Automating Developer Email with MCP and AI Agents
发布时间 2025-03-21 14:00:14 来源
这段对话记录了两位创始人之间关于开发者体验 (DX) 和“代理体验” (AX) 的深入探讨,尤其是在电子邮件服务领域。他们讨论了从以人为中心的应用程序开发到未来以代理(AI 实体)为主要参与者的转变,这需要从根本上重新思考产品设计和可访问性。
对话开始于这样一个观察:现有的基础设施、数据库和通信工具大多是为人类构建和使用的。然而,人工智能代理的日益普及标志着一种范式转变,要求开发者创建专门为代理交互量身定制的体验。讨论强调,开发者不仅需要为人类构建服务,还需要为代理构建服务。
开发者最初的反应是这非常困难,因为他们不仅需要为人类构建服务,现在还需要为代理构建体验。创始人指出,AX 不是要取代 DX,而是要增强它。就像 DX 是所有小细节的总和一样,AX 也是这些细节的总和。创始人在此讨论中提到的一个最重要的例子就是启用发送电子邮件工具的入门过程。在代理开始发送电子邮件之前,它必须先加入平台或系统。
讨论转向了电子邮件领域内的具体挑战和机遇。它提出了这样的问题:代理是否会有自己的电子邮件地址,代表人类处理电子邮件,或者需要不同的编程服务使用说明。他们用自动驾驶汽车在现有道路基础设施上行驶进行类比,暗示代理最初将利用现有的 API、SDK 和协议,而不是要求彻底改革。
演讲者深入研究了电子邮件格式的细微差别,强调了纯文本相对于 HTML 对于代理处理的意外价值。纯文本电子邮件有更多有用的标记,更容易解析,而且成本更低,因为它是一种不太复杂的格式。他们还讨论了代理环境中 API 密钥和权限授予的演变,思考是将它们视为个人用户还是服务帐户。
对话强调了在正确的时间向正确的人发送正确消息的必要性。他们提到了 AI 驱动的 SDR 工具的作用,指出基于 LinkedIn 消息的简单定制是不够的。不同电子邮件客户端之间存在的渲染挑战也得到了认可,强调了保持一致显示的重要性。
讨论进入了准专业级应用程序领域,以及人工智能驱动的内容创建的日益普及。发言人指出,LLM 能够执行各种任务,重要的是要考虑的不仅仅是生成代码。使用代码只是“顿悟时刻”,现在开发者可以利用一个新的行动号召。这种事情很重要,因为它能够定制任何给定环境中正在发生的事情。创始人讨论了如何从电子邮件的想法开始,并在几秒钟内获得电子邮件模板。
他们探讨了如何授权消费者和开发者创建集成电子邮件体验的潜在用例,并提到了“文本到应用程序”工具的趋势及其对重新定义开发者角色的影响。创始人提到他们构建了一个名为 new email 的工具,它可以帮助人们在几秒钟内从一个想法变为电子邮件模板。这不仅可以帮助需要学习编码的开发者,还可以帮助营销人员、设计师、产品经理等。
然后,他们开始讨论 React email 作为一个开源项目,以及他们对电子邮件构建方式进行现代化的目标。关键是将 TypeScript、React 和 Teyawen 集成到这个行业中,以帮助它向前发展并不断创新。借助 LLM,这可以缩短非技术用户的创造性循环,并使他们能够构建更好的东西。
作为 Uber 的一名设计师,他们提到他们已经使用 Cursor 或 V0 或任何工具构建了所有原型,并且他们甚至不再打开 Figma。他们所需要做的就是专注于文案,并思考角度。总的来说,他们强调缩短了创意循环,因为它让那些最有创造力的人能够专注于创意方面,并更快地完成它们。
讨论探讨了生成式体验的本质,质疑它们是否符合代理工作流程。他们将代理定义为执行以完成特定任务的一组工具,可能涉及多个步骤。
最后,他们探讨了 MCP(元聊天协议)作为人工智能代理通信的新兴标准。他们讨论了 MCP 有望简化与各种 API 的交互,使代理能够跨不同平台访问和操作数据。
他们设想的未来是人工智能代理执行大部分在线操作,因此需要转变产品设计,优先考虑代理的可访问性、快速入门和安全权限授予。记录代理活动对于理解和减轻潜在问题至关重要。对话强调了开发者需要调整他们的工具集和工作流程,以适应这种新的现实。
This transcript features a conversation between two founders deeply invested in developer experience (DX) and exploring the evolving landscape of "agent experience" (AX), particularly in the context of email services. They discuss the shift from human-centric application development to a future where agents (AI entities) are the primary actors, requiring a fundamental rethink of product design and accessibility.
The conversation begins with the observation that much of the existing infrastructure, databases, and communication tools are built for and used by humans. However, the increasing prevalence of AI agents signals a paradigm shift, demanding that developers create experiences tailored to agent interaction. The discussion emphasizes that developers need to build not only services for humans, but also for agents.
The initial reaction from developers is that it is very difficult because not only do they need to build services for humans, they now need to build experiences for agents. The founders note that AX is not about replacing DX but augmenting it. Just like DX is the sum of all the little details, AX is also the sum of those same details. One of the biggest examples the founders mention in this discussion is the ability to onboard onto the tool to enable the sending of email. Before an agent can begin to send email, it has to onboard onto the platform or system.
The discussion shifts to the specific challenges and opportunities within the email domain. It questions whether agents will have their own email addresses, processing emails on behalf of humans, or require different instructions for programmatic service utilization. They use the analogy of autonomous cars navigating existing road infrastructure, suggesting that agents will initially leverage existing APIs, SDKs, and protocols rather than demanding a complete overhaul.
The speakers delve into the nuances of email formats, highlighting the unexpected value of plain text over HTML for agent processing. Plain text email has more useful tokens, is easier to parse, and costs less because it is a less complex format. Also they discuss the evolution of API keys and permissioning in the context of agents, pondering whether to treat them as individual users or service accounts.
The conversation highlights the need to send the right message to the right person at the right time. They touch on the role of AI-powered SDR tools, noting that simple customization based on LinkedIn messages is insufficient. Rendering challenges across different email clients are also acknowledged, emphasizing the importance of consistent display.
The discussion moves to the realm of prosumer applications and the growing accessibility of AI-driven content creation. The speaker notes that LLMs are able to perform a variety of tasks and it is important to think about it more than just generating code. Using code is just the aha moment and now there is a new call to action that developers can utilize. This type of thing is important because it enables the ability to customize what is happening in any given setting. The founders discussed how you can start with an idea for an email and in second, you can have the email template.
They explore potential use cases for empowering both consumers and developers to create email-integrated experiences, referencing the trend of "text-to-app" tools and their impact on redefining what a developer is. The founder mentioned they built this tool called new email that really helps like to go from zero from like an idea to an email template in seconds. This can enable not only developers that needed how to code, but marketers, designers, product managers, etc.
They then get into a discussion surrounding React email as an open source project, and their goal on modernizing the way emails are built. It's all about integrating TypeScript, React, and Teyawen into this industry to help it go forward and continue to innovate. With LLMs, this can shorten the creative loop for those non-technical users and enable them to build better things.
As a designer at Uber, they mentioned that they've built all of the prototypes using cursor or V0 or whatever the tools are, and they don't even open Figma anymore. All they need to do is focus on the copy and thinking about the angle. In general, they emphasize that it shortens the creative loop because it enables those who are the most creative to focus on creative aspects and do it faster.
The discussion explores the nature of generative experiences, questioning whether they qualify as agentic workflows. They define an agent as a set of tools executing to accomplish a specific task, potentially involving multiple steps.
Finally, they explore MCP (Meta Chat Protocol) as an emerging standard for AI agent communication. They discuss the potential of MCP to streamline interactions with various APIs, enabling agents to access and manipulate data across different platforms.
The future they envision is one where AI agents conduct the majority of online actions, necessitating a shift in product design to prioritize agent accessibility, rapid onboarding, and secure permissioning. Recording agent activities becomes crucial for understanding and mitigating potential issues. The conversation underscores the need for developers to adapt their toolsets and workflows to embrace this new reality.
摘要
In this episode of AI + a16z, Resend founder and CEO Zeno Rocha sits down with a16z partner Yoko Li to discuss: - How ...
GPT-4正在为你翻译摘要中......
中英文字稿 
When you go to all these different apps, you go to SuperBase. The most databases you see there were built by humans. You go to recent, the emails were sent by humans or drafted by humans and then sent programmatically. I think we're going to see a big shift in terms of who is the actor, who is the creator, and I believe it's going to be the majority of the actions we'll be taken by agents instead of humans. And that's just the reality we're going to live in. So we have to rethink the way we're building product to support that reality.
