为 Agent 设计——Zeno Rocha,Resend
- Resend 的切入口,是在一个成熟的大型邮件市场里补上对开发者的忽视。Zeno Rocha 称,Resend 已有 430 万用户和 12 万付费客户,客户包括 Warner Bros. 和 HBO。其 SDK 的 npm 下载量如今超过 SendGrid 的 2 倍,但 SendGrid 发送的邮件仍多得多:采用势头和基础设施规模是不同的指标。
- Agent 集成正同时成为分发渠道和采购标准。Resend 的 MCP 月调用量大致翻倍,到 8 月底约达 120 万次;CLI 的增长较慢。Rocha 希望产品“没有固定形态”(“formless”),并表示自己从 Plausible 转向 PostHog,是因为想要由厂商官方维护的 MCP。
- 邮件送达率取决于声誉管理和滥用控制,而非单纯的发送能力。向 10 万名收件人发送营销邮件,需要队列、分批发送、按邮件服务商限流,并谨慎安排受众顺序。Resend 不在前期设置审批门槛,但一旦发送信号恶化便会迅速介入:“大多数时候只是误用,不是滥用。”
- Resend 正把开发者体验优势延伸到基础设施所有权。Resend 最初构建在 Amazon SES 之上,如今正加深对发送基础设施的掌控,以管理 IP 池、吞吐量、性能和可观测性。Rocha 称,团队已投入数月,近期将有公告;发送链路也已从 JavaScript 转向 Go。
- 机器可读的产品信息可能提升 Agent 的判断准确度,但能否带来更多分发仍难以确定。主持人讲述了一次自己网站上的测试:在提供 JavaScript 小组件之外的可读价格信息前,模型对 120 万封邮件的场景严重算错了价格。主持人还称,自己在 Opus 4.5 上观察到近期偏好;两位发言者则质疑单项 AEO 手段是否真有作用。
- 拓展企业客户和营销邮件业务,分别暴露出不同的产品与定价要求。Resend 目前只有 1 名客户经理;Rocha 认为企业需求远不止 SSO、SCIM 和审计日志。与此同时,Mistral 等快速增长的客户让按联系人数量计费的营销定价受到挑战;Resend 向自家 430 万用户发送一轮营销邮件就要超过 1 天。
- Rocha 把设计和家庭优先事项明确纳入增长约束。一位 YC 合伙人曾告诉他,对设计过度执着拖慢了进度;他接受这项批评,但仍保留了这种差异化。假设要选,他宁愿经营一家 ARR 为 1,000 万美元、女儿愿意亲近自己的公司,也不愿经营一家 ARR 为 1 亿美元、却失去这种亲密关系的公司——这不是对当前收入的披露。
1. 开发者长期被忽视,为后来者打开了市场缺口
Rocha 用过 Mailgun、Mailchimp 和 SendGrid 等工具,但觉得这个市场对开发者关心不够。主持人披露自己是 Resend 的天使投资人,并以 Mailchimp 和 Postmark 已退出、SendGrid 已归 Twilio 旗下为例,说明市场机会。主持人觉得 Rocha 的切入点格外直接:他从前端开发转行,没有邮件基础设施经验,只是想做“邮件领域的 Stripe”。
Rocha 称,如今 Resend 为 430 万用户提供服务,付费客户已超过 12 万。客户既有独立开发者,也有 Warner Bros. 和 HBO。Warner Bros. 是他偶然发现的:某个周六,他浏览客户域名时点进 wbd.com,才意识到 Warner Bros. Discovery 已经在用 Resend。
主持人追问,为什么老牌厂商的规模优势不能胜出。Rocha 认为 Twilio 忽视了自己的开发者基因:“他们在打盹。”今年早些时候,Resend 的 SDK npm 下载量超过了 SendGrid,如今已是其 2 倍以上;但 Rocha 承认,这个指标有局限,SendGrid 仍凭借大型企业客户发送多得多的邮件。
2. 送达率取决于克制发送,而非蛮力推量
同时向 Gmail 发送 10 万封邮件,很容易触发限流。Rocha 介绍,Resend 会先把互动最活跃的收件人排在队列前面,再根据邮箱服务商观察到的互动情况,调整批次和发送间隔。Gmail、Outlook 和 Yahoo 的行为各不相同;他估计 Gmail 占 Resend 收件目的地的“至少”60%。
Rocha 称,服务商希望发送量稳定上升,而不是先大量群发、沉寂一段时间后又突然冲高。这类突发流量看起来像滥用。因此,新客户会先受到保护性限流,发送速度再随历史记录逐渐调整:“你得做出滥用者绝不会做的事。”
Rocha 举例说,有一个日本团伙购买了“可能有 2 万个域名”,涉及不同顶级域和注册商。Resend 会检查域名注册时间、网站响应和相似域名。新买仅 2 天的域名,声誉起点就较低;另设一个与主域名相似的外联域名,本身也可能像钓鱼网站。
行业过去更看重 IP 声誉,如今声誉管理已更多转向域名层面。Rocha 建议用子域名区分事务邮件和营销邮件;Resend 通过 confidential.resend.com 发送投资人更新。主持人提出购买匹配的 .com 域名是否就能解决送达率问题,Rocha 则提醒,相似域名可能反而引发更多怀疑。
3. 低门槛接入把反滥用责任转移给平台
竞争对手大约 2 天的人工审批流程被取消后,Resend 选择“默认信任用户”,但一旦退信率、投诉率或邮箱服务商反馈显示异常,就会迅速封禁。“大多数时候只是误用,不是滥用。”Mailgun 创始人曾提醒 Rocha,研发工作有 90% 都花在对付滥用者上;Rocha 希望用自动化避免反滥用成为 Resend 的核心业务。
Rocha 称,Resend 会自动从 HTML 提取内容并附加纯文本版本,送达率因此“提升了很多”。主持人反驳说,纯文本无法放入打开追踪像素。Rocha 回应,Apple 的隐私改动和自动化邮件检查,可能在真人阅读前就记录活动,因此打开率本来就越来越不可靠。
点击追踪也有声誉风险。把链接改写成共享追踪域名,会带来“邻居效应”:一个钓鱼客户就可能让该域名被列入黑名单,持续数天乃至数周。Rocha 承认,Resend 起初也这么做过。让链接和图片域名与发件人域名一致,有助于避开滥用行为常见的特征。
4. 把重复问题变成产品功能,客服才能规模化
Rocha 说,Resend 的客服体系大约发展到“重大版本 4”。只雇人坐班没有奏效;光有领域经验,也无法避免重复工单。有一条有效做法是要求员工只能附上文档链接来回答,倒逼团队先完善文档。一个问题最初可能要花 1 个半小时回答,但由此写成的文章之后就能反复复用。
Resend 的用户从 12 月的大约 100 万增长到约 400 万,按比例扩充客服团队并不划算。Lovable、Bolt 和 v0 带来不少无法独立排查问题的用户。问题大多集中在 DNS:用户不知道 CNAME、A 或 TXT 记录该填在哪里。Rocha 希望保留免费层,因为免费的开发者产品曾帮助他开启自己的职业生涯。
Domain Connect 提供了“面向域名的 OAuth”,以授权流程取代手动填写记录。Rocha 称,Resend 说服 Vercel 参与其中,Cloudflare 等厂商也已支持。推广仍很费力:与 IONOS 接洽耗时数月,GoDaddy 收取高额费用,协议维护者也不多。主持人直言它的吸引力:“你什么都不用输入。”
5. 开源和产品打磨拓宽开发者入口
Rocha 称,Resend 开发期间的用户研究发现,设计师文件转邮件模板是个痛点。HTML 表格、内联 CSS 和 Outlook 专用注释让工作变得很折磨。React Email 用 JSX 作为模板语言;Rocha 确认,它并不会在邮件里加入可交互的 React 应用:“你不能在邮件里发送 JavaScript。”关于时间顺序,转录内容有出入:主持人说 React Email 比 Resend 早一年推出,Rocha 则说自己开始做 Resend 后,才听到大家对模板的抱怨。
Rocha 把 React Email 称作“我们的 Next.js”,指的是它作为开源获客渠道的作用。他亲自写了适配 SendGrid、Mailchimp 和 Mailgun 的集成指南,让这个项目不只对 Resend 用户有用。策略是先解决模板制作问题,再提供邮件发送平台:“先不断给予,最后再提出请求。”
产品打磨也体现在 SDK 上。Rocha 称,团队不喜欢 Stainless 和 Fern 等生成器的输出,因此选择由人来编写 SDK。Java、Rust 和 .NET 用户也应该能在自己使用的语言里获得良好体验,即便数百名用户只是数百万开发者中的一小部分。他想到的是自己在巴西长大时,手里只有一台配置差的 Windows 电脑,网络也不稳定。
模板也要适应不同的协作习惯:有些团队希望模板纳入源码管理;另一些团队则希望营销人员用可视化方式编辑,工程师只需调用模板 ID 并传入变量。Resend 的编辑器支持协作和冲突解决。公司内部大约有 20 人用它共同编写全员邮件,最后再统一发送。
6. 从 SES 向下延伸,让 Resend 更掌控发送链路
主持人提到一个低成本基准案例:freeCodeCamp 通过 SES 向 500 万名新闻邮件订阅者发送邮件,每月花费约 5,000 美元。Rocha 认为 SES 是不错的平台,Resend 起步时也使用 SES,最初把精力放在开发者服务不足的环节,比如多语言 SDK,以及从 Lambda 发送邮件等实用指南。
隐性成本来自交付链路周边的各种服务:反馈事件需要 SNS,日志需要 CloudWatch,大流量需要 SQS 等队列,Webhook 也要额外配置。“你的基础设施足迹很快就会比你想象的大得多。”SES 的域名结构和配置集模式,在 Resend 发展过程中也逐渐显出限制。
Resend 如今正构建更深入的基础设施,“一路下探到硬件层”,这项工作已持续数月。具体包括购买 IP 段、管理 IP 池,以及控制发送到各家服务商的吞吐量。服务商关系也很重要:Resend 招来了来自 Postmark 和 Mailchimp 的送达率资深人才,补上了 Rocha 刚创业时尚未建立的行业人脉。
提升可观测性是核心目标之一:开发者希望了解邮件离开服务后,到收件人真正看到它之间发生了什么。性能同样重要。Resend 过去是 TypeScript 和 Node.js 团队,如今发送链路已改用 Go,并开始挑战人们对 P99 操作耗时的固有预期。
