Notion's Token Town — Sarah Sachs, Notion
频道: AI Engineer
视频: https://www.youtube.com/watch?v=-I5W5QVAT8E
原文语言: en
统计: 共 62 轮
[0:01]
[music] Okay. Hello. Okay. Before I get started, you guys, this is a huge keynote room. Can everyone like come forward because I'm talking to like four empty rows and disperse people. Do me a favor. I'm spending 30 minutes telling you all of our secrets. I can see you still. Thank you. Thank you. Thank you. Thank you.
[音乐] 好，大家好。好的。在我开始之前，各位，这是个超大的主会场。大家能不能往前坐一点？因为我面前是四排空座位，然后零零散散坐着一些人。帮我个忙。我要花 30 分钟把我们所有的秘密都告诉你们。我可还是能看见你们哦。谢谢，谢谢，谢谢，谢谢。
[0:38]
We're just going to chat. It's a giant room and there's 500 of us. This room is way larger than that. Thank you. Honestly, I knew you guys had it in you. It's really not so hard. Thank you. I also sit in the back. I also work during talks. I get it. I totally get it. I did all day, but not for me. Okay, I'm gonna start, but I'm gonna still point at you if you're in the back like you. Okay,
我们就当是随便聊聊。这房间这么大，我们有 500 个人呢。这屋子可比那大多了。谢谢。说实话，我就知道你们能做到。这真的没那么难。谢谢。我自己也会坐在后排。我在别人演讲的时候也会处理自己的工作，我懂的，我完全懂。我今天一整天都是这样，但轮到我讲的时候可不行。好，我要开始了，不过要是你坐在后面，我还是会点你的名，就说你呢。好，
[1:03]
I'm Sarah. I'm lead our engineering teams for AI at Notion. Um, welcome to my talk. Um, it's about token town. How do you go from not go from AI pled to AI poor? Okay. Um, I know that today is all about software factories. We're going to talk about that, but we're going to talk about how to do it sustainably.
我是 Sarah，我在 Notion 带领我们的 AI 工程团队。欢迎来听我的演讲。这场演讲叫 Token Town（代币小镇）。讲的就是：你怎么才能不从「AI 狂热」一路沦落到「AI 破产」？好。我知道今天大家都在聊 software factories（软件工厂）。我们会聊这个，但我们要聊的是怎么用可持续的方式去做这件事。
[1:25]
This is me. This is on my first day at Notion in a very sweaty subway. Um, like I said, I lead our AI teams at Notion. Um, and I negotiate AI contracts for a living. My team jokes I act like Anna Winter. So, this is a nice a nice image of me with AI Anna Winter hair after a press article referred to me that externally. Um, and that's kind of the idea, right? Uh, how do you think about negotiating between different vendors?
这就是我。这是我在 Notion 上班第一天，在一节特别闷热的地铁里拍的。就像我说的，我在 Notion 带领我们的 AI 团队，而且我是靠谈 AI 合同为生的。我的团队开玩笑说我做起事来像 Anna Wintour。所以呢，这是一张挺不错的图——有篇媒体报道在外面这么形容我之后，我给自己配上了 AI 生成的 Anna Wintour 式发型。这其实就是核心思路，对吧？你要怎么去思考在不同 vendor 之间做谈判这件事？
[1:52]
um making sure that you maintain taste for your company. I don't do it alone. Um this is launch day of one of our recent launches. This is just a subset. Any good engineering manager points out that we have a whole company of people building this. I'm just the one that gets to come talk to you about it. So,
要确保你为公司守住那份品味。这件事我不是一个人做的。这是我们最近一次发布的当天，而且这还只是其中一部分人。任何一个称职的工程经理都会指出，是我们整个公司的人一起在做这件事，我只不过是那个有机会来跟你们讲这件事的人。所以，
[2:08]
we've been building a lot. Um this is um an example of our AI usage um just in 2026. Um and we've been really proud of how we've been able to grow that usage. and I'm going to talk to you about how you can build an AI native product and an AI native company. Um, but this is just to give me some credit that that we're doing it kind of well. Okay, so for those of you that don't know,
我们做了很多东西。这是我们 AI 使用量的一个例子，单看 2026 年。我们对自己能把这个使用量做起来感到非常自豪。我接下来会跟你们讲，你要怎么打造一个 AI native 的产品、一个 AI native 的公司。不过放这张图，也算是给我自己长长脸，证明我们这事儿做得还算不错。好，那对于不了解的朋友来说，
[2:31]
notion's always been that durable system of record. It's always been the place where you can collaborate with your peers. Um, but today that point of collaboration is a little bit different. It's not just humans. Notion's always been the place for collaboration. And today that collaboration happens between humans and agents. Humans and humans,
Notion 一直以来都是那个 durable system of record（持久的记录系统）。它一直是你可以和同事协作的地方。但今天，协作的形态有点不一样了，参与协作的不只是人类。Notion 一直是那个用来协作的地方，而今天，这种协作发生在人类和 agent 之间、人类和人类之间、
[2:48]