当你使用这些不同的应用程序时,比如SuperBase,你会发现大多数数据库是由人创建的。你查看最近的记录时,会发现邮件是由人撰写或草拟后,再通过程序发送的。我认为未来会发生很大的变化,即谁是行为的推动者、谁是创作者。我相信,未来大部分的操作将由智能代理而非人类完成。这将成为我们生活中的现实。因此,我们必须重新思考如何构建产品,以支持这种现实。
Recently, we've been seeing a lot of agent-centric applications and then developers cannot utilizing LMS to build not new experiences. And then what's interesting is that a lot of the developers are met. They were like, well, nowadays not only do I need to build services for humans, where the access is top of mind, they now need to build experiences for agents. As a founder in the space who's been building developer experience for all across the spectrum, what's your high level stuff here? What have you seen?
最近,我们看到越来越多以代理为中心的应用程序,但开发者无法利用学习管理系统(LMS)来建立新的体验。有趣的是,许多开发者发现,他们不仅需要为人类开发服务,还需要为代理构建体验。作为在这一领域构建开发者体验的创始人,您对此有何看法?您观察到了什么?
Yeah, I've been obsessed with developer experience for the past 10 years. And now you can see that there's a shift and there's a new obsession for me when it comes to agent experience. As a product, as a SaaS provider ourselves, we are thinking about how we can make the product easier for agents to consume. So even small things like adding a recapture on the sign-up to prevent bots, now you have to think, do you really want to prevent bots? Right. Not from signing up, maybe you don't, right? And same as an infrastructure provider.
是的,在过去的10年里,我一直非常关注开发者体验。而现在,我的关注点有所转移,我开始痴迷于代理体验。作为一个产品和SaaS服务提供商,我们正在思考如何让我们的产品更容易被代理使用。因此,即使是像在注册过程中增加验证码以阻止机器人这样的小事,现在我们也需要重新思考,是否真的要阻止机器人?没准你不想阻止机器人注册,对吧?基础设施提供商也是这样。
So, Resand is an email API for developers. It helps people's central transaction and marketing emails. So I'm also thinking a lot about, okay, how can we make the experience of these agents sending emails the easiest as possible? So there's a lot to uncover on each spectrum of product when it comes to making it more agent-friendly, but it's fascinating. It's like we're all trying to figure it out right now.
Resand 是一个为开发者设计的电子邮件 API。它帮助用户处理事务性和营销类的邮件。我在想,如何能让这些代理发送邮件的过程尽可能简单?在提升产品对代理友好度的过程中,还有很多方面需要探索,但这确实很有趣。就好像我们现在都在努力找出最佳方案。
Agent experience is such an interesting term, because like why here the term agent experience, I wonder, is this like a developer experience, but for agents or is this a developer experience that's better for the end human developers to build agents or perhaps is somewhere in the middle? How would you define the term? I think about developer experience as the sum of all the little details. And that's what makes either great developer experience or a bad developer experience, right? So for agents, I believe it's the same.
“代理体验”是一个很有趣的词,因为为什么会用这个词呢?我在想,这是像开发者体验一样的东西,但只是针对代理的吗?还是说这是一种能够让最终的人类开发者更好地构建代理的开发者体验?或者可能是介于两者之间的某种体验?你会如何定义这个词呢?我认为,开发者体验是由所有细节汇总而成的。而正是这些细节造就了出色或糟糕的开发者体验,不是吗?对于代理,我相信情况也是一样的。
You know, you're trying to build a world where agents are the first class citizens. And when you put that in perspective, then you have to rethink a lot of the different things you do. So as an example in our industry, you go to San Grete, you go to Postmark, to Mayo Gun, you sign up for these services. And the first thing you get is a manual verification process that takes two days for you to get approval of your account, right? In an agent first world that's simply not possible. Like an agent will never wait two days to take an action.
你知道的,我们正在努力构建一个代理(自动化程序)成为主要角色的世界。以这个视角来看,我们就需要重新思考很多我们所做的事情。举个例子,在我们的行业中,当你使用像San Grete、Postmark和Mayo Gun这样的服务时,你注册后首先要经历一个需要两天才能通过的人工验证过程,对吧?但在一个以代理为先的世界里,这样的事情是不可能发生的。代理永远不会等待两天才能采取行动。
So recent from day one was all about like removing friction, right? So you sign up, you can send an email in the first minute. When we built that, it was all because we thought this is a better product experience over all for users, turns out this is a huge unlock for agents. So I think it augments GX. It's not a replacement or anything like that. It's more like everything you already do for, for example, docs, right? The more you invest in docs and knowledge base, the more quality of the LLMs that you're going to have, like in terms of the output there.
从一开始,我们的目标就是消除摩擦。你注册后,在第一分钟内就可以发送电子邮件。我们这样设计,是因为我们认为这会为用户提供更好的产品体验,结果证明这也是代理商的一大助力。我认为这提升了GX的表现,而不是替代它。这就像你为文档做的所有事情一样,当你在文档和知识库上投入越多,就能提高大语言模型(LLMs)的输出质量。
So the things you do for one already help you for another. It's so interesting, especially that email, it's like a service every human developer needs either personally or as part of what they're building. Because once someone signs up, you want to send them welcome emails on and so forth. So when I think about agents, what do agents need in the world of email? Do you think of agents as users of the email? Like in the future, every agent will have their own email address and they'll process the email for the two humans?
因此,为一个人而做的事情已经可以帮助到另一个人。这点非常有趣,尤其是电子邮件,它就像是每个开发者无论是个人还是作为他们构建项目的一部分都需要的服务。当有人注册时,你会想要发送欢迎邮件等等。因此,当我思考代理的时候,代理在电子邮件的世界里需要什么呢?你是否认为代理会成为电子邮件的用户?比如在未来,每个代理都会有自己的电子邮件地址,并负责处理两个人之间的邮件?
And or do you think agents will also kind of require a different level of instruction when it comes to leveraging the service in a programmatic way? Using the analogy of autonomous cars, right? Waymo, they couldn't build new roads. They had to figure out a way to use the existing roads to make sure like their product would work. So I think the same for this, like we already have APIs, we already have these established SDKs and protocols. So we're going to have to figure out a way to leverage those instead of creating everything from scratch.
你是否认为代理在以编程方式利用这项服务时也需要不同级别的指导?就像自动驾驶汽车的例子,Waymo不能建造新的道路,而是必须找到方法利用现有的道路来确保他们的产品能够有效运行。我认为这同样适用于这种情况:我们已经有了API、SDK和协议,所以我们需要想办法利用这些现有资源,而不是从头开始创建一切。
But I do believe there are like fundamental differences and limitations as well, right? So there's a reason why LLAMs.xt exists as a format. Because it's so much easier for LLAMs to consume just plain text versus all the different HTML and react code together with docs, right? All you want is the content. Same on the email world. Like whenever you're sending an email, you only have like two formats you can send. You have HTML and you have plain text as well. And the plain text is like the old school format that, you know, people don't really care as much today, right? Turns out in today's world, that's actually better than the HTML version. Because it's easier to parse. You're using less tokens. It's cheaper. So I do think there's a fundamental difference.
我确实认为这是有一些根本性的差异和限制的,对吗?这就是为什么有LLAMs.xt这样的格式存在的原因。对于LLAMs来说,处理纯文本要比处理包含HTML和React代码的文档容易得多,对吧?你只需要内容。在电子邮件的世界中也是如此。当你发送电子邮件时,只有两种格式可以使用:HTML和纯文本。纯文本是那种老派格式,如今人们并不太在意,对吧?但事实证明,在当今世界,纯文本实际上比HTML版本更好,因为它更容易解析、使用的令牌更少、成本更低。所以我确实认为这里存在根本性的差异。
And even the way you think about API keys and permissioning, is it a user or it's a service account? Like I think there's a lot of challenges in. But you have to figure out a way to leverage the existing product surface. Yeah. The LLAMs.xt is such an interesting example. Because like the fact that it exists is because LLAMs just do better with this kind of information. And then as a human, if I want to copy paste the entirety of a site, I wouldn't write a script or to go copy it. It's very hard. I'll go to LLAMs.xt and copy paste it and put it in my chat GPT. I guess in the case of email, because historically, people have been sending a lot of very pretty marketing emails to humans.
即便是思考API密钥和权限管理的方式,也是一个用户还是一个服务账户?我觉得这其中有很多挑战。但你必须找到一种方法来利用现有的产品界面。LLAMs.xt就是一个非常有趣的例子。因为它的存在,是因为这种信息格式下,LLAMs表现得更好。作为一个人类,如果我想复制粘贴整个网站,我不会写个脚本去复制,这是很难的。我会去LLAMs.xt复制粘贴然后放到我的ChatGPT中。我想,在电子邮件的例子中,因为在历史上,人们已经发送了大量非常漂亮的营销邮件给用户。
And humans react to that because we're very visual animals. How would it change now if we're sending emails to agents? Yeah. Content remains the king. I guess the difference now is that we're going to be living in the world where agents are talking to other agents. And these agents are sending emails, but also receiving and parsing and sending to another inbox that is also powered by agents that's parsing, taking action and sending. So the more content focus these emails are, the less markup, the fact that if they have a plaintext version of it, I think this is more important than ever.