7. 官方 MCP 正成为采购决策的一部分
Rocha 的产品理念是“我希望产品没有固定形态”。API 优先的架构让客户能自行选择工作流,包括 Claude Desktop 和 Codex。对一位注重设计的创始人来说,这有些讽刺:客户越来越常告诉他,他们从不打开他花了大量时间打磨的仪表盘。
主持人称,MCP 使用量的月增长率从约 50% 加速到每月大致翻倍。Rocha 表示,到 8 月底,调用量约达 120 万次。最初的实现来自投资方 a16z 的 Yoko:她构建了一个本地开源服务器,之后 Resend 将其采用。Rocha 说 CLI 使用量也在增长,只是速度较慢。
为了做好远程 MCP,Resend 必须重新审视权限模型、构建 OAuth,并梳理 RBAC 和资源访问。之后才能通过官方商店和市场分发。增长并非只是多接入一个连接器:发布集成本身也暴露出底层产品和授权体系需要补齐的工作。
Rocha 提到自己从 Plausible 转向 PostHog,是因为想要官方 MCP。尽管他信任 Sentry,但由 Sentry 维护的 Plausible 集成仍无法满足需求。“我需要软件的所有者也亲自负责这部分集成。”本地集成也让他不满,因为无法在桌面和手机之间通用。
8. 完整开放给 Agent 使用,会暴露产品设计的长尾问题
Resend 的 MCP 工具数量从大约 20 个以内增加到“八十几个”。主持人拿 Cloudflare 只有 2 个工具作对照,问 Rocha 什么情况会让他改变做法。Rocha 会参考客户反馈,也会亲自使用产品,包括制作投资人更新。他眼下更担心响应内容过大、耗尽上下文;相比之下,工具数量过多已不再那么让他困扰。
把仪表盘的 90% 覆盖率补到 100% 很难:账单、用量、邀请、配置和删除功能都需要逐一决策。主持人认为,把数据直接绑定到页面的框架会妨碍开放 API。Rocha 回忆,Next.js 托管成本高昂,团队后来迁移到传统 Node 服务器,还得重复开发公开和私有 API;如今 monorepo 让代码复用容易多了。
Resend 的招聘页面支持通过 CLI 和 API 提交申请,候选人也可借助 skills 让 Claude 代为申请。主持人质疑这会不会招来“低质量内容”,并问创始人是否应该把招聘软件交给外部服务商。Rocha 认为,申请过程中的设置工作本身就是筛选信号;他对 CLI 申请者采用不同的筛选方式,因为认证和安装过程能证明对方确实体验过产品。
Resend 即将推出的另一款产品,会从接收邮件扩展到让 Agent 以编程方式创建几十个、甚至可能 100 个收件箱。接收邮件只是基础抽象;下一层还需要提供读取、标记未读、加星标和加标签等 API。Rocha 称这是 API 驱动的产品,并向主持人提供了一个功能标记后的访问权限。
9. 可读页面能纠正对产品理解有误的 Agent
Resend 起初很难与在 SEO 上投入了十多年、积累深厚的老牌厂商竞争。它的内容基础是范围窄、能解决实际问题的文档:不只讲 Python,还讲如何用 Django 和 Python 集成。随着问答式界面兴起,Rocha 开始在营销页面加入常见问题,让“Resend 是否符合 SOC 2 合规要求?”这类问题可以直接得到答案。
Resend 的营销页面提供 llms.txt,文档也有 Markdown 版本。Rocha 计划为所有营销页面补上 Markdown 版本。用户可在页面地址后加 .md 获取内容,也可以采用内容协商。主持人另举一例:自己定价页面的 Markdown 版本获得了普通浏览器分析无法看到的访问;这些访问只能从 Datadog 的服务端数据中发现。
主持人讲述了自己网站上的一次定价测试:让模型比较发送 120 万封邮件的成本,模型严重算错了 Resend 的价格,还称其定价难以理解。主持人说,Agent 没有操作 JavaScript 小组件。几天后,他加上 Markdown 版本再测,模型就给出了正确数字。Rocha 指出其中的取舍:Markdown 内容和网页内容都要同步更新。
Resend 的迁移页面提供代码转换器,而非传统的竞品对比;Rocha 如今觉得,竞品那些明确的对比页面带来了更多推荐。他也会把图表数据用文字写出来。主持人则指出,能够操作界面和查看图片的 Agent 越来越多;Rocha 区分了这类流程与不运行小组件、只抓取信息的工作流。
10. AEO 已影响商业结果,但因果机制尚未厘清
主持人称,自己“至少在 Opus 4.5 上”观察到了近期偏好:即使传统 SEO 较弱,Resend 这样的新工具似乎也能受益。Rocha 认为,网络各处的自然讨论都会影响推荐。他的建议是尽可能让更多人谈论产品,而不是照着某种所谓公式行事。
主持人说,团队试用了多个 AEO 服务,现在有 4 个订阅,但很少能从中得到具体行动建议;这些工具大多只汇报现状,厂商自己也还在摸索该怎么做。主持人还介绍了一个实验闭环:把 Agent 接入分析数据、代码库、Ahrefs 和 SEO 数据,并要求人在实验开始前签字确认。
降低注册门槛是另一个尚无定论的方向。主持人提出,能否先让用户免注册使用基础功能,再配置并认领资源。Rocha 回应,邮件领域的滥用问题很严重;两人没有就这种模式或配置 API 是否值得做达成结论。
11. 拓展企业客户,考验产品、定价和定位
Resend 的销售体系刚起步,目前只有 1 名客户经理。Rocha 有意循序渐进地引入销售,以保护公司文化。他知道企业销售需要销售赋能和组织机制,但表示首位客户经理正在发挥作用,销售进展也不错。
企业客户需要的“实际上是另一款产品”,而不只是另一个销售渠道。他们关心的问题与每月付 20 美元的用户不同;在定价页面加上 WorkOS、SSO、SCIM 和审计日志并不足够。主持人举的少数几笔大单就能让 ARR 翻倍,是说明销售杠杆的一般例子,并非 Resend 已取得的业绩。
营销邮件通常按联系人数量收费,但 Mistral 等高速增长的 AI 客户让 Rocha 开始质疑这套模式。发送侧也需要动态调整批次,而不是用简单的事务邮件队列处理。向 Resend 自己的 430 万用户发送一轮营销邮件要超过 1 天;主持人认为这不可接受,两人随后讨论是否可以付更多费用,换取不同的队列处理方式。
主持人区分了品类主导地位和品类创造:“AI 原生的 SendGrid”还不足以证明一个新品类已经诞生。主持人心目中的标杆是 HubSpot 的 inbound marketing——这个说法不需要宣言,就能让人觉得其他选择不够理想。Rocha 承认这很难;讨论结束时,Resend 下一句能定义品类的话仍未找到。
12. 设计和家庭优先事项,构成 Rocha 对成功的定义
Rocha 带着 1 岁的女儿进入 YC 2023 年冬季班,当时的目标是做出大约 250 家公司中最好的那一家。起初,他觉得自己和那些来自名校、尚未结婚的创始人竞争处于劣势。后来他意识到“YC 没有赢家”:融资排名会淡去,联合创始人会分开,公司会转型,而这场竞争还会继续。
Paul Graham 曾建议创始人认真听取伴侣的意见。Rocha 便问一位 YC 合伙人,自己应该停止做什么。对方说他太在意设计,否则现在会走得更远。“这句话让我非常受伤”,部分原因是他觉得对方说得没错。但他仍宁愿让 1,000 名用户使用自己喜爱的产品,也不愿让 10,000 人使用一款自己并不想用的产品。
Rocha 正准备迎接第二个孩子,也在增加回巴西的时间。他设想 10 年后,宁愿经营一家 ARR 为 1,000 万美元、女儿愿意和自己玩的公司,也不愿经营一家 ARR 为 1 亿美元、却与女儿关系疏远的公司。“可能对,也可能不对。我不知道。”这些都是假设情景,不是当前的财务数据。
主持人追问,如何兼顾雄心和陪伴:即使人在家里,也可能一直看手机、想着工作。Rocha 没有给出简单的解法。他在家里切换角色时,也会觉得自己不是个称职的父亲、丈夫或 CEO。“我每天都努力让自己多投入一点。”他同时强调,自己的雄心没有减弱。
完整逐字稿
本集逐字稿暂无译文,当前显示原文。
It’s like, “Oh my gosh, how are we ever gonna beat SendGrid? It’s impossible.” Earlier this year, we passed SendGrid in downloads, and I was like, “That’s insane.” Now we’re more than double. Resend is powering 4.3 million users as of this morning. We have all sorts of customers, like Warner Bros. and HBO. There was a Japanese gang that recently came. They brought maybe 20,000 domains at Resend. Those are 20,000 different TLDs registered with different domain registrars, and we’re a $10 million ARR company.
I’d rather live in that reality than live in a reality where we’re a $100 million company and my daughter doesn’t wanna play with me, isn’t interested in me, and we’re not that close.
We’re building the infrastructure and going down to the metal now. It’s a big project. We’ve been working on it for months, and we’re gonna announce something soon about that.