agents and agents. And we like to think about AI transformations going through this journey. And I'm sure some of you are looking at the slide and wondering where you are. AI as a thought partner is when we all started tinkering. We all started just going to the very first version of chat on Thanksgiving when it came out three years ago, four years ago. And we started saying like how can I send this email to my landlord
agent 和 agent 之间。我们喜欢把 AI 转型想象成会经历这样一段旅程。我相信你们里有些人正看着这张幻灯片，在想自己处在哪个阶段。「AI 作为思维伙伴」，就是我们大家开始瞎捣鼓的那个阶段。我们都是从最早那一版的 ChatGPT 开始的，就是三四年前的感恩节它刚出来的时候。我们开始问它，比如：我要怎么给我房东写这封邮件，
[3:15]
to say that I shouldn't pay for repainting right then we'd copy paste it enter it into our email eventually we started getting to a place where we could use AI like an assistant AI was able to maybe execute individual tasks that's how notion AI really took off in the beginning um and it was able to save employee time but functionally was limited in its capabilities based on what humans asked it to do AI as teammates is what we were really excited
跟他说我不该为重新粉刷这事儿买单，然后我们会把它复制粘贴到自己的邮件里。慢慢地，我们走到了一个可以把 AI 当助手用的阶段，AI 也许能够执行一个个单独的任务，Notion AI 一开始真正火起来靠的就是这个。它能帮员工省时间，但功能上还是受限的，得看人类让它做什么。而「AI 作为队友」，才是我们真正激动、
[3:40]
to launch almost a year ago now. Um, but this is true in many products where we can do repetitive work and think about a process and have AI do that process. What I think is really interesting is when AI actually becomes that critical workflow where processes are interfacing with each other and you have entire systems running. How many of you guys feel like you have AI as a system down?
在差不多一年前推出的东西。这在很多产品里都成立——就是我们可以处理重复性的工作，去梳理一个流程，然后让 AI 来跑这个流程。我觉得真正有意思的，是当 AI 真的变成那个关键工作流的时候：各个流程彼此对接，你手上有一整套系统在运转。你们当中有多少人觉得自己已经把「AI 作为系统」这一层搞定了？
[4:03]
Aren't you sad you came up now? Great. None of you. Exactly. We have found that no one has figured out how to do this well. 88% of people can't even get past AI as an assistant. And why is that? We have a thesis at notion. It's because there's too much siloed data and not a sturable system of record for that point of collaboration. And we believe that for your software factory to work, for your company to work, and for your
你们现在是不是有点后悔往前坐了？很好，一个举手的都没有。正是如此。我们发现，没有人真正搞明白怎么把这件事做好。88% 的人甚至连「AI 作为助手」这一关都过不了。这是为什么呢？我们在 Notion 有一个观点：是因为数据太过孤岛化，而且缺少一个 durable system of record 来支撑那个协作的节点。我们相信，要让你的 software factory 跑起来、要让你的公司跑起来、要让你的
[4:30]
systems to work, you need that durable system of record. And that is notion's mission. So doing that is expensive. Um you see a lot of companies that try and commit themselves to this vision and these are just a series of headlines all within a week of how that's painful. So you can put all of your money into a process to try and make a system and you end up feeling like this,
各套系统跑起来，你就需要那个 durable system of record。而这正是 Notion 的使命。可是做这件事很烧钱。你会看到很多公司试图全力押注这个愿景，而这些只是一周之内的一连串新闻标题，都在讲这有多痛苦。你可以把所有的钱都砸进一个流程里，想搭出一套系统，结果最后你的感受就像这样，
[4:55]
right? You end up using a blowtorrch to light what is actually a large cigar. But you kind of get the idea. Cost is a structural barrier to entry. It makes it hard for you to serve products. It makes it hard for you to build factories. And it is ultimately, I would posit, one of the largest reasons why things do not happen at scale successfully today. And I would argue for anyone working in an applied AI company, it's something for them to be
对吧？你最后是拿着一把喷枪去点一支其实很大的雪茄。不过你大概懂我的意思。成本是一道结构性的进入门槛，它让你很难去支撑产品，很难去搭建工厂。而我要说，它归根结底是当今很多事情之所以无法成功规模化的最大原因之一。我认为，对任何在应用型 AI 公司工作的人来说，这都是他们应该
[5:23]
really familiar with to understand the trade-offs that they're making to build durable and exciting and enlightening product for their customers. But that's not really how the market is today, right? I'm not going to name names here, but you guys have search engines. You can figure it out. Exhibit A. A reasoning model gets upgraded.
