### `04:10` AI 交互的“移动球门”与三大技术维度
Jet 提到，他第一次和江洋见面面试时，江洋给他看了一篇为公司第一次全员大会（All Hands）准备的文档。那篇文档的核心论点是：“与 AI 合作的最佳方式尚未被发明（the best way to work with AI hasn't been invented）”，而他们的目标就是成为找出这个解法的团队。
江洋以 Cursor 为例，指出其在探索“如何用 AI 写代码”这一目标上取得了巨大成功，因为他们比其他人更早地聚焦在这一终点线上。但江洋强调，这是一个“移动的球门（moving goalpost）”，你永远无法真正触达市场预期的终点，只能不断做得更好。因为只要你做出新东西，其他有品位的团队和用户就会看到其中的摩擦点。随着 OpenAI 和 Anthropic（推出 Claude Artifacts 等）的不断迭代，许多创始人感觉每天都在面对一个全新的世界。
随着 Context 的增加，江洋直言自己现在“极其讨厌 Prompting”：
> “从现在开始 like I just did not want to prompt. I think prompting is such a... I mean, you have to communicate. But people, we don't think in prompts.”
`25:00` The Inspiration for Paperboy's Interface and Overcoming the "Sidebar" Dilemma
The conversation centers on the underlying inspiration for Paperboy and how the team navigates complex product decisions. One speaker explains the desire to move beyond manual processes—like sketching thoughts on paper—and instead leverage a smart model that intimately knows the user to thread disparate ideas together across different planes. This philosophy directly influenced Paperboy's latest interface, which began development just two weeks prior.
The team encountered a significant product dilemma when considering venture capitalists (VCs) as part of their Ideal Customer Profile (ICP). They realized VCs require highly specific, varied modules: a personal CRM, an upcoming meetings tracker, a quarterly deals overview, and tools to monitor both portfolio companies and LP relationships. Initially, the team considered shipping a standard sidebar with these pre-built modules. However, this triggered a broader architectural question: since operators, founders, and real estate salesmen all require vastly different sidebars, should these features be shipped as "skills, plugins, or recipes" for the user to manually add?
Ultimately, Paperboy itself helped resolve the dilemma by steering the team away from an overly complex sidebar model. It pushed them toward an instant messaging (IAM) framework inspired by WeChat and iMessage dynamics in China. In these ecosystems, users organically and intuitively manage multiple group chats—sometimes creating four separate groups for the exact same three people just to segregate distinct topics. Instead of building a traditional, annoying "infinite list" or a standard inbox feature, the team decided to turn the product itself into an IAM-style inbox, resulting in Paperboy's unique communication and design layout.
* **VC Module Requirements:** Personal CRM, upcoming meetings, quarterly deals, portfolio tracking, and LP relationship management.
* **The WeChat Inspiration:** Users creating multiple distinct group chats with the same participants to naturally isolate and organize different topics.
* **The Interface Philosophy:** Shifting away from rigid sidebars and instead designing the core product to function natively as an intuitive IAM-style inbox.
---
`28:13` The Competitive Landscape and the Limitations of "AI Slack"
The discussion shifts to how a small startup like Paperboy can survive and maintain an edge against massive tech incumbents with superior distribution channels. The team acknowledges that they must rely on having a "taste" edge to beat larger players to new interfaces, drawing a parallel to how the code editor Cursor carved out its own space. While an exceptional individual product experience is vital, the ultimate goal for monetization is entering the enterprise market.
Interestingly, Paperboy’s very first product concept was actually an "AI Slack." The team initially viewed this as the logical answer for how humans should collaborate with AI agents. However, the speaker notes that the AI Slack space has quickly become incredibly crowded, with numerous products attempting to solve team-and-agent integration. Despite a deep admiration for Slack and its leadership, the speaker outlines two massive friction points that ultimately caused Paperboy to pivot away from trying to replace it.
First, the enterprise adoption barrier is incredibly high. To convince a company to switch, a competitor must build a flawless one-to-one data import/export experience alongside a mountain of custom data connectors for all the existing software an enterprise uses. Furthermore, the speaker candidly shares that from a user perspective, Slack is "honestly not a great product" to use due to its rigid channel-and-thread structure, even though companies scale into it because nothing better currently exists. Second, the team does not believe AI agents can effectively learn how to work with humans if their only context comes from chat messages. Employees are simply too busy to train agents manually; they expect software to work out of the box.
* **The Incumbent Advantage vs. Taste:** Small teams must rely on superior design taste and rapid execution, similar to Cursor's strategy.
* **Enterprise Switching Costs:** Replacing Slack requires building extensive data connectors and a seamless one-to-one data migration infrastructure.
* **The Agent Training Flaw:** Relying solely on Slack channels and threads does not provide enough natural context for AI agents to learn effectively without burdening busy users.
---
`31:37` Network Effects and the Complementary "OS Layer" Strategy
The speakers analyze Slack's formidable network effects, comparing its defensive moat to WeChat's dominance. A key element of this moat is Slack's "external connections" feature. For transparent and serious companies, inter-company communication has largely shifted away from traditional email and into these shared enterprise channels. Because of these deep integration network effects, trying to completely replace Slack for any team larger than 50 people is a monumentally difficult task.
Instead of fighting these network effects head-on, Paperboy shifted its strategy to a complementary approach. Rather than trying to build a new chat app from scratch, they view their product as a layer operating above or alongside existing workflows—closer to an Operating System (OS) level. This creates a much more compelling adoption story, as it brings AI agents directly to where people are already working instead of forcing an organizational migration.
The conversation transitions to Paperboy's philosophy on product speed, referencing an unreleased blog post on their website titled *The Last Interface*. This post outlines "five kinds of speed." The speaker clarifies that five is just a sensible categorization inspired by the "Pace Layers" concept originated by the Long Now Foundation in San Francisco. The Pace Layers model views the world through six layers changing at different velocities—ranging from fashion and commerce at the fast end, to infrastructure, civilization, and nature at the slow end.
* **Slack’s Moat:** External connections have replaced email for business-to-business communication, locking companies into the ecosystem.
* **The Complementary Strategy:** Bringing agents to existing workspaces rather than forcing teams to switch off Slack.
* **Pace Layers Influence:** A framework adapted from the Long Now Foundation to understand how different systems and velocities interact within a product architecture.
---
`34:42` The Time-Horizon Spectrum of Human and Agent Tasks
The speakers break down how the concept of pace layers applies to user actions and AI augmentation. They view user behavior as a spectrum of time horizons. On one extreme, a user performs rapid actions, such as taking one to ten seconds to reply to a message on WeChat. On the opposite end of the spectrum, a user might spend several hours reading ten deeply detailed reports to make a major business decision.
The team believes that while AI agents will inevitably augment or automate tasks across this entire spectrum, the required product form factor changes radically depending on the time horizon. Automating a multi-hour task remains an entirely open and highly experimental product frontier that the team is eager to explore.