人类对此有反应,因为我们是非常依赖视觉的动物。那么如果我们发送电子邮件给代理人,这会有什么变化呢?内容依旧是关键。我想,现在的区别在于我们将生活在一个代理人与代理人交流的世界中。这些代理人不仅发送电子邮件,还接收、解析,并发送到另一个由代理人驱动的收件箱,这个收件箱也能解析、采取行动并发送。因此,这些电子邮件越注重内容,标记越少,并且如果它们有纯文本版本,我认为这比以往任何时候都重要。
Yeah, that is so interesting. I've been thinking about generating applications, generating even a CLI tool. I'm a huge fan of cursors. I use it every single day. At the same time, I keep wondering what does it look like if I can incorporate other long-tailed tooling started to use as a consumer? To generate other interesting experiences. What's your view here when it comes to like, consumers or consumers generating like net new contents using AI? I think my view changed drastically two months ago when I was answering users were signing up to reset. And I always asked them like, oh, where'd you come from? What are you doing? And this one day, people were just saying like, oh, it came from loveable. It came from bold. It came from V0. ChatTPT recommended you. Cloud recommended you.
是的,那真是有趣。我一直在考虑生成应用程序,甚至是生成一个命令行工具。我非常喜欢鼠标光标,每天都会使用。同时,我也在想,如果把一些更长尾的工具整合在一起,用户在使用时会是什么样的体验呢?这样能否产生其他有趣的体验?你对于消费者或用户利用AI创造全新内容有什么看法?两个月前我的看法发生了很大的变化,当时我在回答用户们的注册问题。我总是问他们是从哪里来的,在做什么。有一天,有人说他们是从Loveable、Bold、V0来的,还有人说是ChatGPT和Cloud推荐了我。
So I saw that happening more and more every single day. And it was just so clear that there's a new world happening in front of us. And these pro-sumer apps, those text to apps, applications, they're very interesting because they focus on this one vertical. They help this type of user build this type of website, for example. I think we're going to see that across the board. For every single industry, there's going to be one of those. The output might be a little bit different, but you're always going to need something that is tailored to that industry.
所以我看到这种情况每天都在越来越多地发生。显然,一个新的世界正在我们面前展开。这些专业用户应用程序,比如那些文本到应用程序,非常有趣,因为它们专注于一个特定的领域。它们帮助特定的用户建立特定类型的网站。我认为这种情况会在各个行业中普遍出现。虽然输出可能会有一些不同,但你总是需要一些专为该行业定制的东西。
So when I think about these apps, you almost see the same UI. You have like chat on the left, the preview on the right. But what really matters is what's the thing on the top right corner, like that button, the call to action. And for both, lovable V0, it's publishing. It's going live with the website. That's the aha moment. It's not generating the code or building the to-do app. It's actually putting it live. And I think there's going to be apps for each one. And that's the main part of each one.
所以,当我想到这些应用程序时,你会发现它们的界面几乎相同。左侧是聊天区域,右侧是预览。但真正重要的是右上角的那个按钮,即所谓的“号召性用语”。对于初版应用来说,最吸引人的地方就是发布上线。那一刻才是“灵光乍现”的时刻,而不是生成代码或构建待办事项应用。实际上,是将其上线才是关键。我认为,每个应用都会有自己的独特之处,而这就是每个应用程序的核心部分。
What's a new call to action if we think about the email world? I guess like why I think about email. I was thinking about either on the receiving end or on the sending end. On the sending end, if I'm a non-technical user, I want to make sure that things are structured correctly, sent to the set of people. And I want to make sure they open the email. If I'm receiving end, I want to make sure that I get the information I want quickly. Or this is a joy to read. I guess like in the AI world, how would you think about it? Does it change? Is it like the same set of abstractions where kind of playing with just how I'll make it more efficient?
如果我们考虑电子邮件领域,有什么新的行动号召吗?我猜想为什么我会考虑电子邮件呢,我在想的是无论是接收端还是发送端。从发送端来看,如果我是一个非技术用户,我希望确保事情被正确组织,并发送给特定的人群。我还希望确保他们打开电子邮件。从接收端来看,我希望能快速获得我想要的信息,或者阅读这封邮件能让人感到愉悦。如果是在人工智能领域,你会怎么看待这件事?这会改变吗?是否仍然是同样的一套抽象概念,只是我们在考虑如何提高效率呢?
The challenge is always like sending the right message to the right person at the right time. So customization is more important than ever. You see all these AIS-SDR tools. And it's great that we're seeing a revolution there. But just customizing based on your last LinkedIn message. It's not going to be enough. When you're sending email, you have all these different challenges around rendering. So for an email to render the same on Outlook and Gmail and Yahoo Mail, you still have a lot of challenges. So that's one part that needs to be figured out. And then the other part on the receiving end is like, yeah, do you really care about this? And if I do, is this email on the primary box and on this pampholder? Yeah. So there's a lot of challenges on both sides.
挑战总是如何在正确的时间将合适的信息发送给合适的人。因此,个性化比以往任何时候都更加重要。你会看到各种AI-SDR工具的出现,看到这方面的变革是很棒的。然而,单单基于你最后的LinkedIn消息进行个性化定制是不够的。在发送电子邮件时,你会遇到许多不同的渲染问题。比如,为了让一封邮件在Outlook、Gmail和Yahoo Mail上显示的一致,你需要解决很多难题。这是需要解决的一个方面。另一方面,接收者是否真的关心这封邮件?如果关心,这封邮件是出现在主收件箱里,还是在垃圾邮件文件夹中?是的,在这两方面都有许多挑战。
Yeah. And then there are so many interesting experiences one can build with email, whether it's from a developer or a consumer. So as an example, my husband and I, we built this app that sends emails to us and texts us when our cat jumps on the kitchen counter. And then it was like, it was not much code to be written, but it was just a lot of joy to build because everyone needs to be notified of something. And I guess like in this example, from all the text to app, kind of a side of applications, what kind of fun experiences can you imagine that we can better empower consumers and consumers to do. That's better incorporated with emails. What would they be able to do that they may have discovered or what can they do that they haven't discovered?
是的,使用电子邮件可以构建许多有趣的体验,无论是作为开发者还是消费者。举个例子,我和丈夫一起制作了一款应用,当我们的猫跳上厨房台面时,它会给我们发送电子邮件和短信。写这段代码其实并不复杂,但构建这个应用充满了乐趣,因为每个人都需要被通知些什么。我想在这个例子中,从所有基于文本的应用程序来说,我们能想象出什么有趣的体验来更好地赋能消费者和用户,这些体验怎样更好地与电子邮件结合呢?他们可能已经发现了什么,或者还没发现可以做些什么呢?
So in our case, we started this open source project called React email two years ago. And that project, the whole vision around it was, you know what, we need to figure out a way to modernize the way emails are built. So it was all about like, how can we bring TypeScript and Teyawen and all these modern technologies React? Let's bring all that together to this industry that feels like it's not moving forward or not innovating as much. So we did that and it was great. A lot of people using that every single day. But now that we have LLMs, there's a new unlock. So this process that used to take days now takes hours or minutes. But with LLMs, it can take seconds, right? And it can enable not only developers that needed how to code, but these folks that can be just like, I'm a marketer, I'm a designer, I'm a product manager, and I need to send an email. So how can I do that?
我们在两年前启动了一个名为 React Email 的开源项目。这个项目的愿景是找到一种现代化构建电子邮件的方法。我们的目标是将 TypeScript、Tailwind 以及其他现代技术如 React 引入到电子邮件行业中,因为这个行业似乎没有显著的进步和创新。我们实现了这个目标,并取得了很好的效果,许多人每天都在使用它。
但是,现在有了大型语言模型(LLM),我们面临新的突破。过去需要几天的过程,现在可以在几小时或几分钟内完成。而通过 LLM,这个过程甚至可以在几秒内完成。这不仅对需要编程的开发人员有帮助,还能让市场营销人员、设计师和产品经理等需要发邮件的人受益。那么,他们怎样才能便捷地完成这些工作呢?
So we build this thing called new email that really helps like you go from zero from like an idea to an email template in seconds, right? I just see that happening across the board. So whatever the industry is, I think they will have a variation of that, you know what I mean? That's so interesting. In a way like, would be a great way for say, if I'm a marketer or a designer, I want to send an email, like what's the best way to leverage existing tools like new email or others to kind of send high quality emails? I remember this conversation I had with a friend of mine. This is a designer that works at Uber, and he was telling me how like for the past two weeks, he didn't open Figma. He was just building all these prototypes, using cursor, using V0, using these tools that are now available for folks.
