己或你的开发者承诺过的速度，否则它就是一个瓶颈。
指向原始笔记的链接
If you put the infrastructure, if you gather the context, if you start accumulating how code review can be automated for you, then that’s where you’re gonna get the 10x velocity that you’re being promising your CEO or yourself or your developers, because otherwise it’s a bottleneck.
—— Itamar Friedman · [15:59]
我们为工程师处理了超过 250 次自动化迁移，自动化迁移了约 900 万行代码。
指向原始笔记的链接
We handled more than 250 automated migrations, relatively 9 million lines of code automatically for our engineers.
—— Uday Kiran Medisetty · [01:04]
你会怎么对待一个既了解你的代码库又能编写代码来解决你业务中任何问题的人？
指向原始笔记的链接
What would you do with somebody who both knew your code base and could write code to solve anything in your business?
—— Claire · [23:28]
你们时间线上最沉迷 AI 的两个人花了我们大部分时间谈论太多的 AI 代码其实并不能给你带来好产品。
指向原始笔记的链接
The two most AI-pilled people on your timeline spent most of our time talking about how too much AI code doesn’t really get you great product.
—— Claire · [42:13]
然后它还能确保我的测试不会完全糟糕透顶，因为当你让 AI 写测试时，一个很大的危险是——我对 AI 写测试的信任甚至低于我对它写代码的信任。
指向原始笔记的链接
And then um it also um it also makes sure my tests are not completely horrible because like when you have AI write tests, a big danger, like I trust my AI with my tests even less than I trust it with my code.
—— Lada Kesseler · [32:41]
因为根据我的经验，通常架构正确的代码也是运行快速的代码。
指向原始笔记的链接
Because in my experience, usually the code that is architected properly is also the code that runs quickly.
—— Casey Moratori · [85:08]
我们可以看到,在许多不同的模型之间,与试图搜索网页、找到正确的工具相比,能节省近 50% 的 token,不用从你的代码库中拉取 markdown 文件。
指向原始笔记的链接
We can see that between many different models, almost 50% token saving on instead of trying to search the web, find the right tools, pulling the markdown files from your code base.
—— Christopher Burns · [11:22]
未来的代码看起来会像伪代码。
指向原始笔记的链接
Code in the future will look like pseudocode.
—— Martin Casado · [04:33]
我看到有人这么做，我就想，哦等等，我一直完全低估了这个东西。我只是把它当作 Slack 里一个美化版的 Claude Code 在用。
指向原始笔记的链接
I saw somebody do that, I was like, oh wait, I’ve been totally underutilizing this thing. I’ve just been using it as like a glorified Claude Code in Slack.
—— Mike Krieger · [09:33]
它掩盖了一个事实：写代码从来都不是限制性的环节。也许在时间线层面上是，但它从来都不是决定你初创企业成败的东西。
指向原始笔记的链接
It papers over the fact that writing code was never the limiting part. Maybe it was on the timeline perspective, but it was never the thing that was going to make or break your startup.
—— Mike Krieger · [19:29]
今天，Anthropic是什么，它全是关于代码。那就是那句话。那是唯一重要的一句话。
指向原始笔记的链接
Today, what Anthropic is, it’s all about code. That’s the sentence. That’s the only sentence that matters.
—— 嘉宾 · [38:05]
我们最初的尝试是把 Base64 数据直接传进代码，那是个非常糟糕的主意。它直接把上下文窗口撑爆了，方方面面都很糟。别那么做。
指向原始笔记的链接
Our first attempt was just passing Base64 data into the code, and that was just a terrible idea. It just blew up the context window and was bad all around. Don’t do that.
—— Jesse Lumarie · [05:16]
简而言之,代码评审现在成了我们遭遇的瓶颈。
指向原始笔记的链接
So in short, code review is now the bottleneck that we are running into.
—— Will Bond · [01:16]
否则我们可能会陷入智能体的空转:它修复了某个问题,回头又收到一次代码评审,然后不得不反向再修一遍,因为那条评论的质量很低。
指向原始笔记的链接
Or else we can result in dealing with cavitation of an agent where it fixes something, goes back, gets another code review, and has to kind of like fix backwards because the quality of the comment was low.
—— Will Bond · [12:23]
与其把人类从代码评审流程中移除,我们是把他们的职责上移一层。
指向原始笔记的链接
Rather than removing humans from the code review process, we are moving their responsibilities up a layer.
—— Will Bond · [13:55]
你不再是去担心性能优化和 API 兼容性,而是在代码评审中更多地去思考架构。
指向原始笔记的链接
Instead of you worrying about the optimization of the performance and the API compatibility, you’re going to be thinking more about architecture in your code reviews.
—— Will Bond · [14:19]
前沿开发者自己写的代码可能只占他们产出代码的 1% 到 2%,其余的都是智能体写的。
指向原始笔记的链接
Frontier developers write maybe 1% to 2% of the code that they produce. The rest is agents.
—— Clare Liguori · [01:52]
你必须先做有意的工程工作,才能看到生产率提升的那条曲棍球杆曲线,因为我们必须先在代码库里做真正的工作,智能体才能在那里取得成功,尤其是在既有代码库中。
指向原始笔记的链接
You have to do intentional engineering work before you’re going to see that hockey stick curve in productivity improvement because we have to do real work in our code bases first for agents to be successful there, especially in Brownfield existing code bases.
—— Clare Liguori · [09:54]
而我发现,当意图本身就不正确时,与智能体在代码上迭代的效率会低一些。
指向原始笔记的链接
And it is less, I find, less productive to iterate with the agent on code when the intent itself was incorrect.
—— Clare Liguori · [13:14]
你花越多时间去审查是否要真正构建一个新产品的决策,现在构建这个产品就越慢,因为代码只需要一两个月就能写完。
指向原始笔记的链接
The more that you spend reviewing the decision to actually build a new product, the slower it is to build the product now because the code only takes one to two months to write.
—— Clare Liguori · [19:02]
我经常发现,前沿工程团队花在做决策上的时间比写代码还多。
指向原始笔记的链接
Often I find that frontier engineering teams spend more time making decisions than they do writing code.