The host transitions the topic to Jump’s background, noting that despite being born in 2004, his maturity and clarity stand out to everyone who interacts with him. Jump playfully pushes back on this, claiming he is still a "pretty childish person" who likes to have fun with his co-founder, Jet. However, he acknowledges that running a startup requires extreme clarity, efficiency, and focus. He defines his primary job as CEO as bringing absolute clarity to the team: defining what success looks like and ensuring every person has complete, accurate context regarding their goals and operations.
* **The Task Spectrum:** Ranges from short-horizon tasks (1–10 second messaging) to long-horizon tasks (multi-hour report analysis and decision-making).
* **The Product Frontier:** Designing the ideal user interface and form factor to automate multi-hour, complex workflows.
* **The CEO’s Role:** Ensuring high context alignment across the team so everyone understands their exact goals and operational boundaries.
---
`37:28` Early Startup Lessons, Market Selection, and the Role of a First-Time CEO
Jump reflects on his personal journey from high school to college, noting that his insatiable curiosity has been his primary driver. Before Paperboy, he took several internships across different companies to learn diverse problem-solving methodologies. He and his peers initially started a B2B business selling web performance PR bots to large enterprises. They quickly learned that enterprise sales is an incredibly difficult business to scale, especially for 20-year-olds who lack experience navigating enterprise relationships.
This early lesson taught Jump the vital importance of "founder-market fit" and "team-market fit." He argues that market selection is perhaps the most critical trait for an early-stage founder. When picking a market, it cannot just be about the current dollar sign; the market must be highly sustainable and capable of compounding over a decade so the team can continuously evolve its product lineup. Jump points out that history's most successful products are rarely just a single release; they represent a decade-long product lineage aimed at eventually reaching mass consumer adoption.
Transitioning into his first role as a CEO, Jump admits, "Man, it's fucking hard. I gotta own everything." He details the intense leverage a CEO holds, where a single bad idea can cost the company immense amounts of money. As the team scales, the responsibilities shift heavily toward hiring the right people, fostering their continuous growth, and learning basic management skills on the fly—like conducting performance reviews and avoiding micromanagement.
* **Early B2B Pivot:** Shifting away from an enterprise PR bot startup after realizing 20-year-olds face steep structural disadvantages in enterprise relationship building.
* **Market Criteria:** A viable market must have a compounding scale that allows for a decade-long product lineup trajectory.
* **CEO Leverage:** The realization that a CEO's strategic choices carry massive financial consequences compared to isolated execution delays.
---
`41:14` Management Mentors, CEO Coaching, and Structuring an Engineering Team
To navigate his first managerial role, Jump leaned on resources and mentorship. He credits Panpan, the CTO of Amanis (where Jump previously worked), as an exceptional engineering manager who gave him invaluable advice during one-on-one sessions. Jump also devoured classic business literature, citing Andy Grove's *High Output Management* as the best management book he has ever read, alongside Ben Horowitz's *The Hard Thing About the Hard Things* and Bill Campbell's biography, *Trillion Dollar Coach*. Additionally, an investor introduced him to a executive coach whom he meets with weekly for an hour. Interestingly, Jump uses Paperboy during these coaching calls; the AI listens to the audio and automatically helps him organize his action items and commitments.
When asked how he structures his small 10-person team given the abundance of modern AI tooling, Jump explains that serious infrastructure development still requires deep human expertise. Paperboy intentionally hires two distinct profiles on the engineering side. The first profile, exemplified by Jet, consists of young, high-IQ, highly creative individuals who excel at rapidly prototyping creative solutions to any problem they encounter. The second profile consists of rock-solid systems experts who intimately understand operating systems and kernels, such as a recent hire from AWS who specialized in Windows kernels for AWS infrastructure.
The conversation concludes with Jet explaining why he chooses to stay in college at Carnegie Mellon University (CMU) despite being a Founding Engineer at Paperboy, especially at a time when companies like Palantir are offering high schoolers direct paths to skip university entirely. Jet explains that for most people, university provides crucial value because many students are still unsure of what they want to do. CMU allows him to steadily elevate his technical abilities in a structured environment where academic work only takes up a fraction of his time, leaving the rest of his schedule free for self-directed exploration over a four-year period.
* **Essential Management Reading:** *High Output Management* by Andy Grove, *The Hard Thing About the Hard Things* by Ben Horowitz, and *Trillion Dollar Coach* (Bill Campbell biography).
* **The Two Engineering Profiles:** 1) Young, high-IQ rapid prototypers. 2) Deeply experienced systems/kernel engineers from infrastructure giants like AWS.
* **The Value of College in the AI Era:** Jet notes that CMU provides a low-risk environment to compound technical skills while offering ample free time for independent projects.
在具体到如何说服新的候选人（candidate）时，John 坦言自己并不会用一套固定的脚本（script）去生硬地推销公司（hard sell）。他认为招聘必须是双向奔赴的（mutual），他更倾向于针对每个人进行个性化的沟通。在挑选人才时，他强调必须寻找最顶尖的人，拒绝平庸（mediocre）。判断一个人的成色，可以通过深入聊对方过去的工作，而最核心的则是观察他们**“过去的人生选择（past life decisions）”**。特别是对于年龄稍长、拥有十年以上决策经验的人，这些真实的选择是无法伪装的（You can't fake that）。
- **职业选择金句**：很多时候，Slope（成长斜率）比 Intercept（初始截距）重要得多。
- **面试考核核心**：不看包装，最看重候选人过去的人生选择（past life decisions）。
---
`47:11` 独立发展与拒绝大厂 Aqua-hire
当被问及除 Paperboy 之外最喜欢的 AI 产品时，John 毫无掩饰地表达了对 Cursor 的喜爱，甚至直言自己至今不怎么用 Codex，而是对 Cursor 推崇备至。有趣的是，对话揭示了一段关于独立性抉择的往事：在 John 还是十几岁（when he was a billion/teen）的时候，曾经收到过来自 Devin 的母公司 Cognition，以及 Vercel 和 Sentry 等知名公司的收购或加入邀约（offers）。
## Notable quotes
> `01:59` 人类和 AI 配合工作的最佳方式，很可能还没有被发明出来。 — 江洋
> `06:21` 你不该像学新工具一样学怎么用它。 — 江洋
> `16:21` 从现在开始 like I just did not want to prompt. I think prompting is such a... I mean, you have to communicate. But people, we don't think in prompts. — 江洋
> `53:06` Well, I hope we're not losing money every day anymore. — 江洋
## Mentioned
Paperboy, John Yang, 江洋, Jett Chen, Jet, Claude Code, Codex, Manus, OpenClaw, Claude, Cursor, Littlebird, Mini Vivian, Vivian, Auto John, AutoJong, Koji杨远骋, 微信, Slack, iMessage, The Last Interface, Million, same.dev, YC, OpenAI, Anthropic, Claude Artifacts, GitHub, 小红书, 推特, Twitter, Hacker News, Vercel, Sentry, Cognition, Devin, Sky, Apple Shortcuts, Codex Test, xAI, Panpan, Amanis, Andy Grove, High Output Management, Ben Horowitz, The Hard Thing About the Hard Things, Bill Campbell, Trillion Dollar Coach, AWS, Carnegie Mellon University, CMU, Palantir, Sweet-bench, 张阳阳
[1:59] 江洋: 人类和 AI 配合工作的最佳方式，很可能还没有被发明出来。
[6:21] 江洋: 你不该像学新工具一样学怎么用它。
[16:21] 江洋: 从现在开始 like I just did not want to prompt. I think prompting is such a... I mean, you have to communicate. But people, we don't think in prompts.