所以我们开发了一个叫“新邮件”的工具,它能让你从一个想法到电子邮件模板快速完成,几秒钟内就能实现,对吧?我看到这一模式跨行业普遍存在。所以无论哪个行业,我觉得他们都会有一个类似的版本,你明白我的意思吗?这真是有趣。比如说,对于市场营销人员或设计师来说,这是个很好的方式,如果我想发送一封邮件,如何利用现有的工具如“新邮件”或其他工具来发送高质量的邮件呢?我记得我和一个朋友的对话。他是Uber的设计师,他告诉我,在过去的两周里,他都没打开过Figma。他只是利用现在人们可以使用的工具,比如Cursor、V0等,来构建所有这些原型。
And for me, that's incredible. Like that’s so amazing because it redefines what a developer is. I think that's exactly the same for email. Before you needed to maybe hire an agency and you would build this beautiful email template, and then you hand off the Figma file to them, and then a week later, they come with this super ugly markup. The email looks good, but behind the scenes, the how to do a lot of magic to make it happen. We can give that to Ella Lambs now, and empower the actual creator, the person who's thinking about the copy, thinking about the angle, because as a builder yourself, you know that when we're building things, the first version, it's not always the best, it's just the beginning.
对我来说,这真的令人难以置信,非常惊人,因为这重新定义了开发者的角色。我觉得电子邮件也是一样。以前你可能需要聘请一个机构来创建一个漂亮的电子邮件模板,然后把Figma文件交给他们。一周后,他们把超级难看的代码发给你。虽然邮件看上去不错,但背后其实动用了很多“魔法”才做到这一点。现在,我们可以把这个任务交给Ella Lambs,让真正的创作者(即那些思考内容和角度的人)得到赋权。作为一个建设者,你知道我们在创建东西的时候,第一版通常不是最好的,它只是一个开始。
So you see that thing working, and you're like, okay, now that's working, what else can I do? And that back and forth between agencies and non-tactical users, it's just like waste of time that it could be just focused on that creative person that's just keep it and raining on that until they have something they like, and then they can just send. I love that. It really shortens the creative loop. Yes. It just reminds me that, you know, back in the days, obviously, like for oil painting, people like to throw some paint on the canvas before they get started, because the blank canvas is super scary. I don't know what to talk.
所以你看到那个东西在运作,就会想,好吧,这个现在有效,那我还能做些什么?而且,机构和非战术用户之间来回沟通,就像是在浪费时间,本可以让有创造力的人专注于不断完善,直到他们满意为止,然后再发送。我喜欢这一点。这真的缩短了创造的周期。是的,这让我想到,以前在画油画时,人们喜欢在开始之前先往画布上涂点颜色,因为空白的画布让人感到非常恐惧。我不知道该说些什么。
Yeah. So a lot of painters, what they will actually do is to throw gray colored paint on the canvas, and then they will use like a little bit of cloth to like spread it. And then they're like, okay, now there's a bit of color. I'm like, less scared to actually make a action. This actually feels a lot about that.
是的,很多画家实际上会先在画布上涂上一层灰色的底漆,然后用小块布料把它抹开。这样一来,画布上就有了一点颜色,他们就不那么害怕下笔。这种方法让我有种很相似的感觉。
Yeah. You know, it's funny. Yes, this is something that happened yesterday. So we gave this product, the new email product to a couple of friends, and then one friend was like, you know what, I'm trying to build something very creative, you know, but it takes me like a couple of prompts until I get something that feels different. Like the beginning is always like, it feels the same like this, too, is always generating the same kind of emails. Right. And that's because on the system prompt, we told, you know, the LLM to build Apple-like emails with curved borders and black and white vibes versus, you know, go crazy. And we thought we were doing a favor to include that, because then the first version is good, but that also, it's a constraint on the creative person that's trying to build something way more creative, you know.
是的,你知道吗,这很有趣。这件事是昨天发生的。我们把这个新的电子邮件产品给了几个朋友,然后其中一个朋友说,你知道吗,我在尝试做一些非常有创意的东西,但我需要几次尝试才能得到一些看起来不同的东西。开始的时候总是觉得生成的邮件看起来都一样。这是因为在系统提示中,我们让大语言模型生成类似苹果风格的邮件,有弧形边框和黑白基调,而不是完全放开规则。我们以为这样会帮上忙,因为这样生成的第一个版本就很好,但这对尝试做更有创意的人的确也是一种限制。
So it's fascinating. Like, yes, you need the first version to be there, and then you keep it a rating. But if it takes too long for you to iterate, that's also not good, right? There's a balance there. That's so interesting. It really reminds me of in the image generation world, there are different loras, translora, unspecific style, say, impressionism. And then the end user can select any one of these loras and then, you know, create the first generation of an image that's like, like these kind of styles they predefined. And then they can iterate from there.
这实在是很有趣。是这样的,你需要有一个初始版本,然后对其进行迭代。但如果迭代花费太长时间,也是不好的,对吧?这里面需要有一个平衡。这让我联想到图像生成领域中有不同的“lora”,比如说,印象派风格的“translora”或是其他不特定风格的。最终用户可以选择这些“lora”之一,来创建某种预定义风格的第一版图像,然后在此基础上继续迭代。
And then the rest of it's just an editing process. You need to have a last tool somewhere. Like, there's AI versions of all of these now, right? And then the question is like, how can you morph that into something you like, kind of like a sculpture? So yeah, this is like so interesting. Like email templates as where artistic expression, because before it was so hard for people to do this themselves. Last time I looked at email templates, it was all vanilla HTML. And I just remember thinking, this can't be because it's this 2025 now. Remember why I first got into the industry? It was like that it just has a evolved much.
然后剩下的就是编辑过程。你需要在某处有一个最终工具。现在已经有这些工具的AI版本,对吧?关键在于如何将其转化为你喜欢的东西,有点像雕塑。所以这真的非常有趣。就像电子邮件模板成为一种艺术表达,因为以前人们很难自己做到这一点。我上次看电子邮件模板的时候,全都是普通的HTML。我记得当时心想,这不可能,因为现在已经是2025年了。回想我为什么最初进入这个行业,因为它似乎没有太多的演变。
So it's really great that you guys are, you know, creating a new experience at a different level of abstraction too for a new audience. Yeah, you think about how much the web evolved as a platform. And it's, you know, the problems we used to have 10, 12 years ago around browsers rendering different websites, you know, that's gone. Like that's, you know, a jacks and like IE six would render different than opera and Firefox. That's solved. But with email still a big pain, outlook still doesn't render the same as superhuman as notion male.
所以,很高兴看到你们为新受众创造了一种在不同抽象层次上的全新体验。你想想看,网络作为一个平台已经经历了多大的演变。过去在大约10到12年前,我们常常面临浏览器渲染不同网站的问题,比如以前的IE6与Opera和Firefox的渲染效果不同,但这些问题已经解决了。然而,邮件领域的问题依然很令人头疼,比如Outlook的渲染效果仍然不同于Superhuman或Notion Mail。
Like, you have new email clients, you know, being created right now. Right. So it's still an open problem. Like people are still trying to figure this out. Yeah. Speaking of generating things, do you see the generation experience as a agentic workflow? How do you feel about it? Do you think you're building an agent? In many ways, I think yes. And then in other ways, I'm like, is this really an agent? What's your definition of agent?
当然,你知道,现在有新的电子邮件客户端正在被创建。所以这个问题仍然没有定论。人们还在尝试解决这个问题。说到生成东西,你觉得这个生成体验是一种代理工作流程吗?你对此有什么看法?你认为自己在构建一个代理吗?在很多方面,我觉得是的。但在另一方面,我也在想,这真的是一个代理吗?你对代理的定义是什么?
Yeah. I think it's just, yeah. And I don't know if it's the right one. There's no right definition. No one knows. But I think it's just like a set of tools that are being executed and they're trying to accomplish a specific task, right? So it might take a few steps to get to that final result, but it's still like very focused on doing one thing. So I think like in the case of new email, for example, we have one agent to build the email template.
是的。我觉得,就是这样。而且我不知道这是不是正确的定义。没有人知道。但我觉得可以理解为一组工具正在被执行,它们试图完成一个特定的任务,对吧?所以可能需要几个步骤才能达到最终结果,但它仍然专注于做好一件事情。以新邮件为例,我们有一个程序专门用来制作邮件模板。
And then there's another one to actually send it or schedule it. So you can have like multiple agents running at the same time. Right. But they are still very focused on doing one task. Just like, you know, in a company, you would have one person at the smart lane, one person at the right. Design in your like assigning tasks to these people. What is your definition?