—— Clare Liguori · [19:44]
与其等待模型赶上做服务知识工作，如果我们只是使用那些代码知识并将知识工作表示为代码会怎么样？
指向原始笔记的链接
Rather than wait for the models to catch up on doing services knowledge work, what if we just use that code knowledge and represent knowledge work as code?
—— Varun Shenoy · [09:35]
我想这里的很多演讲者都提到过这一点，但投资于验证可能是我们在代码库中能做的价值最高的事情。
指向原始笔记的链接
I think a lot of the speakers here have touched upon this, but investing in verification is probably the highest value thing we can do in our code base.
—— Eyal Blum · [05:00]
如果你告诉你的技能或智能体按「红到绿」的 TDD 风格去写你正在写的代码，它几乎总能给你更好的结果。
指向原始笔记的链接
If you tell your skills or your agent to write the code that you’re writing, like at the red to green, at the TDD style, it almost always gives you better results
—— Eyal Blum · [06:13]
因为那样的话，它会让测试去迁就代码，而不是让代码去满足验证标准。
指向原始笔记的链接
because then it will fit the test to the code rather than fit the code to pass the verification criteria.
—— Eyal Blum · [06:34]
他们之所以持怀疑态度，是因为他们看到了你在验证上缺失的地方、工具失效的地方——所以他们的反馈基本上就是如何改进智能体与代码库交互的路线图。
指向原始笔记的链接
They’re spec haptic because they’re seeing where you’re lacking validation, where your tools fail, so their feedback is basically the roadmap of how to improve your agent interacting with the code base.
—— Eyal Blum · [11:59]
有生产力的、高质量的工程工作,可以说删除代码甚至更好。
指向原始笔记的链接
Productive and high-quality edge work, one can argue that removing code is even better.
—— Mingsheng Hong · [10:06]
有可能存在一个只有 10 行代码的 PR,却花了很长时间,找出并修复了一个并发 bug。
指向原始笔记的链接
There can be a PR with only 10 lines of code that takes forever that finds and fix a concurrency bug.
—— Mingsheng Hong · [11:29]
任何做这行有一段时间的人都知道，代码行数是一个很糟糕的指标。
指向原始笔记的链接
Anybody who’s been doing it any length of time knows that lines of code is a terrible metric.
—— 嘉宾 · [69:59]
而且实际上，这种特异性——能够精确调整你正在审查的内容——在你走向工厂的过程中变得越来越重要，因为说真的，工厂将要产出的量，你必须让人类退出代码审查的循环。
指向原始笔记的链接
And in fact, this specificity, you know, being able to tune exactly what you’re reviewing becomes increasingly important as you move towards factory because really the volume a factory is going to put out, you have to get humans out of the loop for code review.
—— Drew · [20:02]
具体性基本上是所有质量的标尺。比如如果你能给智能体一个非常具体的任务，让它对要做什么有清晰的指示，那就会比一个泛泛的『这样去找代码坏味道』要好。
指向原始笔记的链接
Specificity basically is the ruler of all quality. Like if you can have a very specific task that the agent has clear instructions on what to do, it’s gonna be better than a generic here’s how to look for code smells.
—— Drew · [42:17]
在大多数情况下，已经不再是由人来写代码了。理想情况下，我有一份规格说明，代码是从它生成的。
指向原始笔记的链接
So it’s not someone writing code anymore in most cases. Ideally, I have a spec. I have a spec from where the code is generated.
—— Yorg Shad · [24:45]
逆风，但我们正在构建的代码量字面意义上是 100 倍。
指向原始笔记的链接
Headwinds, but literally the amount of code we’re building is 100x.
—— 嘉宾 · [40:16]
很容易说大多数人不想写代码，但我想法上的大转变是，编程智能体其实是一种与世界互动的通用方式。
指向原始笔记的链接
I think it’s easy to say most people don’t want to make code, but the big change in my thinking is that coding agents are a way to interact with the world generally.
—— Anish Acharya · [51:57]
我们已经越过了人类会阅读每一行代码的时间点。
指向原始笔记的链接
We’ve passed the point where humans will read every line of code.
—— JD · [39:57]
如果你把所有东西都写在同一个代码库里、同一种语言里,那不可避免地你会写得有点草率,你会把不该纠缠的东西纠缠在一起。
指向原始笔记的链接
And if you write everything in the same code base, in the same language, it’s like inevitably you’re going to be a little bit sloppy and you’re going to intertwine things more than you should.
—— Tibo Sottiaux · [19:54]
如果你在做某件事,而且你在搭建一个一万行的代码拐杖来绕过模型的缺陷,那你很可能在做错误的事情。
指向原始笔记的链接
If you’re doing something and you’re building like this 10,000 lines of code crutch to work around the model flaws, you’re probably doing the wrong thing.
—— Tibo Sottiaux · [42:29]
当我们做基准测试时,它们在代码评审上是超越人类的。
指向原始笔记的链接
When we benchmark them, it’s like they’re superhuman in code review.
—— Tibo Sottiaux · [47:53]
所以那大概算是比较不讲道理的事情之一，比如，对，就把这整个 Python 代码库移植到 TypeScript，让它跑起来、能部署，就在一个周末之内。
指向原始笔记的链接
So that probably ranks on like the more unreasonable things like, yeah, just port this entire Python code base to TypeScript, get it working, get it deployable in, you know, a weekend.
—— 嘉宾 · [00:41]
AI 生成的代码,应该让人感觉是那种已经在你团队待了多年的人写出来的。
指向原始笔记的链接
AI-generated code should feel like it was written by someone who’s been on your team for years.
—— Brandon Waselnuk · [01:13]
完全相同的问题,如果所有的代码审查智能体都以同样的方式带有偏见,它们就会漏掉完全同一类 bug,那我们就全完了。
指向原始笔记的链接
Same exact issue, if all the code review agents are biased in the same way, they will miss exactly the same class of bugs and then we’re all sunk.