[21:55] Jet: 我每天晚上就是会让 Paperboy 就类似给我一个 report 就说我今天到底有多 efficient，我让他会去就是相当于骂我，我什么时候就是不够高效、我什么时候分神了。
[45:00] Jet: slope（斜率/成长速度）比 intercept（截距/起点）重要很多
[45:00] 江洋: 过去的人生选择（past life decisions）
[47:11] 江洋: stay independent
[53:06] 江洋: Well, I hope we're not losing money every day anymore.
Original transcript
[0:00] 嗨,我是Koji本周十字路口我们请到的嘉宾是Paperboy的创始人江洋以及他们的一位Funding EngineerJet成你好二位,欢迎来到十字路口Hello,Koji在这里也给大家说明一下因为江用英文表达会比较自然所以我们这一期内容会比较中音夹杂我们同时有音频和视频如果大家需要的话可以去看视频看视频我们就会配字幕和字幕的翻译帮助大家理解我们还是从十字路口的传统快按快答开始有请二位的年龄I'm 2119你们的毕业院校我是上海新欢双语学校高中毕业然后现在在CMU读大一大一刚结束你们的MBTI和星座是什么ISTJ双字座INTJ处女座创业之前你们做了一些什么Payboard是我第二家公司我的第一家公司叫Million我们是在react ecosystem里做了很多开源的dev tools然后我们做了一个产名叫same.dev普通人可以用一句话去输入一个url然后做出和任何网站一模一样的一个UIAnd miller was in YC, winning 24. started broadly with the mission of I am trying to figure out So yeah, Paperboy just started with me trying to hack around different approaches to try to figure out how to solve there are technical problems here but also form factor Like, I shouldn't have to put my information, email, and all this personal information into a situation. If I need to collaborate with each other, I can also talk to an agent. This should be a very easy way for me to, for us to collaborate within the same context window. Agents, if you know my many things, they should be proactive to help me do some things. But in the chat window, you really can't do it. And then all of the products are sessions based. So if you don't find out what context window is, you should continue to do it. So just all kinds of these problems, I think, where generally speaking, it's the gap between model capabilities and deployment. And I feel like there's still a large opportunity to explore and innovate on the product experience and hence the name is Paperboard Products. We'll open it up, but we'll do it again. What's your income? We raised 4.7 million at 25. Oh, cool. Zero and negative gross margins And we lose money every day So we shipped the prototype to friends And that's like our agent that learns from OS activities But there are cost issues It's too expensive It doesn't really work So the next iteration of this product We're tryinto wrap it up by this month. And it's going to try, we're going to try to go to market again. We'll put a link in the description below. If you're interested, you can sign up for it. Oh yeah, you guys will see it, I think, when this podcast comes out. And what's the team's规模? There are 12 people in the community, and there are 10 engineers. I remember when I first met him like this, you gave me a page. The article was written for the interview. The first thing was, you said that the best way to work with humans was not being published. When you wrote this article, what did you see? And then, you thought you saw some changes? Yes, so that doc is the one I prepared for our very first All Hands. That's when I was a company, I was a jet company. and and yeah I think we just started working with one thesis which is the best way to work with AI hasn't been invented and we have shot at being the team that figures out that way so Chris just started really trying to figure out what's the best way to code with AI and
[5:00] they have been largely successful being the first company to really approach on that goal and to figure out how important it was to focus on that goalpost before everybody else. I think what's cool is this is a moving goalpost. Instead of market's expectation, you can't really reach it. You can only keep getting better because every time you come up with something new, Right, everybody else see it And if the other teams have taste If users have taste They will know And they will see the friction points There will always be frictions And so all you have to do You just keep getting better So I think it's really a moving goalpost And we're from Senzai You know, OpenCloud came out We have Cloud Cowork Which is Anthropics You know So actually, many founders Feel like there's a little dream Because in the process of the development There's a new world that's constantly happening Code 因为I think the problem space comes from people. How I think about this problem space from a technical form product point of view is three categories of problems. You have the technology of agents being able to actually learn from the user's environment. It has to be integrated into where the user already works, where the data comes from lots of files on a computer and integrations. And then the second thing is you have to be personalized. So personalization means you don't really have to prompt it as much. You can trust it to work on more and more complex and difficult things. You can trust it to make more higher stakes decisions. And it also means being more reliable. and that means you can actually use the model and just keep running it for longer and longer periods. And also from a design point of view, it needs to be extremely intuitive to use. You shouldn't have to learn it as like a new tool. If your agents are proactive, they can bring up new ideas by themselves. What is the form that holds all of these proactive outputs together? It has to be with the environment and it needs to be personalized It needs to be able to collaborate well with your existing team So when you look at new tools that come out there not really a new dimension that come out yet because these three dimensions are limited by the human teams. We have to explain to you very briefly why we use Paperboy. Because many people are already in the audience of our podcast, or a user, or an agent of the users. Why do we use Paperboy at this time? Yeah, so since I talk about Humanis, to me, they're kind of the most successful forms of agents out there. They are session-based and they are one-to-one like prompt-based. So these are two very important factors here. Session-based means in their sidebar, there is a bunch of, you know, you have workspaces, which are projects, and then per project, you have a bunch of sessions, right? And every time you want the model to do something new, you start a new session and you categorize into a workspace, okay? The second thing is the way you interact with the model is you prompt something and you wait, but then you just send another message and another assistant, a bunch of turns and another message. The problem with this is one the agent is reactive You have to be very specific You can create a bunch of skills or like doc agent files to describe sort of what you doing but you have to actively maintain that And it doesn't matter how much effort you keep into maintaining that. It's very difficult to translate everything you know about your tastes and judgments and what you believe is the best way to do things into these raw text files. That's one limitation. And the second thing is like sessions are not continuous and it sucks. It sucks to have like hundreds, I think many people even maybe have a couple thousand sessions now. And I know in the past, there are some sessions where the information in that context window might have actually contained some very valuable insights. But if I didn't save it, if I didn't save it directly, deliberately, like here, this is the most important part across all of these different sessions that we worked on this project. This is what we need to take away and take with us into the future. that session kind of just gets lost. Yeah, so to me, that's like, so what is Paperboy? Yeah, so we handle kind of these two problems head on. One is the agents have to learn by themselves from observing how you use your computer. And that includes all the information of your screenshots, your keystrokes, mouth movements, your meetings, which includes videos and audios, your browsing history,