非常熟悉的东西——好去理解他们为了给客户打造持久、令人兴奋、给人启发的产品，所做的种种取舍。但市场今天并不是这样运作的，对吧？我在这儿就不点名了，但你们都有搜索引擎，自己能查出来。第一个例子。一个 reasoning model 升级了。
[5:43]
Amazing. The per token pricing is the same. What's not to love, you try it out, it uses three times as many output tokens, right? Exhibit B. A model gets upgraded, but it has an entire new digit, right? Whatever demarcation system that model family likes, it's brand new. It's 40% more than its predecessor, which is being deprecated in the next four months. These are real scenarios that we face at notion. All of you are nodding because these are common
太棒了。per-token 的价格还是一样的，这有什么不喜欢的呢？结果你一试，它用掉的 output token 是原来的三倍，对吧？第二个例子。一个模型升级了，但它的版本号整个进了一位数，对吧？不管那个模型家族喜欢用什么样的编号方式，反正是全新的。它比它的上一代贵 40%，而上一代在接下来四个月里就要被弃用了。这些都是我们在 Notion 真实面对的场景。你们都在点头，因为这些现在基本上
[6:13]
pretty much monthly now. But here's the problem. Are you growing 40% in that time period? Are you making 30 3x more revenue? No. So, how do you navigate the system? If you just auto upgrade your model and everything that you're doing, you're you're giving someone a bad deal.
差不多每个月都在上演。但问题来了：你在这段时间里增长了 40% 吗？你的营收翻了三倍吗？没有。那你要怎么在这套体系里周旋？如果你只是无脑地把你所有在用的模型全都自动升级，那你就是在让某个人吃亏。
[6:31]
Either your customers or your investors, depending on how you charge and where you get your money, neither are good. Fortune 5 million companies have the capability to navigate this. They can hire large consulting teams, have durable teams on their own, and build expertise on how to navigate these trade-offs.
要么是你的客户吃亏，要么是你的投资人吃亏——取决于你怎么收费、你的钱从哪来——但这两种都不是好事。财富 500 强（Fortune 500）公司有能力应对这种局面。他们可以雇庞大的咨询团队，自己养着长期的团队，积累起如何权衡这些取舍的专业能力。
[6:51]
Um, most people don't. everyone else has no ability to negotiate with leverage and they're stuck in these scenarios, right? Part of my job as that Anna Winter joke is to think about advocating for the Fortune 5 million, the non-fortune 500 companies that don't have the mass to have leverage and negotiate, but need to think about how.
但大多数人做不到。其他所有人都没有能力带着筹码去谈判，只能被困在这些处境里，对吧？回到那个 Anna Wintour 的玩笑，我工作的一部分，就是去为那「五百万强」发声——也就是那些非财富 500 强的公司，它们没有那个体量去获得筹码、去谈判，但同样需要好好想想该怎么办。
[7:12]
And I'm going to share some of the lessons that I've learned when I have kind of large amounts of traffic behind me that I think scale to those who don't. Um, this is probably less of a secret now than it was when I started giving talks like this u maybe four months ago.
我会分享一些我自己学到的经验。这些经验是我在背后有大量流量做支撑的情况下总结出来的，但我认为它们同样能适配到那些没有这种流量的人身上。这些经验现在可能已经不像我大概四个月前开始讲这类演讲时那么算是秘密了。
[7:30]
Um, your supplier is your competitor. I know very few people who have convinced me that that's not true. Um, you will always be getting a bad deal on tokens with someone who builds them natively, right? Sometimes the cost of goods served is extremely different. you're basically they're serving a first-party product and then you're buying those tokens at a huge search charge and then selling them again at another search
你的 supplier 就是你的竞争对手。几乎没有人能说服我这句话不成立。在 token 这件事上，你从那些自己原生做 token 的人手里拿到的，永远是一个糟糕的交易，对吧？有时候 cost of goods sold（销货成本）差得非常离谱。本质上，他们卖的是 first-party 产品，而你要以一个巨大的 markup 去买这些 token，然后再加一道 markup
[7:53]
charge. Um, that's not really value you can defend. You're getting a really bad deal. And if you tie yourself to one provider, you have no exit. If you build an AI product that you're selling with this structure, you are crossing your fingers and hoping that you are a viable business. I do not encourage that.
卖出去。这并不是你真正能守得住的价值，你拿到的是一个非常糟糕的交易。而如果你把自己绑死在一家 provider 身上，你就没有退路了。如果你用这种结构去打造并售卖一个 AI 产品，那你就只是在祈祷、指望着自己还能是一门活得下去的生意。我不鼓励这样做。
[8:14]
This is really interesting. Dylan in some analysis posted this. I think it's it says eight hours ago. It wasn't at this point. It was probably a month ago. Um they purchased a subscription plan and they just highlighted right how different what Frontier Labs charge customers for first-party products are versus what they sell. It's a bad deal.
这个特别有意思。Dylan 在一份分析里发了这个。我看上面写着「8 小时前」，不过他发的时候其实不是这个时间点，大概是一个月前。他们买了一个订阅套餐，然后就直接点了出来——Frontier Labs 针对 first-party 产品向客户收的价，和他们卖 token 的价，差别有多大。这是一个糟糕的交易。
[8:34]
Don't play this game or try and let me know how you win. I don't recommend. Think about everyone else. Think about what that structure means and where you have expertise. I don't think that that's winning on the token economics. I think it's about product. It's about building data flywheels and understanding your customers better than anyone else.