—— Chris Potts · [55:13]
这遵循了我们 Prefect 的很多软件都努力坚持的一个原则，那就是一行代码，一个显著的大变化。
指向原始笔记的链接
So this follows a principle that we really try to hold in a lot of our software at Prefect, which is one line of code, one big noticeable change.
—— Jeremiah Lowin · [11:49]
从某种意义上说，现在我们有一个理解设计意图的智能，而同一个智能也能写代码。
指向原始笔记的链接
In a sense, now we have one intelligence that understands design intent and the same intelligence can write code.
—— Jonathan Gordon · [03:26]
工程师的时间只有大约 40% 用于写代码,其中只有约 10% 用于写新代码。也就是说,工程师有 60% 的时间花在非编码任务、非编码工作上。
指向原始笔记的链接
An engineer’s time is only like 40% writing code and only like 10% writing new code. So there’s 60% of an engineer’s time that is non-coding tasks, non-coding work.
—— Robert Lucero · [27:20]
如果你的智能体代码所落地的底层基础不好，它们的贡献也会不好。
指向原始笔记的链接
If the substrate on which your agent’s code lands is bad, their contributions will be bad as well.
—— Jonathan Kelley · [16:15]
就像人类工程师能写出面条式代码一样，智能体也能，只是现在更快了。
指向原始笔记的链接
Just like human engineers can write spaghetti code, so can the agents, but now just faster.
—— Jonathan Kelley · [16:40]
阅读代码一直以来都比编写代码更重要。
指向原始笔记的链接
Reading code has always been more important than writing code.
—— Jonathan Kelley · [17:59]
关于使用编码智能体构建有雄心的软件，我的结语是：代码现在很廉价，但质量不是。
指向原始笔记的链接
So my closing thoughts on using coding agents to build ambitious software is that code is now cheap, but quality is not.
—— Jonathan Kelley · [18:07]
软件工程师的工作从来都不只是把代码行放到屏幕上。
指向原始笔记的链接
The job of a software engineer has never really been about putting lines of code on the screen.
—— Jonathan Kelley · [18:17]
他们拥有的是数百个为他们写代码的 AI 工程师。
指向原始笔记的链接
What they have is hundreds of AI engineers that they code for them.
—— Matteo Franceschetti · [19:51]
所以我们的看法是，做出决策的那部分，也就是定义功能、定义 API 表面积、定义抽象层，那些是非常需要人在环的，我亲自上手和 Claude 一起做。而写代码、执行的部分，才是被自动化的部分。
指向原始笔记的链接
So the way we see it is the part which is where the decisions are being made, right, which is defining the feature, defining the API surface area, the abstraction layer, those are very human in the loop, right, where I’m hands on with Claude. And the code writing part, the execution part, is actually the part which is automated.
—— 嘉宾 · [16:57]
在欧洲的 AI Engineer 大会上，Cursor 做了一个很棒的演讲，讲他们如何用 200 行 agent 文件替换了大约 12000 行 TypeScript 代码来实现类似的东西。
指向原始笔记的链接
So at AI engineer in Europe, Cursor did a great talk on how they replaced roughly 12,000 lines of TypeScript code with 200 lines agent files to create something similar.
—— Philipp Schmid · [15:25]
一个明显的趋势是，随着模型能力提升，我们可以移除编排代码；但如果你的 harness 在模型进步时反而变得更复杂，你很可能是在过度工程化你的 harness。
指向原始笔记的链接
An obvious trend that with better model capabilities, we can remove orchestration code, but if your harness is getting more complex as the model improves, you are most likely overengineering your harness.
—— Philipp Schmid · [16:05]
我们在过去已经见过很多很多次：模型越好，我们能移除的代码就越多，我们需要改变的东西也越多。
指向原始笔记的链接
Like, we have seen in the past many, many times, the better the model gets, the more code we can remove, and the more things we need to change.
—— Philipp Schmid · [17:27]
我们在一旁感叹： 哇， 和我们之前的东西相比， Claude Code 加 Opus 4.5 基本上就是 AGI 了。
指向原始笔记的链接
we on the side were like, wow, Claude Code and Opus 4.5 is basically AGI compared to what we had before.
—— Andrew Qu · [07:50]
Claude Code 并没有被给予一套非常死板规定的工具， 它某种程度上就是被放任自由， 去探索涌现行为。
指向原始笔记的链接
Claude Code was not giving a very prescriptive set of tools. It was sort of just letting it go wild and explore emergent behavior.
—— Andrew Qu · [08:34]
因为 Git 记录的是代码，而不是人类意图。
指向原始笔记的链接
Because Git records the code and not human intent.
—— Kay Malcolm · [03:45]
我学到的教训是：攻击会组合，而代码审查不会，因为我们开发者一次只看一个 diff,但攻击者看的是整个系统，他们会寻找那两个握手并打开一扇门的东西。
指向原始笔记的链接
So the lesson I learned was that attacks compose, code review doesn’t, because us developers all look at diffs one at a time, but attackers look at the whole system and they look for those two things that shake hands and open a door.
—— Sarah Sanders · [08:10]
我相信现在对在座的各位大多数人来说都是如此,基本上从去年年底开始,在我们 iOS 应用中,Claude 共同编写的代码已经超过了团队里的任何一位工程师。
指向原始笔记的链接
So I’m sure this is the case for most of you in the room now, but basically as of late last year, Claude has co-authored more code in our iOS app than any individual engineer in the team.
—— Paula Dozsa · [10:05]
构建这套系统的最大教训是,智能体的上下文主要来自代码库本身,而不是那么来自 Claude MD 文件。
指向原始笔记的链接
And the biggest lesson in building that system is that an agent’s context comes mostly from the code base itself, not so much from the Cloud MD file.
—— Paula Dozsa · [10:33]
然后编码智能体和语音智能体来了，现在他们在给定的一天或一个月内生成的代码比以前多了几个数量级。
指向原始笔记的链接
Then came coding agents and voice agents, and now they generate orders of magnitude more code than they previously did on a given day or month.