Okay, we’re in the studio with Zeno Rocha from Resend. You and I go back to your VS Code theme days, when you were just an indie hacker.
Yeah.
And I was like, “Man, I don’t know what this guy does, but everything he does has taste.” And then now you’ve built an email infrastructure platform. Anyway, welcome. You undersell being an indie hacker.
Top 20 most active GitHub users under 20 in 2022.
Nice. Right. Very, very active open-source stuff.
Yeah, yeah.
What’s your typical introduction these days?
I think I just focus on my recent career. For the longest time, even today, when I say, “Oh, I created Dracula,” people know Dracula more than Resend.
Still.
Yeah, but it’s been nice. It’s changing.
What was the impetus for picking email?
People always bring up React Email and all those things.
Yeah.
I just don’t know if I would have made it a company.
Yeah.
If I knew how hard email was, I probably wouldn’t have started the company either.
Were you primarily a JavaScript front-end—
Totally.
—CSS dev? No infrastructure, right? No email background. The HTML5 guy?
1. Building the Stripe of Email
I was a front-end guy. What drew me to it was looking at all the tools around it. I’d used all of them before: Mailgun, Mailchimp, SendGrid. I always felt like there should be something where someone cares about this space. There are all these players, and it’s a massive market with a lot of competitors, but no one cares about the developer persona. Maybe they care about the marketer or the product manager, but not the developer.
I started building around the premise of, “If this was the Stripe of email, how would that look?” I had no idea or knowledge of email. I was a user of email, but I’d never built an email pipeline or any of that infrastructure before.
Yeah, I mean, typically, if you ask me, Postmark was still a thing.
Yeah.
And then obviously SendGrid and all these folks. They’re still around, but maybe they’re not as innovative as a new YC startup trying to make it.
Yeah.
Just upfront, I like to let people brag a bit. For people who’ve never heard of Resend, what are the ways you talk about it? The most impressive numbers, the logos, anything?
Resend is powering 4.3 million users as of this morning. We have all sorts of customers, like Warner Bros. and HBO—customers that are really big and have all sorts of demands—but also a lot of indie hackers and folks who are just building a website or their own application. We just crossed 120,000 paid customers. And, yeah, the endless logo wall.
Oh my God. Yeah, the big logo wall. Do you think this is effective? You wanted the flex. There you go.
You know what? In the beginning, I felt like I just needed some sort of social proof. In the end, I was like, “You know what? This is actually pretty cool.” I honestly found companies like Warner Bros. just on a random Saturday. This was before AI, so I’d go to my database and think, “Oh, let me see who I can put on the website.” I did the most naive thing you can do: I looked at the domains, filtered by .coms, and filtered by the small ones. There was this domain, WBD—Warner Bros. Discovery. I didn’t know what WBD.com was. I was like, “Oh no, Warner Bros. is using us.” I just love featuring customers.
Pretty good. Specifically for listeners, one of the reasons I reached out for this interview: one, I should declare upfront, I’m an angel. When you sent me the memo, it was super simple. You were like, “Look, Mailchimp has exited. Postmark has exited. SendGrid is under Twilio. Who is doing this?” And I was like, “Okay, yeah, seems right.” Honestly, in terms of the pitches I get, I’ve invested in about 30-plus companies, and that was one of the most straightforward things ever.
And then, secondly, I don’t like to add API keys. I like to have everything on one platform, one API key, copy everything. Cloudflare has built-in email, and I had a project. I set up my Cloudflare email domain DNS and all that junk—DKIM, what have you. I don’t know all the alphabet soup of stuff. I already had this set up, and then I wanted to add another sort of email thing. The Claude model said, “You must add Resend.” I was like, “Who the fuck is telling me this? Resend? I already have Cloudflare email set up, and it’s still recommending it.” It’s on the screen now. This is him praising Resend, like, “I don’t need Resend, but you want me to keep adding Resend.” So who did you bribe at OpenAI and Anthropic to get Resend on there?