别玩这个游戏，或者你要玩的话，回头告诉我你是怎么赢的。我不推荐。想想其他所有人，想想这种结构意味着什么，以及你的专长到底在哪里。我不认为取胜的关键在于 token economics（代币经济学）。我认为关键在于产品，在于打造数据飞轮，在于比任何人都更懂你的客户。
[9:00]
Understanding when you need capability, when you need low price, when you need latency improvements. I promise you, you don't always need what is usually the slowest but the most capable model out there. and then build compelling UI and orchestration. And I'll show you some examples of that to justify the cost on the bad deal tokens that you do resell.
要弄明白什么时候你需要能力，什么时候你需要低价，什么时候你需要改善延迟。我向你保证，你并不总是需要那个通常最慢、但最强的模型。然后，去打造有吸引力的 UI 和 orchestration（编排）。我会给你们看几个这样的例子，好为你转卖出去的那些「亏本」token 所对应的成本正名。
[9:23]
The job is not to train. I mean, some of you might be training the best model and I'd love to serve it and come talk to me afterwards, but most of you are not doing that. Stop trying to win that game and think about the best product that uses many models. Help your customers,
你的任务不是去训练模型。我是说，你们当中也许有人正在训练全世界最好的模型，那我很乐意用它，会后来找我聊；但你们大多数人并不是在做这个。别再想着赢那场游戏了，去想想怎么打造一个用上多种模型的最好的产品。去帮助你的客户，
[9:37]
help your team. Bet on the frontier, not on the lab. and we'll talk about what it looks like to do that. This cost per capability per second trade-off is actually really intense. Um, Citadel came out with this memo um a while ago, maybe two weeks ago. I loved it. The idea is that for the economy at large, simpler models might be the most cost-effective productivity augmenting pathway. They talk about this bifurcation on frontier versus everyday
去帮助你的团队。押注在 frontier（前沿）上，而不是押在某一家实验室上。我们会聊聊具体怎么做到这一点。这种「每单位能力、每秒成本」的取舍，其实非常激烈。Citadel 前段时间发了一份备忘录，大概两周前吧，我很喜欢。它的观点是：对整个经济体而言，更简单的模型也许才是最具性价比、最能提升生产力的路径。他们谈到了 frontier（前沿）和日常使用之间的这种分化，
[10:07]
usage. I really believe that. And for every product, the definition of frontier versus everyday, the definition of saturated capabilities or model capability overhangs depends on your expertise on your product. No one can replace that. And not all traffic is equal. It is a huge miss to send all of these to the latest opus model. Some of these absolutely large scale data analysis when you do it on notion will recommend
我非常认同这个观点。而对每一个产品来说，什么算前沿、什么算日常，什么算能力已经饱和、什么算模型能力有富余（capability overhang），这些的定义都取决于你对自己产品的专长，没有人能替代这一点。而且并不是所有的流量都是等价的。把所有这些请求全都发给最新的 Opus 模型，是一个巨大的失误。这里面有些请求，绝对属于大规模的数据分析——你在 Notion 上做这类事情的时候，我们就会推荐
[10:34]
Opus, right? When you triage an email inbox, if we're charging you to do that on Opus, we're ripping you off and ourselves. Think about where your traffic patterns are. And then think about how Frontier Lab model providers are structured today. I mean, it's functionally an oligopoly,
用 Opus，对吧？但当你只是在整理一个邮件收件箱的时候，如果我们还收你的钱、用 Opus 去做这个，那我们既是在坑你，也是在坑我们自己。想想你的流量分布模式到底在哪里。然后再想想今天这些 Frontier Lab 模型供应商是什么样的格局。我是说，它实质上就是一个寡头垄断，
[10:54]
right? And that's fine because they're racing to the top. And I think the top is really hard and really important. This is not to say that products don't have a place for frontier difficult tasks. I want everyone to nod and understand that's not what this talk is about. Understand when you need those tasks and it's not everything. The problem with those tasks are is keep in mind how pricing is incentivized. You can figure out who these players are.
对吧？这没什么问题，因为他们在争着往顶尖去冲。我觉得那个顶尖既非常难、也非常重要。这并不是说产品里就没有前沿难题的位置。我希望每个人都点点头、理解这一点：这不是我这场演讲要讲的重点。你要清楚什么时候你才需要那些高难度任务，而它并不是全部。这些任务的问题在于——你要记住定价背后的激励逻辑是怎样的。这些玩家都是谁，你自己就能想明白。
[11:18]
Either you are the best model. Everything above what AI can't do today is your market. You can basically price it as high as you kind of want. If you're slightly behind that best model, all you need to be is like a dollar per million tokens cheaper and you have the rest of the market. You know that economic theory about gas stations where the best gas stations are the ones that are right next to each other because they cover east and west the most. Yeah.