—— Charlie Guo · [09:30]
从安全视角看，这是一个非常令人不安的世界，因为突然间你实际上是在一个业务环境里，运行由 LLM 在运行时生成的代码。
指向原始笔记的链接
It’s a very unnerving world from a security lens because suddenly you are literally running code emitted at runtime from an LLM into a business environment, essentially.
—— Samar Abbas · [12:23]
我认为这已经改变了，并且在整个行业迅速改变：随着人类从代码层退出、进入规约层，随着大量编码工作变得完全自动化，代码正越来越多地变成纯粹的实现载体，而不是沟通载体。
指向原始笔记的链接
I think that’s changed and is very quickly changing across the industry as humans get out of that code level into more of that kind of specification level, as a lot of that coding becomes fully automated, is code is turning more and more into purely an implementation vehicle rather than a communication vehicle.
—— Marc Brooker · [40:56]
而且很多团队早期会试图直接跳到最后，说「哦，我要做参数化记忆之类的」——而其实「Clippy 满意并通过所有测试之前不许发布代码」就是一件非常有效、开销非常小的事情。
指向原始笔记的链接
Um and I think a lot of teams early on try and jump to the end and say, oh, I’m gonna do parametric memory or something, when you know you can’t ship code until Clippy is happy with it and it passes all the tests, um, is a very effective thing with very little uh very little um overhead.
—— Marc Brooker · [47:53]
当我们放眼观察，即便是我们自己的、比如客户对智能体产品的使用，人们使用智能体做的事情中，绝大部分——实际上是 80%——他们或许应该使用老式的确定性代码才对。
指向原始笔记的链接
And when we look across like even our own, like in customers usage of agentic products, the vast majority of what people are using an agent for, 80% in fact, they probably should be using actually old fashioned deterministic code.
—— Wade Foster · [15:08]
所以如果你还把 Zapier 想成那种老派的、无代码的、在屏幕上点点框框的东西，我想你会大吃一惊的。
指向原始笔记的链接
So if you still think of Zapier as the old school, no code, clicky click boxes on a screen thing, I think you’re in for a treat.
—— Wade Foster · [63:10]
那种共情，那种认知共情，不仅是人际关系的源代码，它还是民主的源代码，就是这样。
指向原始笔记的链接
That empathy, that cognitive empathy is not only the source code for relationships, it’s the source code of democracy, period.
—— Brené Brown · [09:56]
我们见过的很多其他智能体框架是纵向扩展的——代码看起来越堆越多。
指向原始笔记的链接
I think a lot of other agent frameworks that we had seen, they scaled vertically where they just appeared to get more and more code.
—— Andrew Barba · [48:11]
现在的情况变成了,知道该提示什么、知道该构建什么,比仅仅知道作为 CTO 能够把东西写出来要困难和有价值得多。
指向原始笔记的链接
It’s becoming such that knowing what to prompt and knowing what to build is so much more difficult and valuable than just knowing the CTO being able to code the thing.
—— 嘉宾 · [30:47]
我认为代码审查变成了一种风险管理练习。
指向原始笔记的链接
I think code review becomes an exercise in risk management.
—— Zach Lloyd · [17:04]
会是人类,还是代码?我推断答案是有很多个 9 的概率是代码。
指向原始笔记的链接
Will it be humans or will it be code? And I figured it was many nines of code.
—— Diogo Almeida · [118:25]
写代码很有趣,但真正能产生差距的地方不在那里。
指向原始笔记的链接
Writing code is fun, but it’s not there where you can really make the difference.
—— Daniel Dines · [34:58]
所以如果你写的代码是你能想到的最聪明的代码、最巧妙的东西，那么按定义，你就不够聪明去调试它。
指向原始笔记的链接
And so if you write code that is like the smartest code, the most clever thing that you can come up with, you are by definition not smart enough to debug it.
—— Dex Horthy · [14:12]
如果你在你的软件工厂里交付了坏代码，它会降低未来每一件经过那个工厂的工作的质量，因为模型会看到更多坏模式。
指向原始笔记的链接
if you ship bad code in your software factory, it degrades every future piece of work that goes through that factory because there’s more bad patterns for the model to see.
—— Dex Horthy · [38:30]
我想我们逐渐相信的一件事是:新的无代码就是代码。
指向原始笔记的链接
Well, I think one of the things we’ve come to believe is that the new no code is code.
—— Wade Foster · [65:37]
我们正试图摆脱代码里所有这些 if 语句和分支策略,而是让智能体拥有一些高级工具,能够对无人机下达这些非常高级的指令。
指向原始笔记的链接
We’re trying to get away from having to have all these if statements in the code and these branching strategies and kind of let the agent have some high-level tools to be able to instruct these very high-level commands for the drone.
—— Suchet Bargoti · [19:10]
为什么要生成能生成像素的代码?直接生成像素就好。
指向原始笔记的链接
Why generate the code that generates the pixels? Just generate the pixels directly.
—— Anastasis · [45:14]
我大概八个月没写过一行代码了，因为我只是让 Claude 替我写所有代码。
指向原始笔记的链接
And I haven’t written a line of code in maybe eight months because I just tell Claude to write all my code for me.
—— Zubin Aysola · [14:49]
而智能体应该是真正把制造者的品味注入到代码中的那一个。
指向原始笔记的链接
And agents should be the ones that really take the taste of the maker and put them into code.
—— Roland Gavrilescu · [15:04]
而令人非常惊讶、表现突出的一点是，编程智能体可以一次性完成这件事。它们不需要额外的机器人数据就能用这些代码工作，因为它们已经在如此多的代码数据上训练过了。
指向原始笔记的链接
And what was very surprising, what excelled was that coding agents can do this very one-shot. They did not need additional robot data in order to work with this code because they’re already trained on so much coding data.
—— Ham · [10:11]
但对我来说，只有大约 1% 的代码是完全由 AI 生成的，这并不符合直觉。
指向原始笔记的链接
But it wasn’t intuitive to me that only about 1% of all code was fully AI generated.