I wish there was someone to bribe, because I’d definitely be paying them. But I personally spend a lot of time on that topic. That’s something I feel like every generation has: tools that define that moment. We remember the Heroku days, Netlify, Vercel. To me, it was like, yeah, there’s room for a new generation, and this new generation isn’t going to be distributed in the same places it was before. So from the beginning, I was very focused on, “Okay, we’ve got to be number one when someone asks about that.”
You’re very number one. Claude Code recommends you 60% of the time or more. I think it’ll be fun to see the playbook. How do you win that?
I do want to get back to what you said about having no idea how to start with email. I think a lot of people don’t understand what goes into making a service like this. You might, because you’ve set it up. Some people would just say, “Okay, I can use Twilio, I can use SendGrid, I can use you guys.” But what goes into making a tool to mass-send emails? What do people not think about?
2. How Email Delivery Works
There are a lot of different things you can do, because landing in the inbox, especially when you’re sending high volume, is different from scaling an app where you throw GPUs at the problem or scale vertically or horizontally. With email, you have to respect the rules of the game. Gmail behaves differently from Outlook, and differently from Yahoo Mail. It’s a fascinating, old-school industry that relies a lot on contacts.
What we found early on was that we knew nothing about this world. Coming from an extremely naive perspective was the best thing that ever happened, honestly, because there was no preconception of what was right or wrong. For example, with every app, the first thing you see is a form you have to fill out: how many emails you’re going to send, what the volume is, what you’re trying to do. Then you go through a human verification process. Two days later, they accept or deny you. Most of the time, they deny you. I was like, “This doesn’t make any sense. Why are we blocking people?” So I decided to remove that friction completely, and the friction would be on us. We would figure it out.
Of course, now, three or four years into this business, I get why they do that. There’s so much fraud, phishing, and abuse across the board. I don’t talk too much publicly about that, but we build all sorts of crazy systems internally for it.
To your question about how you architect that: when you’re sending an email to 100,000 people, if you just throw 100,000 emails at Gmail, Gmail will throttle them because it’s way too much, and they want to see the engagement.
So what you do is, they see, “Okay, are people opening the emails? Are people archiving or deleting them right away, or not interacting?” Depending on those initial interactions, what I do with the rest of the emails is dictated by that beginning.
There are clever things you can do. For example, put the most engaged audience members at the beginning of the queue. You have to put that in a queue. It’s an art and a science: you have to space out each batch. You also have to space things out between the different providers. Now we’re buying different IP blocks. You have to look at the history of the IP blocks, distribute people into different IP pools, and then balance the IP pools over time. It’s a whole crazy infrastructure game.
How automated can that be, or is there still a lot of human judgment?
It’s a lot of automation, for sure. Over time, we learned how fast we can send for new customers. Of course, if you’re sending more, maybe you can send faster, but there’s a throttle we have to apply to protect you as a sender.
What Gmail and all these players want to see is a graph that’s consistently sending more and more, versus what most people do, which is send once to a big audience and then never send again. Then it’s empty, and then boom, you send it again.
Once a week? Twice a week?
Yeah. Those are patterns an abuser would have, right? The whole way of thinking about email marketing and email in general is that you need to do exactly what an abuser wouldn’t do. It’s the opposite of that.
What does an abuser do? They buy a domain. We had a Japanese gang come in recently. They bought maybe 20,000 domains, registered across different TLDs and domain registrars, and we look at every single factor. They come in with a domain that was bought two days ago. Very new domains get a bad reputation. Then you look at the website: is it returning 200? Basic things like that. Some people don’t even set that up, or they use a lookalike domain. They might think, “Maybe I’ll use another domain for cold outreach.” That’s terrible because it looks like you’re phishing. You’re using a domain that looks the same.
They do all these little things, and they send a lot of emails all at once and then never send again. You have to reverse-engineer how phishing works and build a platform that does exactly the opposite.
Do you have a list somewhere, or do you recommend people look at some well-known lists?
What I’ve learned is that email is the type of world that requires a lot of education, but people aren’t interested in learning. The platform should do all those things for them. We added insights inside the product, so people can just apply them. We look at sending patterns, and now with AI it’s pretty cool, too. We have link checkers, spell checkers, and all these checkers before you send.
It’s a mix of teaching your audience and having the product do the teaching. The more you can bake in the best practices, the better. For example, when you send an email, you can send an HTML version and a plain-text version. Very early on, we thought, “The whole world is changing. There are way more agents than before. Let’s automatically generate the plain-text version of each email.” We can extract the Markdown from the HTML and then send that.
We started sending that automatically, and it increased deliverability a ton. It’s so much easier for spam filters to look at the content directly without having to parse it. Those are the kinds of things where we think, “Let’s bake this into the product.” Most people don’t do that, but we think it makes sense.
But with plain text, there’s no tracking, right? There’s no tracking because you can’t embed the image.
That’s right.
So you don’t have open rates. How do you know you have deliverability if you don’t have open rates?
This is very interesting because there are two ways of tracking in the industry: open tracking and click tracking. Open tracking includes a one-by-one GIF, a transparent image. If that image is loaded, then you detect that as an open.
The problem is all the changes Apple made recently around transparency and data privacy. They read emails before they land if they’re doing summaries and those kinds of things. So the open rate as a metric isn’t as reliable as it used to be, because all these spam filters are checking emails beforehand.
Then you rely on click tracking. Click tracking can also be difficult because you want the URLs to match the sending domain. You also want the image URLs to match the sending domain. That’s another pattern an abuser will use: they host images somewhere else, point links somewhere else, and send from these domains.
I feel like I’ve used email services that rewrite links for tracking to their own domain. You’re saying that’s bad?
That’s bad. We did that, too, at first.
I thought that was normal.
What’s bad is that you can have a noisy-neighbor problem, too. If one of the customers using that click tracking is an abuser and sends a bunch of phishing, that domain might get listed on a blacklist. Then it takes maybe days or weeks to get off the blacklist, and you’re screwed as a business.
Isn’t that pretty common? You make it very easy to sign up for a free tier and send a bunch of emails. I’d assume you have a lot of spam and phishing. How are you able to avoid that? Is this why?
I think what was different for me—and this is more of a philosophy—is that I saw those other products and the way they behaved, and I felt like it was almost as if you were born into a society where you first go to prison, and then you have to spend 18 years in prison. You have to prove you’re a good citizen, then you get out of prison and can do the things you want to do.
I always felt like, no, I want to live in a society where I’m born free. If I make a mistake, I might be punished just a little bit, or if I do something really bad, I get a bad punishment. That’s how I feel about it, too. You come in, and we trust you by default. Of course, that’s a naive way of thinking, but we’re going to trust you by default instead of making the whole platform worse for everyone because of those bad actors.
What we do differently from others is that, at the very first moment we notice something is off—by looking at bounce rates, complaint rates, and all the feedback we get from mailbox providers—we block you very quickly and then chat with you. Most of the time, it’s just misuse. It’s not abuse. We learned that we need to differentiate.
Skill issue.
It’s a skill issue. Exactly. It’s a skill issue. I think we learned we have skill issues, too, that we wouldn’t have expected.
Yeah, we learned. Email is so important, and it’s so dirty and so secretive. I almost just want you to tell us everything we don’t know.
You have a fun page that you call your philosophy. I got lost just reading around it. And you have a pledge that you’ll make email safer.
Mm-hmm.
If you’re interested in this stuff, you have quite a bit up there. I want to ask about “How we scale support.” You remind me of another investment I had, Jake from Railway. They’re also very permissive and developer-oriented, but you have to scale support, right? The moment someone gets blocked, they come to you and say, “Hey, don’t keep me off your platform.” How do you scale support?