要么你就是那个最好的模型。今天所有别的 AI 做不到的事情之上的那片领域，都是你的市场，你基本上想把价格定多高就能定多高。而如果你只是稍微落后于那个最好的模型，你要做的就只是每一百万 token 便宜大概一美元，剩下的整个市场就都是你的了。你知道那个关于加油站的经济学理论吧——最好的加油站往往是紧挨在一起的那几家，因为这样它们能最大程度地覆盖到东西两个方向的客流。对，就是这个理。
[11:44]
It's the same with model pricing, which means that price does not correlate with capability growth. So for this complex task, understand what capabilities you need, but be the expert on what complexity is. And keep in mind that who handles complexity changes. Um, oftentimes you'll see applied AI companies really be super outspoken on marketing with a specific lab. That's always kind of a red flag for me when they're not model
模型定价也是同样的道理，也就是说，价格和能力的增长并不成正比。所以对于这种复杂任务，你要搞清楚自己需要哪些能力，同时你自己得成为判断“复杂度”的专家。还要记住，负责处理复杂度的角色是会变的。很多时候你会看到一些应用型 AI 公司，在营销上特别高调地站队某一家实验室。对我来说，当他们不是 model
[12:14]
agnostic because if you look at this graph, it basically shows that they're behind every month, right? the new model and the new model provider of the best Frontier capabilities change and if you hit your ride with one particular provider in exchange for for instance a larger discount um you're doing a disservice to your customers like half of the time right so really think about if that discount is worth not actually having a Frontier
agnostic（不绑定特定模型）的时候，这总是一个危险信号。因为如果你看这张图，它基本上说明他们每个月都在落后，对吧？拥有最强 Frontier 能力的新模型、新的模型提供商每个月都在变，如果你为了比如说更大的折扣，就把自己绑死在某一家提供商身上，那你差不多有一半的时间都是在坑你的客户，对吧？所以真的要好好想想，那个折扣到底值不值得让你失去一个真正 Frontier
[12:40]
product and remember that that optionality is your leverage if you don't have the capability to walk at any point you are stuck. And again, I think that's probably the most expensive decision you'll make regardless of what discount you get or the engineering work to have model interoperability.
的产品。记住，这种可选择性（optionality）就是你的筹码。如果你在任何时候都没有掉头走人的能力，那你就被困住了。再说一遍，我认为这大概是你会做出的、代价最高的一个决定——不管你拿到多大的折扣，也不管你为实现模型互操作性要投入多少工程量（都比不上被锁定的代价）。
[13:01]
One option to navigate this is stay model agnostic. Have different models and capabilities in your system so that at any point if pricing seems unfair or untenable, you are not out of business. Notion's auto model does this really well. We have state-of-the-art models available always. Um, but we also have an auto model there at the top that handles about 75% of our traffic. Right?
应对这种情况的一个选择，就是保持 model agnostic。在你的系统里放进不同的模型和能力，这样在任何时候，一旦定价看起来不公平或者难以为继，你也不至于就此做不下去。Notion 的 auto model 在这方面做得非常好。我们始终提供最先进的模型。但在最上层，我们还有一个 auto model，它处理了我们大约 75% 的流量，对吧？
[13:29]
We have the ability to switch between models in our product and we also offer it to our customers so that they have access to these models without vendor lockin. That's part of our AI Switzerland approach. You guys love taking photos of slides.
我们在产品里可以在不同模型之间切换，我们也把这个能力开放给客户，让他们能用上这些模型，而不会被 vendor lock-in 锁住。这是我们“AI 界的瑞士”策略的一部分。你们特别喜欢拍幻灯片。
[13:43]
This is the slide. Okay. Model agnostic playbook. This is how you do it. Build for multimodal. It is hard to kill the cache and switch models mid-transcript. I understand that we invest in that technology. It doesn't even have to be per thread. Just think about your harness as model interoperability.
就是这张幻灯片。好。Model agnostic 的实操手册，就是这么来做的。要为多模型（multi-model）去做设计。在对话进行到一半时清掉 cache、再切换模型，这很难，我理解，所以我们在这项技术上做了投入。它甚至不一定要做到每个 thread 都能切。你就把你的 harness 当成模型互操作性来设计就行了。
[14:02]
Think about the cost per capability per second, not just the tokens. Here's a great example. We posted this review when we um announced our partnership with Parallel as our web search provider. If you were to look at just latency of a single call or just cost,
你要考虑的是每秒、每种能力的成本，而不只是 token。这里有个很好的例子。当我们宣布和 Parallel 合作、把它作为我们的网页搜索提供商时，我们发布了这份评测。如果你只看单次调用的延迟，或者只看成本，
[14:17]
parallel might not be the cheapest. But if you have expertise in entire web search trajectories, you'll see how it differs. The granularity of this eval is what lets us make the best decisions for our customers because we understand all of the trade-offs on entire trajectories, not just single calls.
Parallel 可能不是最便宜的。但如果你对整条网页搜索轨迹（trajectory）有深入的理解，你就会看出差别到底在哪里。正是这份 eval 的这种细粒度，让我们能为客户做出最优的决策，因为我们理解的是整条轨迹上所有的取舍，而不只是单次调用。
[14:35]
Switch fast and often. I think we talked about that. And give them something back. That expertise on use cases is also very valuable to Frontier Labs. We find that our eval program partnerships actually help us a lot with Frontier Labs and is something that we can exchange instead of extraordinarily large commits and I don't think the discount is ever worth the loss in optionality.