然后还有另一个步骤来实际发送或安排它。因此,你可以让多个代理同时运行。但是,它们仍然非常专注于完成某一个任务。就像在公司中一样,你会有一个人在智能通道上工作,一个人在右侧工作。设计时就像是在给这些人分配任务。你对此的定义是什么?
So my definition of agent is kind of like very similar to yours. I think of it as a multi-step LM execution process. It's almost like a processing the systems level. The difference is that you have LM in the middle to make the decisions. And then the rest of it is very much technical detail. On the one spectrum, you can have a one step generation process. You could call it an agent, because that's what co-pilots are.
所以,我对“智能体”的定义与你的非常相似。我认为它是一种多步骤的语言模型执行过程,几乎就像是一个系统级别的处理。区别在于,在这个过程中,你将语言模型放在中间来做决策,而其余部分则涉及很多技术细节。在这个范围的一个极端,你可以有一个一步生成过程,也可以称之为智能体,因为这就是协作助手(如 co-pilot)所做的事情。
On the other spectrum, you could call it a G I an agent. Like that's like an agent that reads and writes everything for you. Maybe you manage your bank account when you're in the mouth. Yep. And awesome. So on and so forth. So for me, I've been just, you know, as a fellow developer kind of developing the space, there's so many different ways to build agents. You know, if you're a platform, you're thinking about how can I make it easy for agents to visit us? How can I make it easy for other people to build tools around it?
在另一方面,你可以称它为一个GI代理。就像一个能够为你阅读和写作所有事务的代理,也许还能在你忙的时候帮你管理银行账户等等,对,就是这样,非常不错。所以就我个人而言,作为一个开发者,我一直在这个领域探索,发现有很多不同的方法来构建代理。对于一个平台而言,你可能会考虑如何让代理更容易访问我们,以及如何让其他人更容易在你的平台上构建相关工具。
And then if you're like a developer trying out certain tools, you're going to work with one of the platforms. The question becomes like, how do I easily integrate the long tails? I don't want to rewrite all the API calls all over again. I guess what's your view? Like what's your advice to developers who want to start building agents like yourself? Is there a standard today? Like how do you think about this?
然后,如果你是一个开发人员,正在尝试使用某些工具,你可能会使用其中一个平台。问题就变成了,如何轻松整合各种小众需求?我不想重新编写所有的API调用。你怎么看待这个问题?你对想要开始构建类似你这样的代理的开发人员有什么建议吗?目前有没有统一的标准呢?你是怎么考虑这个问题的?
Yeah. I think the emerging standard is definitely MCP. There's still like the question of like, is MCP going to be adopted by other AI models? And if that's the case, that will be an even bigger locked in it already is. You know, MCP is on fire now. Everybody's talking about it. Imagine if open AI adopts that then it's it's over. You know, like that will be the de facto protocol. But there's a lot still to be explored in terms of the different interfaces.
是的,我认为新兴的标准肯定是MCP。不过还有一个问题是,其他AI模型会不会采用MCP。如果采用,那它的影响力会比现在更大。你知道,MCP现在非常火,人人都在谈论它。想象一下,如果OpenAI也采用了MCP,那就确定无疑会成为默认的协议。不过关于不同的接口,还有很多需要探索的地方。
So as an API provider myself, like I want to make sure we have an MCP for recent. So all their agents can use that and send emails. They can look at their contact lists and take action, add people to the contact list, remove, you know, see the performance of the emails they're sending. And based on that, they can then decide like, oh, this email is being clicked more than this other one. So let me optimize for that. So there's definitely things to explore around that. But it's an ever evolving ecosystem, right?
作为一个API提供者,我想确保我们有一个用于最近更新的MCP(管理控制平台)。这样,他们所有的代理都可以使用这个平台来发送电子邮件。他们可以查看他们的联系人列表和采取各种操作,比如添加或删除联系人,查看他们发送的电子邮件的表现。基于这些信息,他们可以决定,比如说,哪封邮件被点击的次数更多,然后对此进行优化。在这一领域有很多值得探索的内容,但这始终是一个不断发展的生态系统,对吧?
There's no right answer yet. And you see some like some people trying to build galleries of MCP servers and trying to like create the marketplace for MCP. That's also interesting. Like the same way that when APIs came out, people were like rapid API, they were like, oh, let's build a marketplace of APIs. I think it's good to assert an extent. At the end of the day, I have a problem. I'm going to find whatever the solution is. So if I'm thinking about building, I'm going to go to this stripe API.
目前还没有确切的答案。你会看到有些人在尝试建立MCP服务器的画廊,并尝试为MCP打造一个市场。这样的做法也很有趣,就像当API刚出现时,人们创建了快速API平台,也想要建立一个API的市场。我认为这种做法有一定的好处。但归根结底,当我遇到问题时,我会去寻找解决方案。所以,如果我要构建某些东西,我就会去使用Stripe的API。
If I think about email, go to the recent email API. But the MCP format, it's extremely interesting. But we have to see what are the other interfaces that this is going to expand to. Just cursor and cloud desktop is not enough. And that's what we have today, right? And win server like these other editors. But what else? And I think if we get to the consumer layer, then it can be very, very interesting.
如果我想到电子邮件,就会使用最近的电子邮件API。但是MCP格式非常有趣。不过,我们需要看看这种格式还会扩展到哪些其他接口。仅仅有光标和云桌面是不够的,而这就是我们今天所拥有的,对吧?还有像这些其他编辑器一样的服务器。那么还有什么呢?我认为,如果我们能扩展到消费者层面,那将会非常有趣。
Yeah. So there's obviously MCP servers. It's kind of a philosophical question now. Are you a MCP server? Are you a MCP client? Because you can be both. You have to pick. What's your view on that? Do you think your build like it's new email or recent both or one?
好的,所以显然存在MCP服务器。这现在有些像是一个哲学问题:你是一个MCP服务器,还是一个MCP客户端?因为你可以同时是两者,你必须做出选择。你对此怎么看?你认为你的构建像是新邮件还是最近的,或者是其中一个?
Yeah. Yeah. I think we'll be both. I can definitely see a world where we are the MCP client where you can come in and say, you know what? Get my top 10 linear feature requests and then draft an email based on that and then send. So you have like three different services running to execute a task or go to notion.
是的,是的。我认为我们可以同时扮演多个角色。我完全可以想象这样一个场景:我们是MCP客户,你可以进来然后说:“你知道吗?整理我前10个线性功能请求,然后根据这些请求起草一封邮件并发送。”所以你可能需要运行三个不同的服务来完成一个任务,或者去使用Notion。
Like we are running all of our all-hands meetings on recent now. So it's like one email that we all write in a multiplayer approach. And then by the end of the call, we just send the email. So it's super fun. That's what you say. And yeah, like I see that kind of like workflow where I'm like, okay, just go to linear grab the tickets we closed last week. And then let's use that as a reporting mechanism or from notion or from whatever.
就像我们现在把近期的所有全员会议都这样进行。所有人都会以多人协作的方式写一封邮件,然后在电话会议结束时,我们就直接发送邮件。这样做非常有趣。你是这么说的。是的,我觉得这种工作流程不错,比如我会去 Linear(一个软件)查看我们上周完成的任务,然后用这些作为报告机制,或者从 Notion(另一个软件)中获取信息,或者其他什么地方。
So yeah, I think we're going to be both. So most of the use cases I've actually seen MCP today, it's very local first because of the nature of how the MCP clients are implemented. First, SSE is kind of a pain. So most of the people kind of default to implement MCP as a command. So as a client, the client basically just execute that command locally. And then that process can call into other third party APIs.
所以我觉得我们会兼顾这两方面。目前我看到的大多数MCP(多客户端处理)用例都主要是本地优先的,因为MCP客户端的实现方式决定了这一点。首先,服务器发送事件(SSE)有点麻烦,所以大多数人默认会将MCP实现为一个命令。因此,作为一个客户端,客户端基本上只是在本地执行这个命令。然后,这个过程就可以调用其它第三方API。
What do you think is missing to make MCP more of an ecosystem? I mean, obviously now it's a ecosystem. But like what's missing from pushing it for even more? I think it's adoption by the other models. I think that's the biggest one. There's so many different frontiers that MCP is getting to know that I love, for example, access to your file system, you know, to Apple APIs that are running on your desktop.
你认为让MCP成为更完整生态系统还缺什么?当然,现在它已经算是一个生态系统了。但要让其更进一步,还缺少什么呢?我觉得是其他模型的广泛应用,这是最关键的。有很多新的领域是MCP正在探索的,这让我很兴奋,比如访问你的文件系统以及能够使用运行在桌面上的Apple API。
I've just seen like the latest release from Raycast where you're building all these AI extensions and you're integrating all these different workflows that run on your desktop layer, not in the browser. As we keep exploring that, I think that's fascinating because we keep going down, you know, this different abstraction layers and having more access to do more things.