Man, if support were software, we’d probably be on major version four or something. The very first version was me.
You?
Me. I was so stressed out. I was like, “Dude, I just need someone sitting on the chair. It’s that simple. I cannot be doing this anymore.” And it was just anyone. Anyone, I don’t care.
Then you do that. Of course, it doesn’t go well because you realize very quickly that you need someone who will actually try to prevent the tickets from happening in the first place. So then we went to V2, which was a new person we ended up hiring. They were like, “Okay, we’re answering the tickets. They have domain expertise.” But they still weren’t doing the proactive work, like, “No, let’s make sure this is documented before we talk about it.”
One policy we set was pretty cool: you can only respond to a ticket with a link. You can’t just give the answer right away. You have to stop and write a blog post, write documentation, or write a knowledge-base article. So it takes maybe an hour and a half for you to answer a ticket.
But guess what? Now you’re sending that link all over again. So that was super cool, and it helped us build that knowledge base. But even so, when we crossed maybe 1 million users—we had 1 million users back in December, and now we have 4 million—I can’t throw people at the problem. That’s just not going to solve it, right? I’m just going to keep staffing, and that’s how it is. No. We really have to rethink everything from scratch again.
We want to keep that free tier. I don’t want to get rid of it. I started as an engineer because I was able to use those free products, right? If it weren’t for Heroku’s free tier and all these other products, we wouldn’t be here. So I want to keep Resend that way, but it took us many iterations to get there.
How much of this is that it’s a domain people don’t know a lot about, with unique questions that need human support, versus Command-K with docs and a model? Customer service bots exist, right?
Yeah. I think it’s much better now. There are a few things where there’s so much friction. DNS, oh my gosh.
It’s always DNS.
It’s always DNS, right? The majority of the questions we get today are from people not able to set that up. It’s fascinating because we’re serving a new audience—an audience that doesn’t even know how to troubleshoot. I remember last year, when Lovable, Bolt, and v0 came along, I could tell these were different users. When they have a problem, they get stuck. They don’t know how to get past it.
With DNS, they simply don’t know what a CNAME, an A record, or a TXT record is, or where to put those things. They would paste the wrong value or silly things. By the way, right now I use computer use, like Codex or Claude Code, and it completely manages my DNS records. It’s great. It debugs it, runs dig, and tries all the DNS stuff. I don’t have to worry about it.
One thing we did was find this super obscure protocol that very few people in the industry were investing in. It’s called Domain Connect, and it’s super interesting. It started with some domain providers.
Oh.
Yeah, this is like OAuth for domains, right?
Exactly that. Is this Cloudflare-oriented?
Some people have it, like Cloudflare. Then we convinced Vercel to be there as well, and we started getting more people. You have the two sides: the DNS providers and the service providers. We’re the service providers, for example, and then the DNS providers implement it.
It’s a crazy industry. GoDaddy charges a lot of money for that. It’s so weird. With IONOS, it took us months to get to the first person. No one really maintains the protocol and things like that, but it’s beautiful to have a protocol where you can do OAuth for domains.
Yeah, yeah. For listeners who aren’t familiar, if you haven’t tried something like that, it’s literally like using Plaid or OAuth. You only click buttons. You don’t type in anything. There’s nothing to get wrong. You basically just sign in, as long as you trust this thing called Domain Connect, which you’ve confirmed is legitimate. Then it’s trusted. I literally used it for the first time two weeks ago. I was like, “What is this? I’ve never seen this before.”
Yeah. It’s really cool.
One of the things I’m trying to figure out is how you start this company. You said you tried to get in touch with the IONOS guy. I wouldn’t dare start Resend without knowing someone on the inside at Gmail. Even now, it’s like, “Okay, can we vibe-code our own email thing?” Clearly, no. But that’s even with tools today. You did this years ago. You want people at the domain providers, right? You need the guy.
Yeah, but you need to pick up the phone. There were many times we had to pick up the phone. It’s fascinating. It’s an industry that relies a lot on relationships, and we didn’t have those relationships. So we hired for a few key roles, like deliverability roles—folks who had been at Postmark since the beginning or at Mailchimp 15 years ago—and then we leveraged that network.
What’s interesting now, of course, is that I didn’t know that when I started. I could never predict what was going to happen.
You didn’t do the research?
No. I could never predict what AI was going to do and all that, right? But I think we’re in a space where I’m glad we chose infrastructure. I’m glad we’re infrastructure as a service and not just software as a service.
You had a little bit of that, right? You made React Email the year before, so you did dip your toes in open source work, and then came the company.
Yeah. React Email is our Next.js, right? We wanted an open source channel for top of funnel. I remember doing some things very intentionally. If you go to the docs, for example, you’ll find docs about how to set this up with SendGrid, Mailchimp, and Mailgun. I wrote every single one of those docs. I wanted this to be for everyone, and we were trying to redefine how this works.
It was a strategy of giving, giving, giving, and then asking. If people are building an email template using React, now they’ll need a platform to send it. Where do they do that? Here’s Resend. It was a storytelling kind of way of approaching the market. We’re going to fix the first part of the journey, which is building the template. Once you have the template and that problem is fixed, you need a platform to send it. Here’s the platform.
But did you have the idea for Resend when you started React Email, or was it a side project that then became a company?
We started Resend, and as we were building it, I was doing a lot of user research. People were saying, “The pain is writing the template. I get the Figma file from my designer. It sucks. It’s HTML tables and everything.” They would cut out the images.
Oh my gosh.
It’s in there because there’s nothing else. And then there’s a special syntax of HTML that only works in email.
Right? It’s a mess. Or CSS.
It’s all inline CSS.
That’s right. And then, for Outlook, you have some special comments, and then you’ve got to… It’s such a mess. It’s like a mess. To me, the reason React works is that it is basically a compiler. Somebody writes this once, and then, as long as you fit the API, it’s going to work everywhere. You can import your tokens from Tailwind. Now you have a type definition with TypeScript. No more empty comma in your emails. You can actually have string validation or whatever you want.
I think one of the initial people who pushed back on React Email was like, “Well, you’re not actually using React. There’s no state.” React is supposed to be views as a function of state. So is it mostly JSX as templating, or is there animation or interactivity? Expanding accordions?
No, nothing like that.
Okay. So you cannot ship JavaScript in an email. What you do is—
We basically use—
But CSS is Turing complete, if you—
Yeah.
Okay, cool. I do want to continue on this email path, but I see so many names I used to know. I was wondering if you can indulge us a bit and go through some of these names—what you learned from them, any cool stories, anything.
Man, so much. I remember the very first year of Resend, I was like, “I’ve got to meet the founder of Mailgun,” because I was still a user of Mailgun. I talked to Ev. He built—oh my gosh, what is it? Blogger? Medium? No, it’s a separate product. It’s a security product that he’s building now.
I remember talking to him. I was like, “Man, why did you sell your company so quickly? What was the thing?” He told me, “Oh yeah, 90% of our R&D was just to fight abusers.” It was a terrible business. He said, “You shouldn’t go there.” I didn’t want to live in that world at all, so I was very worried about that. That was a big lesson for me: we’ve got to automate this thing. There’s no way we’re just going to build a massive organization for it. We do have a lot of people interested in safety now, but that’s not going to be our core business. We have to be pushing the industry forward.
That was one. SendGrid has always been my favorite out of those, and it’s always been the behemoth—the biggest, the…
It’s like, “Oh my gosh, how are we ever going to beat SendGrid? It’s just impossible.” Earlier this year, we passed SendGrid in downloads, and I was like, “That’s insane.” Now we’re more than double.
So, in terms of unique users—or what’s the number?
npm downloads of the SDK.