要快速、频繁地切换。这个我们应该已经聊过了。另外，也要给他们一些回报。你在具体使用场景上积累的专业经验，对 Frontier Labs 来说也非常有价值。我们发现，我们的 eval 项目合作其实在跟 Frontier Labs 打交道时帮了我们很大的忙，这是我们可以拿来交换的东西，而不用去做那种超大额的承诺（commit）。我一直觉得，折扣再大，也不值得你为此失去可选择性。
[14:58]
That's a perspective you can choose to keep or not. The second option is moderate tasks understanding open weights place there. um openweight models are really strong enough to handle these tasks and the possibility to RL on top of them has also kind of expanded the upmarket growth that they can cover. I view openw weight models as basically lowering the barrier to entry on cost for our customers and they also give you negotiation leverage. So it's
这个观点你可以选择接受，也可以不接受。第二个选择，是那些中等难度的任务，你要明白 open weight 在这里的位置。open weight 模型其实已经足够强，能搞定这些任务；而且在它们之上做 RL 的可能性，也进一步拓宽了它们能覆盖的高端市场空间。在我看来，open weight 模型基本上就是在为我们的客户降低成本上的准入门槛，同时它们也给了你谈判的筹码。所以它算是
[15:26]
kind of a credible alternative that's putting that downward pressure on pricing that if there's an igopoly of two or three providers at the top is unavailable right now otherwise. I think Kimmy 26 was probably the first time that we really saw a model that outperformed 52, GPT 52, GLM 52 now is another 52 bombshell in the villa that also probably does best here. But it's no longer the case where openway models are good for just SFT on small tasks.
一个相当可信的替代方案，能对定价形成向下的压力——而如果顶端只有两三家提供商形成寡头垄断（oligopoly），这种压力在其他情况下现在是根本不存在的。我觉得 Kimi K2 大概是我们第一次真正看到有一个模型的表现超过了 GPT-5.2，而现在 GLM-5.2 又是一个 5.2 级别的重磅炸弹，在这类任务上大概也是表现最好的。但现在的情况已经不再是——open weight 模型只适合在小任务上做 SFT 了。
[15:54]
Really think about without RL if they're capable enough for what you need. Um, and again, don't just think about external benchmarks. Be able to have expertise on your system. What are your tool errors? What's the actual latency that you need? Right? Here's an example of a benchmark that we posted. It's a little bit stale on purpose, right? But you get the idea.
真的要想清楚，在不做 RL 的情况下，它们是不是就已经足够满足你的需求了。还有，再强调一次，别只盯着外部的 benchmark。你得对自己的系统有专业的理解。你的工具调用错误率是多少？你真正需要的延迟又是多少？对吧？这里有一个我们发布过的 benchmark 例子。它是故意做得稍微有点过时的，对吧？但你能明白我的意思。
[16:21]
Philip at BE 10 this showed this slide once and I've stolen it ever since. Um, thank you. Are you here, buddy? Okay. Well, chat, hi. Um, well, he could come up and say it better, but the idea is that you don't have to be at the top, right? I'm not trying to make a case that open weight is the best model out there. Um, the case being made, however,
Baseten 的 Philip 有一次展示了这张幻灯片，从那以后我就一直“偷”用它。谢谢你。你在现场吗，老兄？好吧。那，观众朋友们，大家好。他本人上来讲会讲得更好，但核心意思是：你不需要做到最顶尖，对吧？我并不是想论证 open weight 就是市面上最好的模型。不过，我真正想说的是，
[16:43]
is that um the gap gets covered eventually. So, if the tasks that you're having today are good enough, then in six months, they're probably covered by open weight. So, be prepared now. And the last thing is CPUs over GPUs. Um, we've we've recently launched something at notion called workers. I don't think that the GPU is necessary for every job.
这个差距最终是会被追平的。所以，如果你今天手上的任务（用 open weight）已经够用了，那么六个月后，它们很可能就完全能被 open weight 覆盖。所以现在就要做好准备。最后一点是：CPU 优先于 GPU。我们最近在 Notion 上线了一个叫 workers 的东西。我不认为每一项工作都需要用到 GPU。
[17:07]
A lot of the jobs that we have are actually serving um discrete pieces of code. Like you don't need an LLM to turn a CSV into a PDF. You don't need an LLM to talk to notion tool calls if we have a CLI. You definitely don't need an LLM to do deterministic SQL queries. This is where people become token poor very quick.
我们的很多任务其实就是在跑一些离散的代码片段。比如说，把一个 CSV 转成 PDF，你根本不需要 LLM。如果我们有 CLI，去调用 Notion 的 tool call 也不需要 LLM。做确定性的 SQL 查询就更不需要 LLM 了。人们就是在这种地方很快变得“token 贫穷”的。
[17:28]
And I think the last option here besides openw weight CPUs and optionality is actually governance. Um there's a lot of AI governance. Um one is visibility. Um understanding who's using the data, understanding its maintainability and control. When you have model optionality, you can offer a lot more to your customers.