[10:00] or iMessage. And of course, you choose to give access to Paperboy what you want to get it. And the second thing is that it needs to exist in chats with persistent conversation histories that are much longer than a context window. And the chats need to be searchable. And the standard product form is really like, okay, you're just like opening your iMessage or opening a WeChat and you have a bunch of chats. And I go to the chat to keep having conversation with the participants in that chat. But actually, the main difference between the paperboy and the product is that The product context is that it mostly based on the user agent past chat history You can learn from this And there also a question that the user gives these agents your email or messages history learn他们也可以就是说通过读这些email去逐渐的去更加personal去学习关于这些用户的更多信息但是我们一开始做的prototype其实也试图去导出用户所有的微信两线数据或者说iMessage的所有两线数据但是我们很快就发现这其实并不是一个非常scalable的方式然后最scalable的方式其实是通过观测用户在日常使用他的电脑就是从一个OS层面的一个形式去collect user data然后我们其实发现这样子的话会对用户每天做什么事情有一个非常全面的了解而且你按照信息的浓度来说用户的一般的computer use他信息的浓度是非常高的你只要观察用户比如说用电脑用个差不多60分钟就可以学习到用很多的信息但是如果你只观察用户过去60分钟的微信聊天数据其实学不到太多东西所以说我们很早的就是Bets就是说通过OS层面的这个context去做这个user adaptation其实十字路口之前有一期播客的嘉宾LG Alley的Founder他们其实在做的也是一个Mac或者一个Windows的桌面客户端来抓尽可能多的上下文所以你们在做的时候和大家是类似的这个实践吗还是你们有什么不一样的一些想法和做法从Mac还有用户使用的电脑上去抓起 raw dataand processing it into memoriesThat as a general thingis going to happen not just to startups that are deliberately focused on it like AirJelly but Codex, CloudCode work, CloudCode they're all going to do it it's sort of like the most obvious next frontier of context available but of course how people treat that raw stream of data and how it gets processed will happen very differently what you choose to and how you choose to structure that information is directly related tothe application of the agentsyeahso what every individual agentand what your application actually decides to help the user onyou shouldn't probably structureand like you can't compressthe raw activity stream in the same waysand so yeah likejust that alone probably gonna havelike for a company that hyper focusedfor example on on studying um how a user replies to emails right then they probably going to have a different memory structure whatever way of compressing the raw emails that they learn from the action space is big enough that there is enough room right now to be able to be the first to get to that it's really understanding who the user is across all their application services across their relationships and then you can sort of model what the user actually does within a day. Yeah, it's just that capability alone, I think it's just going to be so obvious. Everyone's jumping into that, but depends on the application, there's room to, you know, customize and change the algorithm. I also think that now, through the user's user's computer, or computer use, to build the context, it's a certain extent to become a business. And the last thing is, in this process, what are you doing in this process? 目前像是Codex或者说类似Little Bird的这种 他们会把这个屏幕数据当做一个context layer 就比如说对Codex Chronicle 他其实目前的use case是说 你可以通过就是收集用户屏幕数据 去学习用户一般是怎么去develop这个application的 如果你的use case不一样 他确实就你最后这个pipeline也会不一样 我觉得这是一个非常非常新的一个domain 然后你在基础的就是收集用户屏幕数据 还有收集就是个
[15:00] 比如这个情况下你其实还可以干非常非常多的其他的事情然后这个需要非常多的就是engineering还有research就比如说你如何去做最好的proactive agent你是在预测用户下一个keystroke还是说预测用户后面一个小时去干什么还是说去预测什么其他的事情这些都是relatively就是没有被这么好explore的一些problem space然后目前其实我认为也没有任何人发现一个就是说绝对最好的一个recipe去干这件事情所以说我觉得目前就是对于一个公司人说去探索这个Aware还是一个非常好的选择那如果今天用户安装了Paperboy他立即会比如说在第一个小时甚至在前面五分钟他能够感受到的这个价值最大的是什么你们期待用户在第一时间感受到的差异和价值然后从而能够留下来就这个highlight的点是什么We're probably going to start with a meetings prepand一个小时其实很重要的是你要 demonstrate to users一个匡量 of what your product can do you don't have to set expectation one thing for paperboy it gets better the longer you use it and so you just sort of need that initial period where the user trusts it to learn and so you can open it up you can see a chat box and not a chat box like a prompt box but a literal um chat window and once you give it permissions to say a calendar you know emails it's going to start reading where you get it and then it's going to kind of sort of be asking a little bit of a question about who you are from the context of reading value and it's going to start giving you a couple suggestions like hey i see this meeting coming up do you want to look into that for you. This is the way that you can do it by interaction. They sort of figured out the recipe here, which is, well, you just connect to the user's existing context and you need to show the user you are an agent that can actually engage with the personality and proactively in a non-annoying manner to help the user. So you get the user into this expectation that we are the agent that can actually text you in a way that makes sense. Yeah. Yeah. 你们二位或者整个团队用Paper Boy用了多久了? 在你们自己用的过程中有没有什么? 我可以分享的一些啊哈moment Yeah, so Vivian is also here. Hello. Yeah, Vivian is very funny. So she joined from小红书,她之前在红山的月. So we have Vivian's Paper Boy. It's called Mini Vivian. My paperboy is called AutoJone, and they exist in Slack right now. You know, all the time, teammates on paperboy will just ask AutoJone questions. So it's able to handle all the incoming questions and help teammates, product team, design team, find the help that they need. And for Mini Vivian, for example, Vivian does a lot of recruiting work. And so like Mini Vivian is the recruiting intern for her on the team. Because many of you actually understand everything that I have ever said to it across meetings, Slack, which includes my judgment and taste on what kind of candidates we're trying to hire and where we want to source from. It's able to a lot more actually help Vivian surface these candidates across GitHub, Songshu, Twitter. So I think that saved her a lot of time. I think Vivian since February has not used Cloud. she can't use Claude because it just doesn't know you can't use Claude to research candidates for you like you just there's so many things you would have to tell it about the judgment of the person 因为它有了更多的 context 所以你其实在给它 prompt的时候 你甚至可以不 prompt Yeah, I hate prompting 从现在开始 like I just did not want to prompt I think proompting is such a, I mean, you have to communicate. But people, we don't think in prompts. We send texts, you know, and we expect the other person to know what we're talking about. And we enjoy a relationship when the communication is high bandwidth. We also enjoy, I think the smart people, enjoy being told what we don't know. And especially what we don't know what we don't know. And today's models are, smarter than who we are So frankly I look forward to the day when I can just lay back and or John being the smarter higher IQ existence than I am Okay Jensen呢? You have some kind of aha moments in the paperboard process? First, there are many times with John with the auto-jump conversation better than with John. But you're with the auto-jump when he's agent-to-jump conversation you're worried that his and the way I look at it, not just because we are creators of this thing, but the user has to be responsible for your agent eventually. So with how Autogon gets set up, it's not just like one day, boom, you have this existence that you can bridge into Slack. There is an onboarding process where the agent is going to be able to do it. So the user has to be responsible for the agent. So the way I look at it, not just because we are creators of this thing,