Is that the best metric out there?
I think there’s something to that metric in terms of adoption, but if you look at how many emails they send, it’s obviously much more than us because they have big enterprise customers.
One thing that’s surprising is that the whole point of the Twilio–SendGrid acquisition was that you’re no longer an email service; you’re a communications platform. You integrate across everything. It’s not just about cross-selling. It’s about reaching people across multiple channels, or having the scale. Twilio is, I don’t know, let’s call it a $13 billion or $15 billion company. You should have all the relationships, all the talent, all these things. So how do you compete with that?
I admire Twilio so much. They kind of invented developer relations, or “Ask Your Developer.”
Yeah, “Ask Your Developer.”
Now, when I drive past that 101 billboard and see “Customer AI” or whatever, I’m pissed. I’m like, “Why?” You neglected the thing that got you there. Jeff isn’t there anymore.
He bought The Onion.
That’s right. So I feel like they’re sleeping on it. First they did that with SendGrid, and they’re doing that with SMS, too.
My lesson here is that I’m trying to figure out the economics of scale in this industry. Clearly, they were 100 times larger than you when you started. How come they can’t use their size to just win? Maybe there are no benefits to scale. Maybe you just make a better product and slowly get there. I’ve also always wondered why Twilio acquired SendGrid and not the other way around.
Twilio was bigger. SMS is big.
But if email and communications are commodities, like SMS, your margins are even—
Worse.
Yeah, even worse. Have you looked into SMS?
Yeah, for sure.
Resend SMS?
I think one day.
What are the channels that matter? Email, SMS, what else?
WhatsApp is big in many different countries.
Oh, WhatsApp is free, right? You don’t pay anything.
Yeah, but you have to go through Meta, and now business messages are going to be charged. There’s some friction there, for sure.
Phone calls?
Phone calls are a good one. I’m less interested in that, though.
I don’t know if it’s a good one. I never take any robocalls.
That’s true. I’m less interested in that, too. What I care about is what lets me connect with my audience. That’s what I ultimately care about. I don’t care about email.
Direct messages?
Direct messages are one. It’s hard because you can’t message a big crowd. Slack might be one, though.
Oh, yeah, Slack.
I have so many Slack connections, and I need to share news about a new feature we built.
Pylon? I don’t know if you guys use Pylon.
Yeah, Pylon. I think they kind of innovated this Slack support thing.
You can obviously trivially build your own.
Okay. The other one I want to touch on is AWS SES.
My connection to this is that I learned to code with freeCodeCamp, with Quincy Larson and everything. They have about 5 million newsletter subscribers, sent through Amazon SES for about $5,000 a month.
That’s cheap.
It’s pretty cheap. I don’t know what it would cost you.
More than that.
To me, it’s the low-cost player. It’s Amazon; they’re super established. What do you take from them, and what should people know to stay away from?
We started with SES, too. We knew that if we were going to innovate, I wanted to innovate on the developer experience layer because I felt that was the underserved space. So let me have SDKs for every single language. With Serverless, I remember we highlighted SES because no one had guides on how to send an email with Lambda or something. There was this underserved space in developer experience. We started with the highest-level layer, and that served us really well. We were able to scale because we were using SES behind the scenes.
It turns out there are a lot of limitations in how you structure domains, and they have this concept of configuration sets. The hidden cost of SES is that you might be paying only $5,000, for example, but then you need SNS for the events: all the feedback from bounces, complaints, opens, and clicks. Now you get the events. Great. Then you need CloudWatch for the logs. You need a queue because there are going to be a lot of events, so you add SQS. Once you have SQS, you start adding all these different layers. Want webhooks? Now you have to build another thing. SES is a great platform; it’s fine. But you’ll quickly find that your footprint is much bigger than you’d imagine because you need all the apparatus around it.
Another YC startup that came up in the email area is AgentMail. I talked to the founder, too, and he was like, “We’re going to build on top of SES and not really compete with them.”
For us, that’s changing now. We’re building the infrastructure and going down to the metal. We’ve reached the point where it makes sense to innovate on the most basic layer. We can get better throughput and better deliverability if we focus on that. It’s a big project. We’ve been working on it for months, and we’re going to announce something soon about that direction.
What’s the hardest thing? Is it trusted IPs? Is it compute and data centers, or what’s the blocker?
Definitely IPs. You’re buying a whole IP block, and then you’re managing the IP pools yourself. That’s one part. Then you’re almost doing traffic control, controlling the throughput into each mailbox provider, and you need relationships with the providers. Once you have that infrastructure established, you have much more control over what you can give users.
One of my biggest things is observability. As an engineer, I want to see exactly what’s happening, and I’ve always felt like there wasn’t enough information in this space. When an email goes out, is it really delivered? Is it in the spam folder? Is it in the Promotions tab? I never know. You just trust that the email goes out, and then you hear from a user, “Oh, I didn’t get the email.” I’m like, “What happened?” There’s this empty space between those things. The closer we get to the metal, the more data we have to surface for users. I’m really excited. It’s the flagship change we’re making now.
You’re becoming—It’s really interesting for all these developer-experience businesses. You start with a thin layer, like React Email, and build on other people’s infrastructure. Once you get big enough, you can build your own infrastructure. Are you doing anything differently now that you own everything?
I think we’re approaching it with the same naive mentality we brought to all the other problems. We’re like, “Okay, this shouldn’t be this way.” We’re challenging the most basic things: this shouldn’t be this way. The P99 shouldn’t take this long. How can we make it faster? That’s one of the things we’re doing now: focusing very deeply on performance. We’ve always been a Node.js shop, with TypeScript everywhere. Most recently, in the sending path, we stopped using JavaScript. We have Go there, and we’re changing other parts to be more performant in other ways.
Because now languages don’t matter.
Exactly. It’s so cheap to pick a problem and rewrite the whole thing in whatever the fast language is. I’m running Rust projects, and I don’t really know Rust. I know some theoretical things about Rust, but…
I’ve never really bothered to learn it.
Mm-hmm.
I feel like we’re not really covering the recent product surface area.
Yeah.
Maybe you want to go.
No, no. Everything we’ve talked about is what you’re solving, and then there’s also just sending emails from the CLI, Super Agent Forward. You have some takes on this. I think the other day you tweeted, “MCP is so back,” right? Over a million MCP calls. What’s the AI side of email?
Yeah, everyone was saying MCP was dead a couple of months ago. The only thing that matters is the CLI.
What I’ve noticed is that I’ve stopped paying attention to what’s in and what’s out. I really think about these problems as, “Okay, how do I want my product to be?” I want my product to be formless. Whatever people want to do with the product, it should be possible.
That’s why we’ve always been API-first. I had no idea how much this would help us down the line, but I knew I’d never be able to cover all the use cases, so I thought, “Let me just give them the API. You do whatever you want.”
With MCP, it’s fascinating. We’ve had at least 2x growth every month. In August, the numbers were even higher. This was before the end of August. I think it was 1.2 million by the time the month ended. People are starting to work from Claude Desktop or Codex. If work starts from there, then you need those integrations. It’s super important.
I just want to clarify. You showed us the month-by-month numbers for MCP.
Mm-hmm.
It picked up from, let’s say, 50% month-to-month growth, which is already super high. Then it starts to double basically every month.
Mm-hmm.
Did you invest in MCP, or did you just have an MCP and, after some time, it got picked up?
The fun fact is that our very first MCP was built by Yoko from a16z, our investor.
Oh, wow. Value add.
Yeah, right.
Brilliant.
She built the open-source version. It was a remote— a local MCP. Then it started to grow like crazy, and I was like, “Man, why are people using this? It’s so hard to set up MCP.” This was last year.
Yoko.
Yeah, exactly. She built that, and then I just asked her, “Hey, can you transfer it to our org?” It was just an engineer building something. It was nice to see the traction it was getting.