我刚刚看到Raycast的最新发布,他们正在构建所有这些AI扩展,并且把各种不同的工作流集成到桌面层,而不是在浏览器中运行。我们持续探索这一领域,我认为这很吸引人,因为我们不断深入到不同的抽象层,并且获得更多的权限去做更多的事情。
So go to Apple Notes, grab my notes and then do this other thing. I think that's fascinating. And I hope that people continue to adopt. What's your prediction on the kind of MCP workflows that will take off? Obviously, there's a very productivity focus like getting notion notes, put in an email send a email or send emails from cursor.
去苹果备忘录,把我的笔记拿出来,然后做其他事情。我觉得这很有趣,也希望人们继续使用。你认为哪种MCP工作流程会流行起来?显然,非常注重生产力,比如从Notion获取笔记、放入电子邮件并发送,或者从光标发送电子邮件。
It's like every day like kind of workflow. Do you think there are around time workflows by that? I mean like when a surface is running and CP server will stepping in and do something, I actually haven't seen that happen in production, but just curious what your thoughts are there. I think all of the system of records type of applications, they will be front and center at this, you know, new revolution. Like because all of your issues are already on linear, all of your emails are already on Gmail, all of your notes are ready on notion or Apple notes. So they're going to be in a very good position to then, you know, based on that information, do this other task. Interesting. So there's definitely a data gravity. I think so. To it. Yes. That's so interesting. Do you see most of the MCP related, like clients or server around their own databases? Wow. That's a good one because you need to store state.
这段话大概是说:
就像每天工作流程一样,你觉得是否有实时的工作流程呢?我的意思是,比如当一个系统在运行时,CP(检查点)服务器会介入并做一些事情,我在实际生产中其实没有见过这种情况,只是好奇你对此的看法。我认为所有记录系统类型的应用程序会在这场新的革命中扮演重要角色,比如因为你所有的问题已经在Linear上,你的邮件在Gmail上,你的笔记在Notion或Apple Notes上。因此,他们会处于一个很好的位置,可以基于这些信息完成其他任务。很有意思,因此数据重力在其中很重要,我是这么认为的。是的,这很有趣。你是否看到大多数与MCP(多处理器计算)相关的,比如客户端或服务器,它们围绕自己的数据库吗?哇,这是个好问题,因为你需要存储状态。
Exactly. Yeah. Yeah. I haven't thought about that, but I think that will be a pretty interesting approach. I think when you look at even we're talking about this tax to apps applications, right? So you have lovable V0 and these apps they're using different databases. So for lovable and bold, to the super base for raplet and create.xyz it's neon. And I think the race with those is going to be around like who can reduce friction the most who can spin up a new postgres database faster. So I think in a world where MCP is at the front and center, then in it, yeah, like how can we have these databases? There's, you know, this pin up so quickly we can save state and then maybe we'll need a new kind of database.
好的。是这样的。是的,是的。我还没有考虑过这个,但我觉得这可能是一个非常有趣的方法。我认为,当你看到我们正在讨论应用程序与数据库的交互时,这很有趣。比如说,你有一个初级版本的应用程序,这些应用程序使用不同的数据库。例如,lovable 和 bold 使用 super base,而 raplet 和 create.xyz 使用 neon。我认为,竞争的关键在于谁能最大限度地减少阻力,谁能更快地启动一个新的 postgres 数据库。所以在一个 MCP(多方计算协议)处于核心的位置的世界中,我们要考虑如何让这些数据库快速启动并保存状态,也许我们还需要一种新的数据库类型。
Yeah. That's very true. Obviously I've been playing this. Recent is always recommended by agents at all. Because the distribution of the training data, I guess it's there's just so much of that. Every time I ask like chat GPT like, oh, I want to draft this email and just gave me a react email template right away. I'm like, how did you notice? Where did it come from? When you think about more creative use cases, developers can do either with agents with our agents or some other L and driven processes with recent what comes to mind. You know what? There's just a lot of systems that developers rely every single day. Like GitHub, you know, they are storing their poor class in one place, but they also have branches that are running locally on their machines that they haven't pushed to a remote system yet. The combination of that stop and web, I think that's beautiful.
是的,这很对。显然,我一直在使用这个。最近代理们总是推荐这个。我猜是因为训练数据的分布,非常庞大。每次我问像ChatGPT这样的工具要起草邮件时,它都会立刻给我一个React邮件模板。我想知道它怎么注意到的,信息从哪里来的?当你考虑更有创意的使用场景时,开发者可以通过我们的代理或其他学习驱动的流程来实现。你知道吗?开发者每天都依赖很多系统。比如GitHub,他们把代码存放在一个地方,但也有一些还没有推送到远程系统的本地分支。我觉得这种本地和网络的结合很美妙。
There's so many fascinating angles that you could do. I saw a demo last week where someone was using like a raycast extension that was not powered by MCP, but very similar in a way when you look at implementation. And it was integrated with Bob like an HR system. So if we go to this HR tool and you would only ask like, oh, yeah, like who is celebrating a birthday this month? So based on those four people like, let's draft a message that's unique about each one of them. Wow. Or look at the different interests from these people based on there's lack messages on top of, you know, this HR data. Yeah. So you can go like, you know, there's so many interesting use cases where you can just pipe data from one place to the other and then you can get something that's extremely tailored on the other end.
有很多有趣的角度可以探索。我上周看到一个演示,有人在使用类似射线投射扩展的工具。虽然它不是由MCP驱动,但从实现上看非常相似。这个工具与一个叫Bob的HR系统集成在一起。比如,如果我们使用这个HR工具,你可以询问“这个月谁过生日?”然后根据这四个人的信息,起草一条关于他们各自独特的信息。或者从这些人的兴趣出发,根据HR数据顶部的消息来分析。这样你可以把数据从一个地方传输到另一个地方,从而得到非常个性化的结果。
I guess in the past few weeks to your past couple of months, what is the craziest use case you have seen that developers or AI developers use recent work? Yeah, there's some really interesting AI generated newsletters that are being triggered for recent. I saw one last week where every single day they generate a new one and they send just like the content is so interesting because you can fast from so many different places. You get like the latest news from X and the latest news from like different publications and then you put all together and then here you have like a newsletter that's tailored for you versus, you know, like this big poll that you're trying to almost like, please everyone and you end up not pleasing like anyone. So there's something about that that I really love like, yeah, like there are people with thousands of domains in one recent account and they're spinning up new domains as new applications are built.
我猜在过去的几周到几个月里,你有没有看到开发者或AI开发者使用最近的成果去做一些疯狂的事情?最近有一些非常有趣的AI生成的新闻简报让我印象深刻。我上周看到一个,每天都会生成新的内容,并且发送出去。这些内容非常有趣,因为它们从很多不同的地方汇集而来。你会得到像是X处的最新消息,还有来自不同出版物的最新动态,所有这些整合在一起,就像是为你量身打造的新闻简报,而不是那种试图迎合所有人的大杂烩,最终可能谁也没取悦到。我真的很喜欢这一点。确实,有人能够在一个最近的账号里拥有上千个域名,并且在新的应用程序建立时不断添加新的域名。
So there's just very different domains. Can you do that on recent? You can. Yeah, programmatically. Yeah. Yeah. I think pre-warming domains or getting new domains. Yeah, so for example, payload CMS, every time you sign up for one of their cloud product, so they have like this, it's like a WordPress alternative and you can just create a cloud version of it instead of deploying your own server. Yeah. Every time they provision one of those, they provision a new domain, powered by recent and then email sending is already done for you like all the configuration, all the SMTP setup, the domain is there already verified, you don't have to add a DKM or SPF record. It's all there.
因此,有非常不同的领域。可以在新的领域上实现吗?可以。是的,使用编程方式。是的,是的。我认为可以预热域名或获取新域名。比如说,Payload CMS,每次你注册他们的一项云产品时,他们提供的就是一个类似于 WordPress 的替代品,而你可以创建一个云版本而不必自己配置服务器。每次他们配置这样的服务时,他们会开启一个新的域名,由 Recent 提供支持,而邮件发送等功能已经配置好,如 SMTP 设置和域名验证,你无需再添加 DKM 或 SPF 记录。这些都已经配置好了。
So from day one, you have email sending capabilities, which is something that traditionally was very hard to do. I can't wait for the day when agents can have their own domain, like just thinking about how many agents site projects. Yeah. They will feel that they don't utilize way more than humans. Maybe they will come up with like new project ideas and they'll be like, oh, let me see if the domain is free. Oh, yeah, it is. Bye. Totally.