—— Daksh Gupta · [03:47]
通过把其他源代码系统和项目跟踪系统中正在发生的活动以 webhook 的形式喂给它，我们开始达到那种自主性水平，我们的工厂在执行产品工程工作，而不仅仅是代码工作。
指向原始笔记的链接
By feeding our web hooks activities that are happening in other source code systems and project tracking systems, we’re starting to get to that level of autonomy where our factory is performing product engineering work, not just code work.
—— Ryan Cooke · [05:07]
给它们一个不会被杀掉的沙箱，在那里它们可以探索你的代码库，可以处理非常困难的任务，并且知道当你合上笔记本盖子时它们不会被关掉。
指向原始笔记的链接
Give them a sandbox that won’t get killed, where they can explore your code base, where they can work on really hard tasks, where they know that they’re not going to get shut down when you close your laptop lid.
—— Charlie Holtz · [08:50]
写代码的智能体总是会告诉你它能正常运行,所以你让它交回一段东西实际运行的录像。
指向原始笔记的链接
The agent that wrote the code will always tell you that it works, so you make it hand back a recording of the thing actually running.
—— Andrew Orobator · [10:41]
模型写代码,我写判断。
指向原始笔记的链接
The model wrote the code, and I wrote the judgment.
—— Andrew Orobator · [12:43]
我现在仍然在频繁地发布产品，但过去六个月里我一行代码都没写。
指向原始笔记的链接
I am still shipping frequently, but I haven’t written a line of code in the last six months.
—— Zach Lloyd · [00:40]
我会让一个智能体先做代码审查，然后随着时间推移，这变成一种风险管理练习，比如，什么时候引入人类来做代码审查？
指向原始笔记的链接
I would have an agent do code review first, and then it becomes, over time, like a risk management exercise of, like, when do you bring in humans to do code review?
—— Zach Lloyd · [11:14]
所以在座的每个人都会写更少的代码，但会发布更多东西，这会是一种权衡。
指向原始笔记的链接
So everyone in here is gonna code less, but they’re gonna ship more, and that’s gonna be a trade-off.
—— Zach Lloyd · [15:43]
软件工厂不只是一个编码智能体，甚至不是一群编码智能体，哪怕上千个也不算——因为生成代码、写代码，相比其他所有环节，是容易的部分。
指向原始笔记的链接
Software Factory is not just coding agent, and it’s not even a swarm of coding agents, even thousands of agents, because generating code, writing code, that’s the easy part compared to all the others.
—— Tereza Tížková · [02:59]
我们实际上发现，这样做最终会有更新鲜的上下文，算是「头脑清醒」——就像人类有其他同事来验证他们的代码一样。
指向原始笔记的链接
We actually found that if you do this, you end up with more fresh context and kind of fresh heads, same with when with humans have like other colleagues verifying their code.
—— Tereza Tížková · [13:04]
重要的一点是，验证者评判的是他们自己没写过的代码。
指向原始笔记的链接
Important thing is the validators judge code that they didn’t write.
—— Tereza Tížková · [13:39]
如果你的代码库没准备好，没有结构化的代码库和所有重要的东西，那么转型成软件工厂反而可能让你变得更糟，让你的代码不断退化。
指向原始笔记的链接
If your code base is not ready, if you don’t have structured code base and all important things, then adopting or turning into a software factory can actually make you end up worse and make your code degrading.
—— Tereza Tížková · [17:34]
AI 会复制你代码库中已有的模式。如果你有五种不同的做事方式，AI 会变得非常困惑，然后创造出第六种、第七种做事方式。
指向原始笔记的链接
AI copies existing patterns that exist in your code base. If you have five different ways of doing something, AI will become very confused and it will create six ways or seven ways in ways of doing it.
—— Liz Fong-Jones · [11:13]
因为机器永远不会感到无聊，让它们给你的代码库添加遥测是一件很棒的事情
指向原始笔记的链接
because the machines never get bored, it’s a great thing to do to have them add telemetry to your code base
—— Liz Fong-Jones · [13:24]
现在你的代码库就被窃取泄露出去了，因为这个智能体被提示词注入了，而它对你的数据有这么多访问权限。
指向原始笔记的链接
And now you’ve exfiltrated your code base out because the agent has been prompt injected and it has all this access to your data.
—— Thariq Shihipar · [56:53]
不要抵押你的代码库,指望一年后会出一个聪明到能修复你所有问题的模型。
指向原始笔记的链接
Do not mortgage your code base in the hopes that in a year the model will come out that’s smart enough to fix all your problems.
—— Dex Horthy · [01:08]
这是一个完整的基准测试,专门用来证明模型会让代码库随着时间推移变得越来越糟。
指向原始笔记的链接
It’s an entire benchmark designed around proving that models make code bases worse over time.
—— Dex Horthy · [02:38]
疯狂的事情是这样的:在 Claude 3.5、3.7,大概 Opus 4 那个时代,我们经常这么做——好,是的,我知道代码很潦草,但等到这个问题真正要紧的时候,那就是 GPT-7 的问题了。
指向原始笔记的链接
Well, so here’s the crazy thing is like we used to do this thing in the like Claude 3.5, 3.7, maybe like Opus 4 days where it’s like, cool, we’re just gonna, yes, I know the code is sloppy, but by the time that matters, that’ll be GPT-7’s problem.
—— Dex Horthy · [06:08]
你没法捕捉所有东西,而且真相之源应该是代码,而不是某个规格说明或计划。
指向原始笔记的链接
You can’t capture everything, and the source of truth should be the code, not some specification or plan.
—— Dex Horthy · [09:56]
我从来没有见过一个提示词或一个确定性工具,能够在没有人类参与的情况下,以任何有意义的方式驱动代码库的可维护性并对抗模型写出不可维护代码的倾向。
指向原始笔记的链接
But I have never seen a prompt or a deterministic tool that can drive code-based maintainability and fight the model’s tendency to write unmaintainable code in any meaningful way without a human in the loop.
—— Dex Horthy · [27:15]
我的建议一直是,不要拿你的代码库去抵押,指望一年后、六个月后或十八个月后会有一个聪明到能解决你所有问题的模型发布出来。
指向原始笔记的链接
And my advice is always like, do not mortgage your code base in the hopes that in a year or six months or 18 months a model will come out that’s smart enough to fix all your problems.