[20:00] This actually asks you questions. And it asks questions like, how much am I allowed to share to this specific person? Right? By default, it mirrors, like for example, I share a lot with Jet. Maybe I don't share as much with a new engineering hire. And the agent actually knows that from observing all my Jet. And I think that usually when you interact with someone, it's actually a lot of different. In the end, you're probably... If you're a person, I sometimes just like with a guy talking to me. Or if I need to do a thing that I don't have the rights, I need to approve a thing. 这些东西我觉得目前可能是就是我不会去和AutoJong说的但是实际上就是在工作场景中你很多时候和对方的一些communication是说这样他有整个对Payboy一个公司的context然后我作为一个工程师或者我作为一个IC就是我每天需要就是我如何能对这个公司产生最大价值我去做什么research或者说做什么engineering是最有用的然后这个时候就因为这个agent第一就是factually他有这样有的大部分context然后第二他通过观测这样工作的时候 the heuristic actually with the job is a bit similar so I was doing the job about context and heuristic the things that I think is more useful and the other aha moment actually I think I have a moment is when we did the text complete actually moment moment Complete就已经挺有用的因为我平常用的现在有很多AI Terminal很火但是就AI命令行很火但实际上就是那些命令行不如就是很多传统的命令行做的就是来流畅但是传统命令行那些APP很多时候并没有就是AI Integration然后很多时候你去写脚本的时候会很烦但是我们做出text completion的时候我完全可以就是说写完一堆代码然后我要去发个get commit的时候我直接就是说我就在命令行里面输入ats pb commits然后他就会自动把这个整个commit message写好然后我enter就可以发个commit这要不要稍微展开讲一讲这是一个什么样的feature因为刚开始我们没有介绍到OK OK我们PingWay就是开发流程大概是说我们先把收集操作系统这个用户数据到就是说有一个比较好的memory系统这一套做出来了然后我们做出这道系统出来之后我们其实在工程上就有一个类似就是framework你都会有一个就是比较好的一个markdown document这个markdown document它是会适时更新的包括这个用户的职业它包括用户就过去几天做了什么 Agent一直会有这个context然后我们有了这对context之后我们其实在做一件事情就是我们在寻找应用场景然后我们找到第一个就是说比较好的应用场景我们可以在操作系统任何一个地方去做一个类似自动补权autocomplete就比如说你在发微信的时候你可以就是说在你的文字框里面输入就是at pp这是个激活词然后后面就是输入一个比较简短的一个instruction或者说prompt就比如说你可以说也可以不输入也可以不输入他就会猜你在这个时候你at我就找我了对吧对你不输入他也会猜一猜你此刻找我是要我来干嘛对因为他有这个context比如说有一个同事过来可能我不需要对他说什么就一个眼神他就懂了为什么我在找他That's great对就比如说很多时候你和你的就是工程师你并不需要就是描述你只要点一下屏幕就你指下屏幕他就知道就是说看这儿对看这儿他一看就哦就这个问题对对然后当时Payrplay已经差不多达到这个效果OK所以说就不管你在回微信的时候做任何事情的时候他都会就是事实的有这个context description description实际上会比cursor或者说Hotcode传统的那些就是会好因为很多时候我开发一个就比如说比较大的feature或者说就是一会儿在和Cloudcode打交道一会儿我在微信里面和John去发什么消息然后一会儿我在浏览器里面去做research就是我也可以把这所有的这个context都聚合在一起然后并以一个PR的形式去给我做这个draft然后我会发现就他这个对我去做什么事情的这个理解程度会比就是任何机具单应用context的这个的这种agent或者AI会考很多还有吗我自我怀疑我有ADHD然后我在工作的时候很容易就是分心我经常就是在那个Cloud Code的时候旁边会看hacker news然后hacker news如果有什么比较有意思的article我可能就是点那个article然后点到时候我看了30分钟回来就是Cloud Code可能已经结束了十几分钟了其实对于这种情况就是我我记得我当时是我每天晚上就是会让Puribway就类似给我一个report就说我今天到底有多efficient我让他会去就是相当于骂我我什么时候就是不够高下我什么时候分神了我觉得对于我来说就是我平常工作的时候当我知道就是Puribway在盯着我工作的时候我会效率更高
[25:00] This is not Big Brother, is what I'm asking. the network effects businesses um you know sort of the successes and failures after the research i want to actually connect to that paperboy at that point i really have a bunch of just like things that's happening in my head and i need something to kind of like help me thread it through yeah it's like connecting the dots yeah yeah and then be able to tell me wait have you thought about this real quick you know you tried this you know maybe a couple months ago and this is the product form and these are the problems that you recognize the product right like i know about these things but they're when you're doing like that sort of thinking across different planes you can do it manually I should just send myself in front of the you know piece of paper I can sketch it out but it a lot faster to be able to have a very smart model that knows me to help me through that process this is the inspiration oh yeah uh i mean paperboy we started working on this new the latest interface two weeks ago it was um so i can talk a little bit about problems we had for example if you're shipping paperboy to vc and vc is definitely part of our ICPs. They want a personal CRM module. They want like meetings coming up module. They want like deals of the quarter or deals from module. They want to keep track of everything that's happening across all the portfolio companies. They want to keep track of their relationship with LPs, right? And so maybe you can ship like a sidebar where you have these like existing modules in place. Then the question is like, okay, we know this thing works for operators, for founders, for real estate salesmen. They all need a different sidebar. So what do we do? Do we ship these things as skills, as plugins, as recipes, right? And allow the user to like pick which one they want to add into their product? So you're thinking about this problem. And then Paperboy helps you. That doesn't make a lot of sense to me, right? So Paperboy pointed out to me that it was sort of like what application do we use that is well you obviously need a list but it an infinite list but it actually not annoying and then it came back to sort of the iam idea which is when you think about uh what's going on in our iMessage and we chats today we chat especially in china there are contacts and there are a bunch of group chats and then there are group chats where it's sort of like maybe you have like three people okay it's just three people, but like your three people can have four group chats for different topics. And it's actually a very intuitive way for people to organize their topics and things that are going on. Actually, we will get a new group and we will get a new group. Every group is a single group. Because there are several speakers in the same way to do the same thing. And then in the first place, it will be so混乱. So every time we open a new group, we will open a new group. Everyone will get a new group. And the great thing about IAM is, you don't have that group in the chat, it will be turned out. You can also choose the right key or the right key I do it all the time So my WeChat is sort of like this inbox So that was kind of the inspiration We know we also need an inbox feature So what if we don't design an inbox feature We just turn what is the thing that itself is an inbox It an IAM So that kind of how the latest interface came about So actually when we were talking about this if you haven used Paperboy it may need to be able to copy We've already explored a lot of Paperboy with different features and features, and unique communication and design. We'll also talk about this here. Actually, as we mentioned earlier, Capture and Context are a common theme. Everyone can do it. And what we mentioned about the new communication strategies or thinking about it I think if it really makes sense I think everyone will also follow up quickly So I want to ask the two how to fight against the competition Do you think the biggest opponent is who? And how do you keep the advantage and change in the competition? computer now and they have great distribution all we can pray for is we're like cursor you know where we have some edge in taste you know where we beat them to you know sort of this new interface and all of that is for is for individuals and product experience that's not going to go away but then also second thing is enterprise