从第一天起,你就具备了发送电子邮件的功能,这在传统上是很难做到的。我迫不及待地希望有一天,智能代理可以拥有自己的域名,想想看有多少代理项目网站。是的,他们会觉得这些资源比人类还要充分利用,也许他们会想出新的项目创意,然后就想着,哦,看看这个域名是不是可用的。哦,是的,是可用的。太好了。
Yeah, I saw human developer, like this is just like something that I always love doing, even though I don't use, I've probably spent hundreds of dollars every year just paying for domains I don't use. The other day, a friend of mine, he sent me a text, he was like, did you know, Dalai, L.I. It's a top-level domain. I didn't know that. It happens to be my last name. And Yoko is not a very common name. So I was able to, you know, the domain right away. So I now did redrips to my website or I think I tried to get the .no, which is the Norwegian TLG. I couldn't Z.no, I couldn't get, but I'm still trying. Was it because someone took care of the ISC?
是的,我看到了人类开发者类似的事情,这就是我一直喜欢做的事情,即使我不使用它们,我可能每年花费数百美元购买不使用的域名。前几天,我的一个朋友给我发了短信,他说:“你知道吗,.L.I.是一个顶级域名。” 我之前不知道这件事。碰巧这是我的姓氏。而且Yoko不是一个很常见的名字。所以我能够立刻注册到这个域名。然后我为我的网站做了重定向,或者我试着去获取.no域名,这是挪威的顶级域名。我没能拿到Z.no,但我仍然在尝试。是不是因为有人已经处理了ISC的事情?
Well, we can write an agent to get domains. Just check every day if the domain is free and then let me know when it's good. And then just buy. Do you have advice for developers who are now navigating this whole AI landscape who are building their app for the agents or using the agents or just entering the AI domain? Like, how do you tell them they should focus on what matters the most nowadays?
我们可以编写一个代理程序来获取域名。每天检查一下域名是否可用,然后在合适的时候告诉我。然后直接购买。对于那些正在探索AI领域的开发者,无论是构建应用程序供代理使用、使用现有代理,还是刚进入AI领域,你有什么建议?你会如何告诉他们在当下应该把注意力集中在哪些最重要的事情上?
I think in my case, it's interesting because I'm both a developer and a founder. And I have like many different hats that I have to wear every single day. And for the past two years, I've been ignoring AI, which was a certain extent. Like, I knew the power. I was a user, power user, but I was never building with AI because we didn't have time. We had to like, build the company, right? So I had to like just one day stop and be like, okay, let me look into this thing.
我觉得在我的情况下,这很有趣,因为我既是开发者又是创始人。每天我都需要在不同的角色之间切换。在过去的两年里,我基本上忽略了人工智能。虽然我知道它的强大,也算是个资深用户,但我一直没有用AI来做开发,因为我们没时间,我们得全力去打造公司。所以有一天我不得不停下来,决定好好研究一下人工智能。
And I think it's around like changing your tool set in a way. Like, okay, I'm used to VS Code. I have to switch to cursor. Even for a little bit, you know, even if it hurts in the beginning, like, I remember my extensions were not ready. And I was like, I hate this cursor thing. Like it broke my whole workflow. Now I can't leave without it. Same with ReCast and the AI on the desktop level, you know, and same with so many other tools.
我认为这就像是改变你的工具集。有点像,我习惯了使用 VS Code,但我必须切换到 Cursor。即便只是稍微用一下,甚至即使一开始有点难受。我记得我的扩展都还没准备好,我想,我讨厌这个 Cursor 东西,它打乱了我整个工作流程。但现在我离不开它了。在桌面层面上使用 ReCast 和 AI 也是如此,还有许多其他工具也是这种情况。
So I think it starts from there, like just looking at your two chain and thinking, how can I add AI enabled apps? And then the second part is just like trying to figure out the use case. Yeah. So I remember a conversation I had with the Stripe team. And they were building the MCP server. And I asked them, how are you approaching it? Like the Stripe API is so big. There's so many endpoints. This is Stripe MCP server gives agent access to their two accounts. Wow. That's very powerful.
我认为事情就是从这开始的:先看看你的流程,然后思考如何将人工智能应用程序添加进去。接下来的步骤就是弄清楚具体的使用场景。我记得我曾经和Stripe团队讨论过。他们正在构建MCP服务器,我问他们如何处理这个问题。因为Stripe的API非常庞大,包含了很多端点。这个Stripe MCP服务器能够让代理访问他们的两个账户。哇,这真的很强大。
You can create invoices, payment links. And with Stripe, there's a lot of different. If you want to create an invoice, you need a customer object. You need a subscription object. So there's a lot of chaining that you have to do. And I remember them telling me like, don't start through via the API, like just grabbing your open API back and generating an MCP server. No, start from the use case. What are people actually doing for product? And then you build like, you don't need full coverage of your API, just the most important things.
你可以创建发票和付款链接。而使用Stripe有很多不同之处。如果你想创建一张发票,你需要一个客户对象和一个订阅对象。这就意味着你需要进行很多关联操作。我记得他们告诉我,不要直接通过API开始,比如抓取你的开放API并生成一个MCP服务器。相反,要从使用场景出发。人们实际上在产品中做些什么?然后你开始构建,不需要完全覆盖你的API,只需要最重要的部分就好了。
I think about that for like as a traditional software engineer that's been doing this for the past 15 years. Now trying to convert to a AI engineer in a way, right? I have to be building with those tools and I have to start from the use case. So what is something I can optimize or make my life a little bit easier? Like the app you built for yourself, you know, that's fascinating. Like it's a real pain or like you're really curious about something.
我在思考这个问题,就像一个有着15年经验的传统软件工程师,现在想转型为人工智能工程师,对吧?我必须开始使用那些工具,并且从实际应用需求出发。比如,是否有某些事情我可以进行优化或者让生活更轻松些?就像你为自己开发的那个应用程序一样,真的很让人着迷。它解决了一个实际问题,或者满足了你的某种好奇心。
And that's always the best when you're really curious about something. I guess that led me to the other question, which way I love asking founders. I know you probably don't have time to work on site projects. But if you did have time, what are the site projects you want to do? Wow. I actually do have a lot of site projects. I love that. I have a theme called Dracula that I build. And I just like solving problems that I'm facing every day. So with that theme was like, you know what? I hated the fact that I had a theme on my code editor and a different theme on my browser and a different female swear. I wanted to reduce my cognitive load when I moved from different tools. So let me just build one that works everywhere. The same for recent, because just solving my own pain. So yeah, today I just have a lot of different pains.
这总是最好的,当你对某件事情真的很感兴趣时。我想这促使我提出另一个问题,这是我喜欢问创业者的问题。我知道你可能没有时间做副业项目。但如果你有时间,你想做哪些副业项目?哇,实际上我确实有很多副业项目。我非常喜欢这一点。我创建了一个叫Dracula的主题。我喜欢解决我每天面临的问题。比如说,针对这个主题,我讨厌在代码编辑器上有一个主题,而在浏览器上是另一个主题,还有不同的风格切换。我希望当我在不同工具间切换时能减轻我的脑力负担。所以我就想着创造一个能在所有地方使用的统一主题。最近也是这样,因为我只是在解决我自己的痛苦。所以是的,现在我有很多不同的“痛点”。
For example, I don't like dealing with personal life things like going to the DMV and renewing my, you know, like car registration or whatever or dealing with insurance. I would love to automate all of those things. I can't wait for someone to build an agent for DMV. That'll be amazing. I will be a customer like day one if you're watching this. Please But the Dracula theme is so cool. I remember when I first met you, I couldn't believe you were the person who made the Dracula theme. Like when you open up the item to like cover theme and it's like one of the top themes you can add to. I know we probably will take the entire episode to talk about the theme too because it has that incredible story. But like if you want to briefly talk about how the theme came to be and yeah.
例如,我不喜欢处理个人生活中的琐事,比如去车辆管理局(DMV)更新我的汽车登记或者处理保险之类的事情。我希望能自动化这些事情。如果有人能开发一个处理DMV事务的代理,那就太好了。如果你在看这个视频,我会在第一时间成为你的客户。请一定考虑这个想法。
说到Dracula主题,它真是太酷了。我记得第一次见到你时,我简直不敢相信你就是那个制作了Dracula主题的人。当你打开界面去选择主题时,它是其中一个最受欢迎的主题。我知道可能会花整集的时间来谈论这个主题,因为它背后有一个令人难以置信的故事。但如果你愿意,可以简要谈谈这个主题是如何诞生的。
Yeah, it's an insane story. Yeah, Dracula has now 9 million users and it's a side project, right? Recent is my main project. This is only a side project. And it started because of stravelling. I was in Germany. I was in Germany and I was traveling to Spain and I got sick. In the plane, I was like feeling so sick. I was like, oh my gosh, what's happening. I was alone. I've never been to Spain before. So I'm like, okay, I was so I was feeling so much pain to the point that I had to call the flight attendant and I was like, hey, I need help. So they landed in Madrid. They took me out of the plane in an ambulance. And I was like, okay, this is bad, right? They take me to the hospital. And they started to give me like some medicine. I'm feeling great. I'm like, okay, I'm ready to leave this place. Thank you for the help, you know, but I'm ready to go to the hotel or something.