—— Dex Horthy · [30:27]
如果你每一千行就检查一次,交付一万行代码会容易得多。
指向原始笔记的链接
Because like, it’s a lot easier to ship 10,000 lines of code if you check it every thousand lines.
—— Dex Horthy · [39:31]
开启自主智能体的开发者多写了 741% 的代码，但实际发布的软件只多了 30%。
指向原始笔记的链接
The developers who turned on autonomous agents wrote 741% more code, but only 30% more software shipped.
—— Laurie Voss · [01:47]
该研究指出，评审者一次读超过 400 行代码就基本发现不了缺陷，一小时超过 450 行效率就完全断崖式下跌。
指向原始笔记的链接
And the study says that reviewers stop finding defects effectively if they try to read more than 400 lines of code in one sitting and their effectiveness completely falls off a cliff if they try to review more than 450 lines of code in an hour.
—— Laurie Voss · [03:54]
在她谈论这件事的文章里，她讲清楚了模型为什么在生成代码上进步得这么快——因为编译器是一个免费的验证器。
指向原始笔记的链接
In the essay where she was talking about this, she made clear why models got so good so fast at generating code, and that is because a compiler is a free verifier.
—— Laurie Voss · [09:36]
所以对拉取请求的机器评审，已经是全球最大代码托管平台上的主流默认做法。
指向原始笔记的链接
So machine review of pull requests is the mainstream default on the world’s largest code host.
—— Laurie Voss · [11:24]
模型倾向于看着代码说，嗯，这在我看来不错，所以就发布吧——而这恰恰是人类在那种情况下会做的事。
指向原始笔记的链接
The model tended to look at the code and say, well, that looks good to me, so ship it, which is exactly what a human would do in that situation.
—— Laurie Voss · [12:54]
他说，我错了。请、请去读代码。我们有大约六个月的时间尝试不读代码。结局并不好。我们不得不把那个系统的大部分拆掉重做。
指向原始笔记的链接
He said, I was wrong. Please, please read the code. We tried not reading the code for like six months. It did not end well. We had to rip out and replace large parts of that system.
—— Laurie Voss · [19:01]
但正在发生的是，人类的角色并没有消失，它正在向栈的上层移动，可能移动了栈的好几层——从直接检查代码，转到设计和调优那些检查代码的系统，以及设计「好」的定义。
指向原始笔记的链接
But what’s happening is the human role isn’t disappearing, it is moving up the stack, possibly several levels of the stack, from inspecting the code directly to designing and tuning the systems that inspect the code and designing the definition of good.
—— Laurie Voss · [20:21]
所以自动化审查会栽在那些以自信姿态包装的坏代码上，而以自信姿态包装的坏代码恰恰是智能体非常擅长生产的那种代码。
指向原始笔记的链接
So automated reviews fall for confidently framed bad code, and confidently framed bad code is exactly the kind of code that agents are very good at producing.
—— Laurie Voss · [21:33]
一旦合并前的审查全部由机器完成，观察代码实际做了什么就成了最后一个屹立不倒的审查者。
指向原始笔记的链接
Once the pre-merge review is all machines, watching what the code actually does becomes the last reviewer standing.
—— Laurie Voss · [22:07]
而未来几年胜出的团队，不会是生成最多代码的团队，而是那些能够拿出证据说明他们为什么信任自己所发布内容的团队。
指向原始笔记的链接
And the teams that win the next few years won’t be the ones that generate the most code, they’ll be the ones who can say with evidence why they trust what they shipped.
—— Laurie Voss · [23:15]
即使工程师从模型那里得到的是 100% 准确的即时代码——事实并非如此——你仍然只是在攻击整个价值流中大约 14% 到 16% 的部分。
指向原始笔记的链接
Even if engineers are getting like 100% accurate instant code coming from the models, which they are not, you would still only be attacking anywhere from maybe 14 to 16% of the overall value stream.
—— Justin Reock · [15:04]
这背后有多种原因，但确实值得一提：每多一行代码，就多一个潜在的 bug、一个潜在的漏洞。
指向原始笔记的链接
So there’s multiple reasons for this, but it really bears mentioning that every extra line of code is a potential bug, a potential vulnerability.
—— Justin Reock · [08:00]
现在就像，好吧，现在先把东西弄出来，而被生成的代码本质上会是平庸的，因为这是平均律。
指向原始笔记的链接
Now it’s like, well, now let’s get the thing out and the code that’s being generated is gonna be essentially mediocre because it’s law of averages.
—— Justin Reock · [07:34]
我想说，直到最近，Meta 的货币是代码。
指向原始笔记的链接
I would say until quite recently, the currency at Meta was code.
—— Moritz Beller · [04:55]
模型在编写和调用代码方面比调用工具更好，因为工具调用基本上是一种人工构造，我们必须教会 LLM 去做这件事。
指向原始笔记的链接
Models are better at writing and calling code than calling tools because tool calling is an artificial construct, basically, that we have to teach the LLMs to do.
—— Jan Čurn · [06:28]
我们所相信的是,仅仅阅读代码已经不够了。
指向原始笔记的链接
What we believe is that reading code is not enough anymore.
—— Zilvinas · [11:01]
人们对被推入生产的代码的理解却更少了。
指向原始笔记的链接
People have less understanding of the code that’s being pushed into production.
—— Eric Schwartz · [02:11]
像 Claude Code、Codex 和 Cursor 这样的工具非常出色,真的很出色,但它们会带来更多代码和更多复杂性。
指向原始笔记的链接
Tools like Claude Code and Codex and Cursor, which are fantastic, they are fantastic, but they result in more code and more complexity.
—— Eric Schwartz · [02:58]
如果你有穷尽式的验证，你就能以你的智能体写代码的速度发布代码。
指向原始笔记的链接
if you have exhaustive verification, then you can ship code at the speed that your agents write it.
—— Gabriel Spencer-Harper · [00:45]
通常，按我们带立场的观点来说，代码覆盖率对世界上除 Meticulous 之外的每一个工具来说都是一个糟糕的指标。
指向原始笔记的链接
Typically, I would say that code coverage in our opinionated view is a terrible metric for every tool in the world apart from meticulous.