我觉得，除了 open weight、CPU 和可选择性之外，这里的最后一个选择其实是治理（governance）。AI 治理有很多方面。其中一个是可见性——搞清楚是谁在使用数据，理解它的可维护性和可控性。当你拥有模型上的可选择性时，你能为客户提供的东西就多得多。
[17:48]
Um here's an example of how that governance works in notion. So final tips again, think about architecture, think about open weight, and build value that transcends tokens. So we're going to depart Token Town. I know I said welcome to Token Town. We're going to spend the next 10 minutes really thinking about what to do next.
这里有个例子，展示这种治理在 Notion 里是怎么运作的。那么，再总结几条最后的建议：多想想架构，多想想 open weight，去构建那些超越 token 本身的价值。接下来我们要离开 Token Town 了。我知道我一开始说的是“欢迎来到 Token Town”。接下来的十分钟，我们要认真想想下一步该做什么。
[18:09]
So I think the challenge of the next six months doesn't have to do with capabilities. I think it has to do with security. Let's start there. There's this concept called the lethal trifecta. Simon Wilson, I think, crafted this. If you have access to private data,
我认为接下来六个月的挑战，跟能力无关。我觉得它跟安全有关。我们就从这里说起。有一个概念叫“致命三要素”（lethal trifecta），我记得是 Simon Willison 提出来的。如果你的系统能访问私有数据，
[18:23]
exposure to untrusted content, whether it be through ingestion, MCP, email, right? And the ability to ex to communicate externally, and that can include like payloads in a web search. The second you have that system, you're exposing risk. And in fact, the more autonomous your system is, the more unsupervised this risk is. I think that this is what builds valuable product,
又暴露在不可信的内容之下——不管是通过数据摄入、MCP 还是邮件，对吧？再加上对外通信的能力，这里面可能就包含了网页搜索里夹带的攻击 payload。一旦你的系统同时具备了这几点，你就在暴露风险了。而且事实上，你的系统越自主，这种风险就越是无人监管。我认为，真正能打造出有价值产品的，正是这一点，
[18:45]
not just capability. Same with sandboxes and computers. We talked about this, but it really is something that builds better determinism in your product and also better token economics for your customers in multi-agent orchestration. Understanding what agents see and do and what persists. I think persistence of enterprise knowledge is something that's actually really not discussed enough.
而不只是能力。sandbox 和 computer（电脑操作）也是一样。这个我们之前聊过，但它确实能让你的产品拥有更好的确定性，在多智能体编排（multi-agent orchestration）里，也能为你的客户带来更好的 token 经济性。要理解智能体看到了什么、做了什么，以及有什么东西会被持久化保留下来。我觉得，企业知识的持久化，其实是一个远远没有被充分讨论的话题。
[19:13]
It's starting to be with some recent launches, you know. Oh, there is audio. So don't have your workflows look like this. And I think this is where most software factories are today, right? It's like actually your entire engineering time just spends time babysitting the factory, right? I mean, I get it. Ours started off like this. Agent orchestration is one of the most difficult tasks of making factories work.
随着最近的一些产品发布，它开始被讨论了。哦，原来是有声音的。所以，别让你的工作流变成这个样子。我觉得，这就是今天大多数“软件工厂”（software factory）所处的状态，对吧？基本上就是，你整个工程团队的时间全都花在给这个工厂当“保姆”上了，对吧？我懂，我们一开始也是这样的。智能体编排，是让工厂真正跑起来最难的任务之一。
[19:38]
So okay, this is me telling Tay to tell people to buy notion AI. And the reason I included this slide is I am going to sell notion for a second. It's my job. Always be closing. Always be selling. Always be hiring. come find me. But I'm going to talk for a second about how notion does this today. We already have the ability to inspect tasks.
好，这一页其实是我在让 Tay 去劝大家买 Notion AI。我放这张幻灯片的原因是，我要花一点时间给 Notion 打个广告。这是我的工作嘛。永远在成交，永远在推销，永远在招人——欢迎来找我。不过我想花点时间讲讲，Notion 今天是怎么做这件事的。我们已经具备了检视任务（inspect task）的能力。
[19:57]
And you can imagine any task that you look at um in a notion document. You can have Claude actually go ahead and scope out what you need. We've launched this manage agent capability today. So if I go ahead to the top of this task, I can actually ask Claude agent to scope out the task. Right?