But to do MCP right, to do the remote one, we had to go back and revisit our entire permissions model because we needed OAuth. Then we were like, “Okay, now we have a blocker. Let’s build OAuth.” So we built OAuth, and then it was, “Okay, RBAC. Let’s look at that. Let’s look at how users have access to each one of the resources.”
It was a whole journey to build OAuth. After we had OAuth, we could build the remote MCP. Then came distribution. We had the remote MCP. Great, great, great. Observability was in place. Great, great. All the Datadog dashboards. Okay, now we’re all in. Let’s talk to everyone.
That was hard, too. It’s not as easy—I wish I had all your connections.
You can talk to me.
Yeah. I’m just supposed to talk to you. But talking to the labs and trying to get into the official stores and plugin marketplaces and stuff like that was hard.
Do you think that actually matters?
For sure.
I think sometimes I’ve been on the other side where I try so hard to get these partnerships, we launch it, and then nothing.
Oh, man.
Which one matters?
As a consumer, I make decisions based on whether I have access to that data or not. As a buyer of B2B software, if I go to the cloud connectors and I don’t see the company listed there with an official MCP, I don’t trust it as much.
We had this with Plausible and PostHog. Plausible had an MCP built by Sentry. I don’t care. If Plausible doesn’t maintain that and it’s not official, I can’t trust it as much. I love the Sentry team and I trust them, but I need the owner of the software to own this piece of the integration, too.
I made the call: “No, we’re getting rid of Plausible, and we’re just going to use PostHog here because I have the MCP there.” I did that as a consumer, too. I’m into running, and I thought, “Oh, let me…” Strava has an MCP. I can access my runs, so I use Strava because of that.
If it’s, “Oh no, we have an MCP, but you have to connect like this,” like the xMCP, for example, where I have to use the local one, it’s just so frustrating. It works on the desktop client, not on my phone. To me, if it’s not there, I don’t use it.
Apart from being officially maintained by the vendor, what makes a good MCP versus a so-so MCP?
Even for ours, as I interact with it, the tool calls are always collapsed. I’m always opening them and trying to understand what it did. Just a couple of days ago, there was one call where it couldn’t put everything on the context for some reason. I thought, “No, this shouldn’t happen. How much data are we sending? Maybe we’re sending way too much data to the client.”
The number of tools was a bigger concern earlier. People thought if you had way too many tools, it would get confused. I don’t see that as much anymore. In fact, we’re just adding more to the MCP. We had maybe 20 tools when we launched, or even fewer than that. Now we have 80-something tools.
What the fuck?
Yeah, and I’m adding more and more. That’s the long tail of the tool footprint, what I’ve noticed. You can do only 90% of what your dashboard does, or you can do 100%. Getting to 100% coverage of what your dashboard can do is actually very, very difficult. Now you’re doing usage, billing, team invites, account provisioning, deletion. Should that be exposed as an API or not? It’s really hard. It’s a difficult problem.
In fact, you’re one of the few Latent Space guests I can ever talk about this with. I think this is one of the mistakes of Next.js.
Oh, really?
Because Next.js, Server Components, and the Pages Router or App Router make it easy for you to bind data to your page. So you never expose the public API.
Yeah.
By default, the thing that performs well and that they encourage you to do means you won’t have an API.
That’s so true.
Then you have to create separate code paths or whatever. You should always have a completely headless backend, and then your frontend just happens to be one of many clients that connects to your backend.
We started with our API on Next.js. I remember that. First of all, the bill was just so high that we were like, “There’s no way we can pay for this.” So then we moved to a traditional Node server, or whatever you call it. Our dashboard was always a private API consuming the same thing. We had to duplicate a lot of code. It was a mess.
Now it’s so much better. Everything’s in a monorepo, and you can reuse stuff. Ideally, those frameworks just give us those components by default. We had to build a public API separately from that, and it sucks to maintain both.
Okay, one more thing on the MCP front. You’ve got 80 tools in there, and you’re adding more. You’re very counter-consensus compared to Cloudflare, which only has 2 tools: search and execute or something like that. So far you haven’t felt that issue. What would make you change your mind? The numbers are going up, but do you actually know? Do you talk to your customers and hear them say, “This is the best MCP on earth”?
I do try to get a lot of feedback on that. The way I like to build products is that I’ve got to use the product myself. The investor update that I sent you yesterday was built using the MCP. It was fetching data from one place, from another place, and then—
You’re making me feel bad now.
Don’t do all this and don’t use MCP now, basically, is what I’m hearing.
Well, the other thing is, you do have a CLI and a regular SDK. Are those also growing a lot?
The CLI is definitely growing, but not as much as the MCP. I have a lot of fun with the CLI, too. I’m proud of this: I built our Careers page on a weekend or something. If you go to any job and try to apply, you have three ways of applying. You can apply via CLI or API. You can log in, and it has an interactive mode. We also have skills baked into the CLI, and you can just ask Claude to apply for a product designer role at Resend.
Okay, so you’re very— It feels like you’re filling yourself with slop, especially in Careers.
Yeah, yeah. I think maybe those are also the people you want, right?
Totally. I actually have a filter. If the source is the CLI, I look at it differently because I know people are going the extra mile, setting it up: “Oh no, I have to download this thing, authenticate, and go through that flow.” But people who do that always tell me, “Man, I’ve never applied for a job using a CLI before.” Even if they’re not accepted, it’s a different experience from other job applications.
Yeah. Even applying for one job through the API.
Exactly. It’s the same thing as the CLI.
Yeah. So one of the things I’m trying to figure out is product philosophy. You’re clearly maxing out everywhere: every SDK, every language, every surface area. Obviously, that’s easier with agents. But some founders I know would say, “Hey, this doesn’t matter. Give that to Greenhouse or Ashby.” It’s your job application; you’re not going to differentiate this way, right? You clearly have a different philosophy. You care about everything, even though some of it is probably wasted.
The way I see it, I’m trying to optimize not only for privileged users—the ones with the best Wi-Fi and MacBooks in San Francisco or New York—but also for people at the margins in other countries with terrible Wi-Fi.
You mean Brazil?
Brazil. Growing up, that’s how it was for me: a terrible Windows computer and terrible internet access. That’s why we have a free tier. We don’t have it without a reason.
When we have all the SDKs, I’m maxing out on them. Do you think building the .NET SDK was fun? We looked into Stainless, Fern, and the third one, but we didn’t like what was generated. Every one of those SDKs is generated by humans—the Java one, the Rust one, all of them. I do that because if you’re using a product, you should have the best experience in the language you’re using.
“But there are only 500 people out of the 4 million.”
I care about those 400 people.
Wow. Why not? You have 4 million users.
That’s why not. There’s a lot to do, but it’s good to have that opinionated side toward building this.
Any other product surface area we should cover? You covered the CLI, templates, and MCP. I think it’d be good to get your high-level perspective on how much of this is traditional. You have dashboards and templates for manually writing emails, you have programmatic tools, and you’re also fully agent-pilled. Where do you see it going? How do you think about building for a fast-changing world?
There’s definitely still a component of email moving out of single-player mode and becoming more multiplayer. We even built our editor that way. It’s multiplayer, and it fixes all the conflicts if two or three people are working together. We actually run all our all-hands through a private Resend account I have with all the teams. The all-hands is pretty much: you go there, and it’s a huge—
You make an email?
It’s an email, and we send it together at the end of the all-hands. It’s super fun. You have about 20 people in the same document. That editor alone is so hard to maintain, but I love it. I love that we have the UI for people who need it.
Yeah.