[30:00] Favor Boy definitely want to go into enterpriseI mean that's where the money isFavor Boy一开始其实想做的第一个产品是AI SlackThat was our answer initially to what to buildIf the best way to work with agentshaving been invented yet所以这也是现在非常热闻的一个赛道就是有很多的产品都在尝试解决当agent进入团队以后我们应该build一个什么样的产品去工作所以你怎么看现在市场上一些大家讨论比较多的产品比如说 Slack, Multica, Moxt,等等等等有没有你想讲的 Well first of all I love Slack I think Richard is brilliant he has a great background his co-founders are great and now they're building up a product team as well the hard part about AI Slack and this is from my personal POV how do you get enterprise to switch? AI Slack, the business model that makes sense is you have to sell it to a team it's the team that's paying for it and for enterpriseif there is a wayto exportall the datafrom slackand import itinto slackand it's a one to oneexperiencethen maybe it can workandbut you knowyou still have to builda bunch ofconnectorslike the integratorsfor all the otherdata the enterprisehasyeahso thatinfo workisn't going awaythe se second thing is that Slack is honestly not a great product. You know, like if you just try to build interface that's similar to Slack with channels and threads, to me, I don't enjoy using it. I still, well, you know, obviously at a scale, you kind of have to switch to it because nothing better exists right now. But it's not an enjoyable product to use. So that's all on the user side and getting adoption side. And then there's the agents. As Jed said, we don't believe that in the learning rates of the agents, if the only sources that they get for understanding how to work with people comes from the messages you send to them. People with jobs are busy. They do not have time to care about training agents. They expect things to work for them out of the box. They expect the product to keep getting better. So, and the reason we went away from Slack is that it's just like the agents in Slack is still We need more context for them to learn, you know, naturally Cool I think Slack actually has quite a lot of network effects Just like the John said, to stop They're being replaced by new players For example, Slack has a function called external connections Currently, all of the more transparent companies They actually use Slack And the communication between the complayer去替代掉的就比如说slack它其实还有一个功能叫external connections目前就是所有比较认真的公司其实都会用slack然后公司和公司之间的交流其实现在很大程度上并不是email而是通过external connections去进行然后这个是很难被替代的我觉得slack肯定还有就很多其他的这种integration方面的这些network effect就和现在微信一样让他们到一个就是某种程度上不可替代的一个方式然后相比就是如果你去把agent作为一个就是a layer above更加就是在OS层面或者说和Slack形成一个互补形式的一个东西的话我觉得反而会有更好的一个这个adoption story所以这是你们在做的事情吗就是尝试去和Slack有一些互补Yeah, we assume that initially if you try to go into enterprise there will be communicating on SlackIf you're building a Slack right now you can't really try to sell a thing to a team larger than maybe 50 peoplehow much it would take for them to switch off a Slackit would be a bit of work and it makes more sense again to bring the agents to where people already work instead of trying to replace and build everything from scratch我其实有注意到这个PaperBall目前的官网啊产品没有发布但是有放了一篇blog post然后这个标题叫做The Last Interface然后在那篇blog里面其实有提到这个5 kinds of speed就5种ive kinds of speed, so I can open it up a little bit. I understand this behind me, it seems like you're talking about how to understand the context layer and the memory system. Yeah, I don't want to go into too much detail, because actually the choice of using five isn't because we only have like five layers, it's just because that was like a sensible categorization. The important thing is like if you think about, Okay, so in SF, there is a nonprofit called the Langnao Foundation, and its founders do a brand that's published something called Pace Layers. So it's sort of like at the top, it's like fashion, and then there's like commerce, politics, or governance, and then infra, and then it's like civilization, it's like nature. And these are six places that changes at different velocities. So fashion obviously changes as fast as nature. Our understanding is that it doesn't really change physics. And the world is made up of these kind of different paces and sort of all have a relationship with each other. And they're reflexive relationships between every layer and largely the makeup of our society. I have been kind of fascinated with that since high school. And it's influenced definitely my worldview of what is the world that we live in, both the natural and social systems.
[35:00] and you know when I think about what's actually efficient and what can scale for representing this world that we live in in the agent's world you need these different you know pace layers too yeah so it's quite interesting that what will do the same for nature? what will do the same for fashion? I think for the product to use if you use the task to make the task of the task to make a separate way 其实会有蛮多不同的种类 就比如说用户在微信上回一条消息 他只需要一秒钟或者说十秒钟 然后我觉得这是Spectrum的一端 然后Spectrum的另一端可能是说 用户去读十个非常非常长的一个report 然后去做一个business decision 去做一个决策 然后这个可能要花个几个小时 That's probably like closer to the left side too The long side is more like over a month Yeah like over a month Yeah 就basically有个Spectrum 然后我们认为就这个Spectrum I think it's every time they will be some kind of AI agent to augment or self-image But every time they need to do the things, it's not the same If you want to self-image, you can use the form factor You can use the form factor You can use the form factor You can use the form of the form This is what we have done But the time horizon is more and more Actually you have a lot of product Horizon你去自动化用户一个几个小时Task这个最好的form practice是什么这个其实目前还是一个非常值得被探索的一个区域我们也很感兴趣其实在录制一些播客之前我和Paperboy的团队也有一些交流然后发现团队里面好几个成员都有讲说他们认为每次和Jump沟通的时候感觉不到他是04年的就觉得他有超越年龄的成熟我实际上有这样的感觉我想问一下你觉得自己的这样的状态它是怎么来的是因为你过去做了什么或者看了什么或者被什么激发激励I think I'm still a pretty childish personespecially with chat aroundYou knowWe have a lot of fun togetherI think you're oldI think for the team at leastI definitely just have a passion for business我们要做什么很难的在很短的时间所以 clarity efficiency and focus这要怎么做到呢这个是一型难的那开始从我 define success job is to bring to the team define what success looks like and be able to make sure that the context every person has about their goals and how they actually operating are complete and accurate. Actually, even for me, the best part of being able to start a company is to get to learn so much of what I've always wanted to explore and just understand about the world. And that curiosity has probably been the biggest driver for me since high school you know between high school and college i took a couple internships and all of that are at different companies i just do different type of problem solving activities and then million obviously we started in the b2b business right selling yeah the web performance you know pr bots to large enterprises and then we learned that it's a difficult business to scale. We as 20-year-olds are not good at scaling up of enterprise relationships. So since then, I think I've been on this trajectory where I'm just trying to learn about, okay, what are the markets that actually make sense for different players? And the way I like to learn is not just for startups, but across different business scales to go with it. It like a founder market fit Yeah team Team market fit So I think market picking is one of the most important if not the most important trade of a founder early on It's making sure that you understand what you are going into, where are all the leverage points, where are all sort of the challenges coming down on you. And then you can actually go build a team, and then it's a product experience, you know, finding customers, technology. So here you have something to share with you, how to choose a market, the right market. Yeah, I think I'm about to say a bunch of common sense again. It has to be big enough. And this big isn't just dollar sign, but how sustainable is the market? And that scale has to keep increasing so your team can continuously develop your product and your market position across a decade. So when I look at the most successful products in history, it's never the single product. It's actually sort of like a product lineup where the expectation is when you build this thing, it's going to take maybe a decade for it to get to a form where it's really mass consumer or everybody's using it. Yeah, that's the first thing. You got to start with a problem close to you. It's all just sort of like a, to me, a method of understanding market. But then you need a lot of history to be able to understand if the business dynamics, you know, the economics are actually favorable and there have been success.