是的,这个故事真是疯狂。是的,Dracula 现在有 900 万用户,而这只是一个副业项目,对吧?Recent 才是我的主项目,而这只是个副业项目。这个项目的开始和我的旅行经历有关。那时候我在德国,然后要去西班牙,结果生病了。在飞机上我感觉非常不舒服,心想天哪,这是发生了什么。我一个人,从来没去过西班牙,所以我心里很慌张,我疼到必须叫空乘人员过来帮我。他们在马德里降落后,用救护车把我送下了飞机。当时我心想,事情很严重,对吧?他们把我送到医院,开始给我用药。我感觉好多了,想着好了,我可以离开了,非常感谢他们的帮助,但我准备去酒店休息一下。
And they're like, no, no, you're not leaving. You got to stay. Turns out I stayed there for three weeks. It was like that bad. But in the first few days, like I was already feeling a little bit better. So I asked my co-workers in my dread. I was like, hey, can you bring me my computer? Because it got stuck like in the airplane or something. Like, you know, like I just laughed in an ambulance. So like my whole luggage was there. So they bring my computer and I'm like coding in the hospital. I'm like, super happy. I have my computer there. And then one day I leave the room just to get some water and then someone comes in and steals my computer. No, in the hospital. So I came to my room. I was like, what's going on? Maybe they really want to program to you. No, it was so bad. I felt it was like the worst day.
他们就说,不不不,你不能走,你得留下来。结果我在那里待了三个星期,那真是不太好。不过在最初的几天里,我已经感觉好一点了。所以,我就跟同事开玩笑说,嘿,能不能给我带一下我的电脑?因为它好像被留在飞机上或者别的地方了。当时我直接被送上救护车,所以行李都还在那儿。他们把电脑带来了,我就在医院里开始编程,超级开心能有电脑在身边。有一天我只是出去接水,然后有人进房间偷了我的电脑,居然是在医院里。所以我回到房间的时候,就觉得发生了什么事?也许他们真的很想学编程吧。不,那天真的是太糟糕了。
And then, you know, the next day I called my co-workers again, I tell them what happened. And they're like, don't worry about it. We're bringing you a new computer. So they bring me a new computer. And as a developer, you do that thing with a new machine. You're like, start to configure all your hotkeys and shortcuts and themes. So that's what Dracula came about. Like, I just wanted a thing that worked everywhere. So I built the first version in the hospital. And then it just took off after that. Were there already other themes when you built Dracula? What's the most popular theme before Dracula? There was Monokai. There was another one that's like creamy collar. I forgot the name now.
然后,你知道,第二天我又打电话给我的同事,告诉他们发生了什么事情。他们说,别担心,我们会给你带来一台新电脑。所以他们给我带来了一台新电脑。作为一名开发者,你会在新机器上做一些事情,比如配置你的所有快捷键、快捷方式和主题。这就是Dracula主题的起源。我只是想要一个可以在任何地方使用的东西。所以我在医院里创建了第一个版本。然后它就流行起来了。你创建Dracula的时候已经有其他主题了吗?在Dracula之前最流行的主题是什么?有Monokai,还有一个是奶油色的,我现在忘记名字了。
Why did you name the theme Dracula? I don't know. Like, I completely blanked on that. I don't know. I was like, is it related to Germany? I don't remember if Dracula was. I think just because it was a dark theme. But it really helped me on my entrepreneurial journey because at some point, I built a pro version and then I sold like $300,000 with a theme. It's just like six colors. So I would never imagine that you could sell colors online. And people would buy. But that gave me the confidence then be like, you know what? Let me build my own company. If this works, maybe something else will. I love that because there are people who are selling email templates too. And now, like, what you're building and basically empowering a new type of an audience to be able to build their own Dracula and not for coming online before emails. And they can go sell it if it turns out to be very good looking and people want to adopt it because emails are hard to implement.
你为什么把这个主题命名为德古拉?我也不知道。就像,我完全空白了。我不知道。我当时想,是不是和德国有关?我不记得德古拉是否有关系。我想可能只是因为它是个暗色主题。但这个主题确实在我的创业历程中帮了我很多,因为我开发了一个高级版本,然后卖了大约30万美元。它其实只是六种颜色组合。我从没想到颜色也能在网上卖,而且有人会买。这让我有信心去创办自己的公司。我想,如果这个能成功,其他东西也可能成功。我很喜欢这种感觉,因为有些人也在卖电子邮件模板。而现在,像你这样的人正在创造新的东西,基本上是在赋予新用户创建自己的德古拉主题的能力。他们可以在线上出售这些模板,如果这些模板非常好看,并且有人愿意采用,因为电子邮件的实现是很难的。
If it converts, have you seen actually users of a recent selling like, react to email templates? Not yet. But there's a lot of libraries that are being built around that. Yeah, that's amazing. Okay. So I guess last question, if you have a crystal ball, and you can predict the future, what do you think the future of email and future of AI will look like?
如果它转化了,你有没有见过最近销售的实际用户对电子邮件模板的反应?还没有。但是有很多围绕这个主题建立的库。太棒了。那么,我想问最后一个问题,如果你有一个水晶球,可以预测未来,你认为电子邮件和人工智能的未来会是什么样子?
I believe today, there's a lot of actions that we take as humans. And that's the majority of the work when you go to, you know, all these different apps, you go to super based, the most databases you see there were built by humans. You go to recent, the emails were sent by humans or drafted by humans and then sent programmatically. I think we're going to see a big shift in terms of like, who is the actor? Who is the creator? And I believe it's going to be the majority of the actions will be taken by agents instead of humans. And that's just the reality we're going to live in.
我相信,目前有许多行为是人类执行的。当你使用各种应用程序,比如访问超级数据库时,你会发现大多数数据库是由人类创建的。打开邮箱时,邮件也是由人类撰写后通过程序发送的。不过,我认为我们将看到一个巨大的变化,那就是行为的执行者和内容的创作者将主要是智能代理而不是人类。我相信在未来,绝大多数的操作都会由这些代理完成,而这将成为我们生活中的现实。
So we have to rethink the way we're building product to support that reality. Do you want to have more on how to rethink how to build product?
因此,我们必须重新考虑我们构建产品的方式,以适应这种现实。你想更多地了解如何重新思考产品构建的方法吗?
Yeah. Because it's such a great point. Yeah. And to rethink these products, you really have to, you know, go down the whole journey of the user. It starts from the own morning, you cannot have like 10 steps or you cannot wait two days to get access to your account. You cannot, you know, wait for the account manager to schedule a call with you or to book a demo. From that point, you need to experience the aha moment as soon as possible and the agents should be able to take action as soon as possible.
是的。这真是个很好的观点。要重新设计这些产品,你必须真正去理解用户的整个体验过程。这个过程从用户早上的第一步开始,你不能让用户经历十个步骤或者等上两天才能访问他们的账户。你不能让用户等账户经理来安排电话或预约演示。从这个角度来说,用户需要尽快体验到“恍然大悟”的时刻,而客户服务人员也应该能够尽快采取行动。
And you have to rethink role-based access control. You have to rethink permissioning and authentication for these agents. So every single layer of the product, you have to rethink. Like, even the way you store the activities, you know, like every SaaS has like this activity space there. They can record everything you did as a human.
你需要重新思考基于角色的访问控制。你还要重新考虑这些代理的权限管理和认证问题。所以,你需要重新审视产品的每一个层面。比如说,甚至存储活动的方式也需要重新思考。几乎每个SaaS产品都有一个活动空间,可以记录你作为人类所做的所有事情。
Yeah. Like actually recording what every agent is doing is more important than ever. Because if they screw up, which they will in the beginning, then you need to know what happened. If they take a destructive action, you need to know. So there's just a lot that will change in terms of how we build products. And we have to take that in consideration, because humans are not going to be the only user anymore.
是啊,现在记录每个代理的行为比以往任何时候都重要。因为他们在一开始可能会出错,你需要了解发生了什么。如果他们采取了破坏性的动作,你也需要知道。所以在产品开发上会有很多变化。我们必须考虑到这一点,因为人类将不再是唯一的用户。
That's powerful and amazing. Well, thank you so much for coming. This is a lot of fun. Yeah, thank you.
那太厉害了,真让人惊叹。非常感谢你的到来。这次真的很有趣。是啊,谢谢你。