你可以想象一下，在 Notion 文档里你看到的任何一个任务。你都可以直接让 Claude 去把你需要做的事情梳理清楚（scope out）。我们今天上线了这个 managed agent 的能力。所以，如果我滚动到这个任务的最上面，我其实可以让 Claude agent 来梳理这个任务的范围，对吧？
[20:18]
Ideally, it's working. Um, and you'll see it'll actually populate um an entire spec of what needs to be done. In this example, it's not ready. It's going to ask me a question. Keep in mind, this isn't a markdown file. This is an active document. Um,
理想情况下，它是能正常工作的。你会看到，它其实会自动生成一整份“需要做什么”的 spec（规格说明）。在这个例子里，它还没准备好，它要问我一个问题。记住，这不是一个 markdown 文件，这是一个“活的”文档。
[20:35]
let's say I don't actually know the question and I go ahead and I ask my team um what to do. Imagine that you can kind of tag in your team into these systems. MJ's our PM. So in this example, she doesn't know. Usually she does, but multi-agent orchestration is important. Maybe Claude Code isn't the best at customer voice,
假设我其实并不知道这个问题的答案，那我就去问我的团队该怎么办。想象一下，你可以把你的团队成员“叫进”（tag in）这些系统里。MJ 是我们的 PM。在这个例子里，她也不知道。通常她是知道的，但这恰恰说明多智能体编排很重要。也许 Claude Code 不太擅长把握客户的声音，
[20:58]
but Decagon is, right? You can ask Decagon agents, we're proud partners of them as well, to collect the right data that you need. Okay, in this example, we think we know enough. We're going to go ahead and actually um iterate through some of this flow. I'm going to skip ahead a little bit. We asked our TL what we needed. He replied again. It's a collaborative file, not just a markdown. And we can have Claude actually go ahead and spin up the PR.
但 Decagon 擅长，对吧？你可以让 Decagon 的智能体——我们也很自豪能成为他们的合作伙伴——去帮你收集你需要的正确数据。好，在这个例子里，我们觉得已经了解得差不多了。我们就继续，实际地把这个流程往下推进几步。我快进一点。我们问了我们的 TL 需要什么，他又回复了。这是一个可协作的文件，不只是一个 markdown。然后我们可以让 Claude 直接去创建这个 PR。
[21:27]
Hopefully, this is looking a little familiar now. This is kind of the vision of software factories. It's what we're trying to host. Okay, Claude put up a PR. Maybe that's not enough. Um maybe I want to go ahead and ask Codeex what it thinks.
希望现在这个画面看起来有点熟悉了。这差不多就是“软件工厂”的愿景，也是我们想要去承载的东西。好，Claude 提交了一个 PR。也许这还不够。也许我想接着去问问 Codex 它怎么看。
[21:48]
Great. Found two issues. You can think about this scaling in an actual factory. So today in notion, you're actually able to orchestrate these agents together and you're not committing to a lab. You're committing to the concept that AI is augmenting and automating what you do.
很好，发现了两个问题。你可以想象一下，这个模式在一个真正的工厂里规模化会是什么样子。所以在今天的 Notion 里，你其实可以把这些智能体编排协作起来，而且你并没有把自己绑定到某一家实验室。你绑定的是这样一个理念：AI 是在增强、并自动化你所做的事情。
[22:08]
This is real. I asked Rejieve if I could post this. This is how it works today internally at notion. Almost all of our polish and large feedback like this is actually coordinated um through our software factories both in terms of writing to the right teams and also having coding agents take the first step. Verscell does this as well from staging to shipping to closing.
这是真实的。我问过 Rajeev 我能不能把这个发出来。这就是今天 Notion 内部实际的运作方式。我们几乎所有的打磨工作，以及这类大量的反馈，其实都是通过我们的软件工厂来协调的——既包括把内容写给正确的团队，也包括让编码智能体（coding agent）先迈出第一步。Vercel 也是这么做的，从 staging 到 shipping 再到 closing 全流程。
[22:33]
And we see massive ROI gains from our customers. That's over three minutes saved on a given task. Imagine that at scale. So I think we're trying our hardest to think about the factory lens. We cannot do this without optionality and we cannot do this without conviction that we understand what models are required for which tasks.
我们从客户那里看到了巨大的 ROI 提升。单个任务就能节省超过三分钟的时间。想象一下这个在规模化之后是什么效果。所以我觉得，我们在拼尽全力地用“工厂”这个视角去思考。没有可选择性，我们做不到这一点；如果我们不能笃定地知道哪些任务需要哪些模型，我们同样也做不到这一点。
[22:53]
It's really wild out there you guys. I get it. The market is really young. It's exceptionally opaque. It's moving fast. I'm super grateful for communities like AI engineer to bring us together and like talk openly about these things and how we navigate it. Um I think we owe it to all of our customers to get it right and to be critical thinkers about how we navigate this together. I'm chronically online fortunately. Um you can always DM
外面的世界真的很疯狂，各位。我懂的。这个市场还非常年轻，异常不透明，而且变化飞快。我特别感激像 AI Engineer 这样的社区，能把大家聚到一起，坦诚地聊这些事、聊我们该怎么去应对。我觉得我们有责任为所有客户把这件事做对，也有责任在一起摸索这条路的时候保持批判性思考。好在我是那种常年泡在网上的人。你们随时都可以私信我——
[23:19]
me on Twitter, you can email me, you can find me after this. Um but thank you for yapping with me and thinking about this problem and have a good day. >> [music] [music] [music]
——在 Twitter 上找我，也可以给我发邮件，或者会后直接来找我。总之，谢谢你们陪我一起唠这些、一起琢磨这个问题，祝大家今天愉快。>> [音乐] [音乐] [音乐]