—— Gabriel Spencer-Harper · [09:26]
我们用确定性代码的时候日子有多好过。就像，天哪，我还记得计算机完全按我们说的去做的那些日子。
指向原始笔记的链接
how good we’ve had it with deterministic code. Like, oh my god, I remember the days when computers did exactly what we told them to do.
—— Amjad Masad · [39:15]
这就是 AI 发挥作用的地方——最终更快地学习、交付更好的产品，而这是很多人在写更多代码时错过的东西：运用产品品味，运用那种学习。
指向原始笔记的链接
And that’s where AI kicks in to actually learn faster in the end and shipping better products, which a lot of people are missing out as they write more code is applying the product taste and applying that learning.
—— Harald Kirschner · [04:34]
对我们来说，TypeScript Go 让构建获得了 10 倍提升——在智能体运行自动化 PR、需要 CI-CD 循环来获取反馈改进代码的规模下，这是一个巨大的提升。
指向原始笔记的链接
And for us, TypeScript Go was a 10 times improvement in our builds, which at the scale of automatic PRs running with agents and them needing that CI, CD loop for feedback to improve the code is a massive improvement.
—— Harald Kirschner · [07:36]
你如何搭建你的技能、你的环境、你的代码库，我认为这些最终才是打造产品的新食材。
指向原始笔记的链接
And how you set up your skills, your environment, your code base, I think are ultimately the new ingredients that go into building product.
—— Poteto · [15:00]
你会在我构建的很多其他技能里看到这一点，就是我尝试把确定性的部分提取出来变成代码，只把真正需要判断力的部分留给智能体。
指向原始笔记的链接
And you’ll see this in a lot of the other skills that I’ve built, is like I try to extract out the deterministic parts and turn that into code and just leave only the parts that actually require judgment to the agent.
—— Poteto · [21:46]
关于这一点，我要说的最后一个大要点是，我实现 2000 个 PR 的大突破，是从问题出发倒推：我怎样才能达到让我的智能体自己合并自己的代码的地步？
指向原始笔记的链接
And really the big, the last thing I’ll say to this is like the big unlock for me for getting to 2,000 PRs is starting from the question and working backwards of how do I get to the point where my agent can merge its own code?
—— Poteto · [39:37]
现在你的重点不仅仅是构建一个能完成工作的好的智能体系统，你可能还想开始考虑编写自愈代码，考虑创建一个能审视自身、查看有哪些 bug 并自我修复的智能体。
指向原始笔记的链接
So now your focus is not just building a good agentic system that can get the job done, you would also maybe want to start thinking about healing, writing self-healing code, about creating an agent that can sort of go over itself, look at what bugs are there and fix itself.
—— Sandhya Subramani · [14:20]
一股代码的巨浪正在向你袭来。
指向原始笔记的链接
A tidal wave of code is coming for you.
—— Dan Adler · [02:12]
但与此同时，实际发生的是，这些代码库，这些维持世界运转的庞大代码库，正在开始腐化。
指向原始笔记的链接
But meanwhile, what’s happening is that these code bases, these massive ones that keep the world running, are beginning to decay.
—— Dan Adler · [02:56]
正是那些让我们所有人提速、让我们比以往任何时候都更快写出更多代码的工具，也正是在制造那些会导致这些支撑我们世界运转的庞大代码库开始崩溃的条件。
指向原始笔记的链接
The very tools that are speeding all of us up and letting us write more code faster than ever are also the tools that are creating the conditions that will cause these massive code bases that run our world to begin to fail.
—— Dan Adler · [04:15]
当然，Claude Code 可以做出这个修改，但我有 9 万个仓库要做这个修改。
指向原始笔记的链接
Sure, Claude Code can make this change, but I have 90,000 repositories to make it in.
—— Dan Adler · [06:45]
它真的会生成一万行代码，没有人能检查，也没有人知道它到底是不是在做对的事情。
指向原始笔记的链接
it generates literally 10,000 lines of code that nobody can inspect and nobody knows if it’s actually doing the right thing.
—— Jan Oberhauser · [04:37]
他们花太多时间找文件，而不是写代码、做自己真正喜欢做的事。
指向原始笔记的链接
They’re spending too much time looking for the files instead of writing code and doing things that they really like to do.
—— Jeff Koss · [05:18]
别只让它替你写代码。
指向原始笔记的链接
Don’t just let it code for you.
—— Matt Brockman · [18:29]
我们看到很多智能体在写代码，即使它们不是编码智能体。
指向原始笔记的链接
And we see a lot of agents writing code, even if they’re not coding agents.
—— 嘉宾 · [12:10]
我由衷地相信，我们会拥有主动联系其他智能体来解决问题的智能体，甚至更进一步，去部署代码来解决那些环节上的问题。
指向原始笔记的链接
I wholeheartedly believe that we will have agents that are reaching out to other agents to solve problems and then even take it a step further, go and deploy code to solve the problem on those pieces.
—— Dan O’Connell · [48:19]
我是说,如果你用编码智能体写代码,很多情况下,你甚至不知道里面有什么。
指向原始笔记的链接
I mean, if you write your code with a coding agent, in many cases, you don’t even know what’s inside it.
—— Alon Schindel · [32:21]
在某些情况下，人们甚至不再亲手写任何一行代码了。AI 实际上在写你的所有代码，审查你的所有代码，而你只是在引导这些智能体去做这些事。
指向原始笔记的链接
In some cases, people are not even writing any lines of code anymore. AI is actually writing all of your code, reviewing all of your code, and you’re just sort of guiding these agents to doing that.
—— Ben · [11:44]
而我愿意这样想：总有一天智能会像数据库一样，当你在代码中需要智能时就直接调用它。
指向原始笔记的链接
And I like to think of it as like one day intelligence will be like databases where you just call it when you need it, when you need intelligence within your code.