[40:00] best cases in the past or not. Yeah, there's so many details, and I think I am just barely scratching the surface. Yeah, Paperboy is just my chance to kind of prove out what I know. Paperboy, you should be the first CEO. Because the company is a co-founder. So what do you think of the biggest change in your CEO? Man, it's fucking hard. I gotta own everything. I mean, as a co-founder, you definitely feel that too. I get frustrated when an engineer, for example, waits like an afternoon. But as CEO, no, I am actually the person that can cost the company the most amount of money because I have so much leverage, right? If I say we go do this and that is actually not a good idea, it's the position where literally all of your ideas can have such a large cost and reward. And then as the team scales, obviously, who do I choose to hire? Where do they come from? And then once they're hired, how do you continuously improve together for the individual and for the team? You know, making sure every person is obviously on the road that they deserve. I feel like these are basic management skills. This is basically my first time as a manager, too. So a lot of that I have to figure out on the spot. How do you give performance reviews How do you have one How do you make sure that you know sometimes I have to just talk to the team leader right and not to everyone and micromanage Before Paperboy, the one stint I had where I actually stayed at a larger company was Amanis. Amanis, man, Panpan is an amazing, amazing CTO. I mean, I haven't worked with any other, you know, real engineering managers, but he is a great engineering manager. And we've had the opportunity when I was a man is to have a couple one-on-ones where I just ask him all the questions I wanted to ask. How do you organize everything, man? And then I also spend time just like reading. I think the best management book I've ever written is still just like high output management by Andy Grove. And then Ben Horowitz, you know, he has the thing hard thing about the hard things. And then Bill Campbell, the trillion dollar coach, great biography. And I got myself a CEO coach. We meet weekly for like an hour. That's helped me a lot, especially like the first few calls, just helping me, having a space sort of to talk about all of the specific problems that I have. How did you find her? Investor intro. Okay Yeah It actually the same coach for the investor too Can you probably introduce her to her But then you can actually also talk about business You can talk about everything that's happening in the business And so you know it's confidential Also it's been really fun to use Paperboy With the coaching calls Because Paperboy listens to the calls And it can actually help me follow up On things that I committed to doing On calls and organizing everything Yeah So you know how do you see For example today Because you said there are 10 people But actually today Not every team has so many people You have no small team size. So I'm curious, in your opinion, in the situation of today, there are so many tools, AI agents, to build a team and to build a team and to build a team. There are some big changes that are happening. If you try to make a serious infra structure build out you need people who understand infra I think we hired two kinds of people on the engineering side One is like Jet who are like young, high IQ, creative, just prototypes of shit out of every problem that he sees. And then second kind is like, you got to be real solid at what you do. You got to understand the systems. You got to understand the OS. So we hired this guy out of AWS who worked on like Windows kernels for the AWS like infra for the running window. I don't think that's going to go away anytime soon The human experts其实这个Jet现在大一对吧对我们看到硅谷有一些公司比如Palantir直接给高中生offer让他们不要读大学了直接工作因为认为在AI时代不需要大学了那很显然其实Jet也是优秀的这个大一学生但你也在PaperBall是Funding Engineer那你为什么现在还要读大学啊He's got Chinese parents意义的话其实对大部分来说大学还是能提供非常大的意义因为 至少我在CMU这边看见就大部分其实和我同学的人他们其实并不是很确定未来想要做什么事情 然后你在不确定的情况下就是你可以做的是什么你可以提升你自己的就是technical能力上大学是个比较好的一个选项 就学业只会占你就相对比较少的一个时间你大部分其他时间就是说是你自己决定的然后相当于四年的时间
[45:00] 去做这个exploration然后这个四年时间其实是对大普普来说会是比较有用的但是同时就如果你在上大学的时候已经很确定你想要做什么事情然后并且就是说你做事情这个方面确实机会我觉得辍学也是一个理性的选择对这个Jump你做一个非常年轻的founder也是first time founder你怎么吸引到同学们愿意加入你的就是你要怎么说服大家这样什么方面吸引到我你关键到的因为对大家来说职业选择是非常重要的这个人生投资很多时候你选一个founder也是你有intercept你有slope然后很多时候slope比intercept重要很多然后我觉得John是一个就是high slope的founder怎么说就是他现在就如果你看他的这个track record的话他现在就差不多比我大所以相当于他就是在一个很小的时间内就是说实现了特别多比较厉害的事情然后就这一定程度上他是比较就比较有agency然后就是他这个可以很快的去学习然后并且可以去做非常多的新事情然后对于一个创业者来说这个arguably是最重要的一个quality就是说你得要能你头脑心心你得要能就是很快的去适应各种的情况首先你得要认为他是一个就是good founder对于我来说就首先我和张已经就是工作了比较久所以说就是我在和他合作我有比较高的confidence说就是我们可以就是比较顺畅的合作然后张他也有不错的research engineering甚至product的这个taste然后我觉得就是说和他在工作的过程中我可以就是进去很多这种经验然后让我个人也有更好的成长我个人觉得张还是就算比较就是charismatic的一个人大型表白现场 I appreciate it So how do you tell yourself Do you have some ways to summarize your抽象比如你现在去 approach一个新的 candidate你会有一些什么样的方法去说服他 I treat every candidate individually and there's not like a script I stick to It has to be mutual I never like to sell the company to a candidate like hard sell and recruiting is one thing what's more important是关于我的信息我认为我其实会负很多关注是在我会的我会叫做一个人我会把我送给你我会把我送给你可以把我送给你让我送给你我会送给你你给我送给你你给我一个人的你给我送给你你给别人的你给别人的你不要跟别人的你不要跟别人的你可以从从你的经济从你的经济从我的经济尤其是 You've got to pick people who are great, man. You can't talk to mediocre people. And you can tell that from conversations about their past work. And what's especially important is their past life decisions. I really like to learn about, you know, if they're older, they have a decade sort of decisions behind them. And you can really learn a lot. You can't fake that. Can you share your most of your AI products? Apart from Paperboy. Yeah, I still love Cursor, man. I still don't use Codex. I really like Codex. Cursor has always been a great inspiration to me. I'm really happy for them. I hope that whatever acquisition happens go through. This is John's first co-founder, when he was a billion, that was actually a purchase from Devin. That was from Devin's company, Cognition. Yeah, there were other offers too, like from Brussels and Century. Wow. So you never had these offers. Yeah. And at that time, you can tell us about how to purchase offers. 首先, these are aqua hires. You know, whatever offers for the talent, there's not going to be a lot of money. Okay. And you're, you know, you're joining the companies and sort of just like another employee. And you have to work on somebody else's ideas. I've never been interested in debt tools that much. So if you ask me to join Sentry or join Vercel or join Cognition as an individual, I think I wouldn't do it, right? So it doesn't really make sense. If you can stay independent and if you can achieve your dreams like by staying independent, you should stay independent, right? And that's what every founder wants. Let's say back to you. You mentioned the most favorite thing is Cursor. And then Codex. Can you tell us why Codex is codex而不是 Cloud Code Cloud Code它理念不一样就是它那个相当于以最ambitious的形式就是说你未来就是说这个soft engineer会是怎么work我觉得 Cloud Code它那个最重要的东西是体现Opus有多好不是说它那个命令行有多好我appreciate那个Opus的模型我不是很appreciate Cloud Code那个DLI本身Codex我喜欢它几个点一个就是说它核心的agent是开源的它的命令行还有那个整体的agent loop它实际上都放在Github上然后如果你用他们那个桌面端的app的话 把那个中边端的app去和就比如说高购那个中边的app比他那个整个app就是他那个craft的水平就是会非常非常好他们前面买了一个团队叫sky然后sky就是他们是做apple shortcuts的那个团队然后open 就是会还是这种人去做一个非常polished的产品然后你去看codex的话你会发现他们就比如最近出了一个叫codex test的一个东西
[50:00] 就非常polished然后也很好玩然后包括Codex的computer usebrowser use其实都做得非常polished然后我觉得Codex其实也是第一个就是高并行式的就是说你不同agent的同时去工作的一个东西当做主要UI的一个产品然后也push出来然后还有就是OpenAI在那个infra层面其实比anthropic优势大非常多所以说如果你用Codex的话模型大差不差但是稳定很多大意思就是OpenAI在infra上面其实可以subsidize更多的compute然后anthropic他们现在就是投资的不够所以他们没法subsidize然后导致了一系列风波就比如说你的repo里面有什么open cloud或者说hermes agent然后cloud code可能会直接收你十倍的价格反正就anthropic现在因为compute受限去做很多奇怪的操作如果可以买他们的secondary就可以买他们的股票你现在会买open AI还是买anthropic还是你会各三期开买还是怎么样我感觉我会买open AIAll in OpenAI可能就是比如说OpenAI 80%然后Antwerp的20%OK什么原因呢就是他们两个公司都会do very well但是我更喜欢OpenAI在做的事情我觉得Antwerp做的最好的事情是他们的interpretability还有他们的social impact那那两个部门做得非常非常好然后我特别喜欢那两个东西但是anthropic在就比如说很多模型我觉得他们过于opiniated就是说用户该怎么用这个language world会非常就比如说不管是道德上还是具体使用习惯上会非常opiniated然后相比之下open AI的理念是说就是在一个最小的一个局限性之下就是用户想要怎么用这个模型就怎么用然后我个人会比较libertarian我会喜欢就是OpenAI这种更加unapenient的这种values然后还有就是我觉得OpenAI在compute上面会占优势I'm not gonnaIf I'm making investment decisionsI'm not gonna buyeither of themat the current valuationBut I actually preferAnthrope'sI think safetyCommitment and safety那你们怎么看就是模型吞噬一切这个说法我觉得你完全可以argue就是很多公司会提供差不多的模型然后产品的差异性还是在产品公司里面就是这不一定会是真的但是我觉得这是一个possible的然后还有cursor我觉得是一个不错的一个example就是他们一开始产品公司但是现在他们既是产品公司又是模型公司他们现在可以用XAI的这个compute去就是frontier coding model所以我觉得就是公司不能把自己局限为就是模型公司还是产品公司it's usually both. If you can invest 300 million dollars, you can invest three people in the world. Slack is good. Just a bet on Richard. Robotics companies, I think that's definitely going to be big. And then Infra, people who are focused on Infra. And then I believe the biggest company, similar to Cursor, is going to come out of consumer. It's not going to come out of enterprise AI. It's going to come out of consumer. And obviously Paperboy is in that market because of all that belief. But yeah, consumers are such a big market. There's so many opportunities. I think robotics will be very helpful. Because I think in this kind of automation, The problem with that is how do you compete with Anthropic getting into that business? Anthropic OpenAI will have the best model. The problem is that you have a good design After the optimization layer, what is the best way to do? What is the best way to do? Is it a better way to harpec OpenAI肯定会有最好的模型问题是就是你有最好模型之后你下了一个optimization layer是什么你是做比较好的harness吗你是把人类security researcher的各种就是heoristic big到agent里面还是说你去做infra就比如说非常庞大的一个multi agent系统去不断的去挖这些漏洞我觉得这个探索空间会非常大就比如说就你这个产品挖的漏洞会比另外一个产品比如说多个10%那么你这个产品就是非常有用的这是一个非常 meritocratic的一个 market所以我觉得这是个不错的我们还没有说过Long Horizon emails yet但Harvey shipped this thingLong Horizon law然后我认为其他张杨阳 shipped this new sweet bench今天或最近 reproducing complete co-bases instead of stocks or specs那是不会走过我觉得安全能力对于国家安全是非常非常重要的一个事情但我觉得对于创业来说这也是非常好的一个 opportunity因为就是模型最强的人他有这个incentive去把这个模型说只有我们自己国家可以用这个东西那么相当于你其他国家怎么搞呢你其实很多很多不同的就是国家或者说力纪元都需要这种模型所以说这个market是非常大的我们最后问个问题你会期待就一年后的这个
[55:00] 此刻Paper Boy是一个什么样的状态? 你有一些什么样的期待? 或者你有一些什么样的害怕做错的事? Well, I hope we're not losing money every day anymore. I want to keep hiring the best people I can find. So the talent density needs to be higher than, you know, where we're at today. I mean, there's so much specifics we can talk about, but I think from a team and like business point of view, it's just that we actually need cash flow, and we need to keep building a better team. Okay,好,谢谢二位. 谢谢,非常开心. Bye bye