—— Diogo Almeida · [74:25]
你不能把自己的思考委托出去、把它当作拐杖，因为那样你会得到这些庞大的、你根本解释不了的垃圾代码库。
指向原始笔记的链接
You can’t delegate your thinking away and basically rely on it as a crutch because then you get these massive slop code bases that you can’t actually explain.
—— Yash · [48:59]
我们总是可以这样拆解：服务图是正在发生什么，代码图是它如何发生以及为什么发生，基础设施图是它在哪里发生。
指向原始笔记的链接
So we can always break it down as saying, you know, the service graph is what is happening. The code graph is how and why it’s happening. The infograph is where it’s happening.
—— Matt Jones · [46:34]
你基本上是在依据代码自己怎么想的来判断，而不是它实际在做什么。
指向原始笔记的链接
You’re kind of going on what the code thinks, not what it does.
—— Matt Jones · [46:31]
我认为人们缺乏创造力，无法理解在今天的 Claude Code 订阅和按 token 定价模式之外，回报还能如何被看待。
指向原始笔记的链接
I think people lack the creativity to understand how returns can be viewed beyond the Claude Code subscription and per token pricing regime of today.
—— 嘉宾 · [15:11]
我们碰巧通过代码让那个程序运行起来，代码其实是我们如何协作的一个副产品。
指向原始笔记的链接
The fact that we happen to make that program work through code, the code is really a side effect of how we work together.
—— Sam Newman · [17:23]
如果你只是不看代码，而且也不在意那些其他的事情，那么接下来发生在你身上的任何事都是你活该。
指向原始笔记的链接
If you’re just not looking at the code and you’re not caring about those other things, well, then you deserve everything that’s going to happen to you.
—— Sam Newman · [40:33]
根据我们的数据，现在每个开发者的代码行数比以前多了 218%。
指向原始笔记的链接
According to our data, it’s 218% more lines of code per developer than it used to be before.
—— Eli Cohen · [01:49]
我现在可以把一堆 CLI 命令硬编码,把我的 API key 硬编码进 markdown 文件,然后称之为进步,对吧?
指向原始笔记的链接
I can now hard-code a bunch of CLI commands, hard-code my API key inside markdown file, and call it a progress, right?
—— Rafal Wilinski · [07:38]
其中最重要的一点是，代码已经在磁盘上、在本地了，尤其是如果你的智能体有上下文的话，那为什么要用一个预检索系统把它复杂化呢？
指向原始笔记的链接
And the largest part there is code is already on disk and local, especially if your agent has context, so why complicate it with a pre-retrieval system?
—— George He · [03:49]
你没法真正对一个公司数据集的文件夹做 grep，因为它不像代码那样组织有序。
指向原始笔记的链接
You can’t really grep the folders across a company dataset because it’s not organized like code.
—— George He · [04:13]
任何涉及设计或架构的东西,你还是需要一个人类,因为智能体只会在你的代码库里局部地乱搞一通,把事情弄得更糟。
指向原始笔记的链接
For anything that concerns the design or architecture, you still kind of need a human because the agent will just locally shit into your code base and make things worse.
—— Mario Zechner · [04:01]
它几乎不可能在不增加代码的情况下修复一个问题,这看起来很奇怪,因为我认为很多 issue 实际上本不应该让代码库净增加代码。
指向原始笔记的链接
It’s almost impossible that it will fix a problem without adding code, which seems weird because I think a lot of the issues actually shouldn’t be net additive to the code base.
—— Armin Ronacher · [04:17]
工程师正在从敲代码转向治理智能体。
指向原始笔记的链接
Engineers are going from typing code to governing agents.
—— Davis Palmie · [02:07]
如果你拉取你的 Dora 指标，从代码写完到进入生产环境的时间，我敢打赌它远远超过实际做那些工程改动所花的时间。
指向原始笔记的链接
If you pull your Dora metric, the time from code written to time in production, I would bet dwarfs the time actually making those engineering changes.
—— Davis Palmie · [05:38]
举例来说，我们发现你在代码审查上用 GPT 5.2 可以达到和最新 Opus 模型相同的性能，但价格只有一半。
指向原始笔记的链接
We found, for instance, that you can achieve the same performance on code review with GPT 5.2 as with the latest Opus models, but at half the price.
—— Davis Palmie · [08:25]
但这是不需要 50 万行你需要搬过来的 Python 或 TypeScript 代码的 loop engineering,那样你的智能体才能真正互相提示工作。
指向原始笔记的链接
But it’s loop engineering without 500,000 lines of a Python or TypeScript code that you need to bring over so your agents can actually prompt each other for work.
—— Vlad Luzin · [15:45]
智能体优先，智能体写所有的代码。你只是在管理一个智能体团队，并驱动它们。
指向原始笔记的链接
Agent first, the agents write all of the code. You’re just managing a team of agents and driving that.
—— Matt Garman · [53:22]
所以我认为我们应该终结这场辩论了，对于 90% 的使用场景来说，你用的是 cloud code 还是外面的 codex 都不重要。
指向原始笔记的链接
So I think we should end this debate like for 90% of the use cases, it does not matter if you are using cloud code or the codex out there.
—— Ankit Shukla · [35:24]
创建代码的成本在下降，但理解一个软件程序的成本并没有。
指向原始笔记的链接
The cost of creating code is falling, but the cost of understanding a software program is not.
—— Gabriel Martinez · [03:13]
AI 可以生成代码，但它无法神奇地给一个混乱的系统带来连贯性。
指向原始笔记的链接
AI can generate code, but it cannot magically give a messy system coherence.
—— Gabriel Martinez · [03:49]
如今大部分代码是由 Astra 写的。
指向原始笔记的链接
The majority of the code these days is written by Astra.
—— Tibo Sottiaux · [10:27]
我们其实不认为瓶颈在于产出更多代码。
指向原始笔记的链接
We actually don’t think the bottleneck is in producing more code.
—— Arvind Srinivas · [42:38]
② 出现在这些集
1 集
③ 关联
点进去有真内容 —— 本页主要出口
Nico · Dylan Field · Figma · GitHub · Notion · Slack · 智能体 · 提示词 · 代码层 · Make