Some teams hate having the email template pushed into their source code. They just want to build a visual editor, make a template for the marketing team, and call the ID. “I don’t care about the template. I just want to call the ID with variables, and I’m done.” Other teams say, “No, I want to commit. I want source control over this and to own this piece of work.” I try to tailor the product to those different audiences and build something that feels like we care, which isn’t the feeling I get elsewhere.
By the way, what does the Resend team use to read email? Superhuman?
Yeah.
Have you seen the latest Superhuman update this morning?
I haven’t.
You’ve been busy with other launches too, right? For listeners, we’re recording this on the day of the Astro launch, which I was pulled into. I’m not sleeping.
What’s wrong with it?
They just updated the UI a little. There’s a sentiment that it’s going in the direction of every other acquisition: now you’re fitting a brand that has other interests, like Grammarly and other things.
Okay.
This is just email. I do want to keep up to date on email tech in general. You’re on the sending side; other people are on the reading and receiving side.
Yes. I think increasingly, agents are reading my email. I actually don’t read most of my email. I have Poke and two humans helping me with it.
Nice. I love that.
Poke is pretty good. Honestly, Instinct is super hot now—a $2.5 billion company with no revenue. People are trying to read email with agents. We’re definitely looking at that space. That’s another thing I’m spending a lot of my time on: how you can programmatically create dozens of inboxes, and how those inboxes are accessed by agents.
Yeah, because you enable receiving email. What’s it called? Inbound?
Inbound, yeah. Email inbound is the most basic abstraction. Now we’re going one step further: maybe you need 100 inboxes for some reason, and here are the APIs you need on top of that. You need a read API, a mark-as-unread API, a star API, and labeling. We’re going to launch a product around that soon. But again, it’s all API-driven.
Fully. The comparable products I’ve been looking at are Missive, Help Scout, and other shared inbox products like Front.
Yeah, so hopefully this can be a version of that. I can share it with you. There’s a feature flag for you if you want to try it out.
I’m very interested in email tech. And then, what percentage of your email goes to G Suite? What’s the market share between Google, Outlook, Apple, and Hotmail?
I don’t have the numbers off the top of my head, but the vast majority—at least 60%—is Gmail. Then you have the big three after that, like Outlook and Yahoo Mail, and then a long tail.
Apple isn’t that big.
Apple is big.
Yeah. Especially with the privacy thing they did, it’s harder—
You can’t sign up with an Apple ID?
What? Sorry?
Apple auto-generates an email address for you.
Oh, yeah, yeah, yeah. If you’re signing up for something—
Yeah, yeah.
I use that. Is it a problem for you?
They’re stricter about what can get into those inboxes. As a domain owner, we have an in-product wizard to help you register with Apple so you can get your emails delivered there.
Yeah. I understand. Obviously, this is your job. The whole thing is deliverability. If you don’t have deliverability, you don’t have a business.
Exactly. That’s kind of what they do. We can have the most beautiful Rubik’s Cube on the website, the coolest handbook ever, but if it’s down, it’s all for naught.
And how correlated are they? If you get banned, downranked, or put in the Promotions tab in Google, does that transfer over to the other providers, or are they very decentralized?
Thankfully, now there’s a lot of reputation associated with the domain. At the beginning of the email industry, it was very IP-based. If one IP pool was damaged, you got the consequences. Now people create different subdomains depending on what types of emails they’re sending: one subdomain for transactional emails that maybe don’t have click and open tracking, and another for marketing emails. Those subdomains help segregate the reputation of the stuff you manage.
Would it be something like transactions.resend.com?
Yeah, updates, notifications, whatever you want to use.
Could I have more dots if I want, like a.b.c.resend.com?
You could, but we just see that having five subdomains isn’t really practical.
People do some special ones. We do that at Resend for our own emails. Marketing has a different stream from transactional emails. Even the investor emails have a different one. I use confidential.resend.com to send those emails.
I didn’t even know you could do that.
Yeah. That’s good advice. It’s good practice.
So before, it was very IP-based?
Yeah. Now it’s very domain-based.
I’ve received so much bad advice. I had to buy a separate domain because we use ai.engineer and latent.space. These are all weird TLDs, and nobody recognizes them.
That’s right.
Nobody recognizes them, right? So I was told to get .coms for everything, and then maybe that would improve deliverability. The problem with that is, let’s take Coca-Cola. Coca-Cola has a domain. Now you’re Coca-Cola, and you’re like, “Oh yeah, let’s buy another domain that looks like our domain.”
It looks like phishing. It looks exactly like phishing.
So the chances of damage are way worse. Is this still a thing when you’re buying the .com? I’d get it the other way, right? You’re latentspace.com, and I buy latent.space. That looks like phishing, right? But the .com still?
Man, you’d be surprised. The things I see on a day-to-day basis are crazy.
So what I’m hearing is that .com is worth it.
Yeah.
That’s all the sort of business of email. Obviously, some part of it is that you just got started before ChatGPT, and you were a leading player that people liked, so you popped up organically. You didn’t do anything special for AEO. What did you do for AEO?
3. Getting Recommended by AI
I wouldn’t say we were the leaders. Coming up in 2023, very few people knew about Resend, and we were so behind. Twilio, Postmark, and Mailgun had invested in SEO for more than 10 years, so getting ranked for anything was an uphill battle. It was so difficult.
We decided to focus on the docs. We were very intentional about the docs and the long-tail searches they could target. For example, instead of just having docs for Python, we’d explain how to set it up with Django.
How to do something with something.
Exactly. Super short articles focused on that.
That’s traditional SEO.
Traditional SEO. Nothing fancy there. Then, when models started to come along, I thought maybe I needed to change the content on the page. If people are asking questions, I need to create content in an ask-and-answer format. So we added a lot of FAQs. Now, almost every page at Resend has an FAQ.
You’d go to the pricing page, and of course pricing almost always has an FAQ. That’s pretty standard.
Yeah.
But you do FAQs on all sorts of pages.
Security, too. You see the pattern: all pages have that, because people ask, “Is Resend SOC 2 compliant?” Then the answer is right there.
We added llms.txt to our docs, thankfully because of Mintlify, which we use. Then we asked how to bring that to the marketing site, too, so we added llms.txt to every marketing page. We did the same with the docs’ Markdown versions. I’m going to make Markdown versions of every page on the marketing website, too.
Are you using content negotiation, or do you just append .md?
You just append it. If you go to the security page and add .md, it works. You can also use content negotiation, and it’ll render the Markdown version on that page without going to another page.
For listeners, I’m not sure that matters. I’m really not sure.
I mean, you’re doing really well, so…
One example where it changed things—and maybe it doesn’t help with distribution, but it helps with accuracy: go to the pricing page. I remember one random Sunday, I was doing competitor analysis. I asked, “If you send 1.2 million emails using Resend, X, Y, and Z competitors, who’s the cheapest and who’s the most expensive?” It got it completely wrong. It got the Resend number so wrong, and it actually said very bad things about us, like, “Resend’s pricing is very obscure.” I was like, “What the fuck? It’s not obscure at all.”
Oh, so you made a calculator?
Yeah. The problem was that agents don’t really run a JavaScript widget.
Okay.
So I made the .md version of this page.
Which gets a tremendous amount of access, by the way. You get analytics on this?
Yeah. Not through the regular analytics, because we can’t track that.
Yeah, it’s server-side.
So we have to go to Datadog to figure that out. It was really crazy. I made that change and pushed it, then I asked again a few days later, and it got the numbers right.
I see.
I’m like, “Okay, this already—”
It’s a good way to live-update and fix stuff in your search entry.
Exactly. This is what matters: accuracy.
But it does mean you don’t have a single source of truth. You have to update the Markdown, and then you also have to update the—
That sucks completely. But if I wanted the happy path and the easy path, I wouldn’t be doing email. That’s the price of it.
Well, if you’re good at React, you can write a custom renderer.
You know how to do this. What else? We talked about the basics: FAQs everywhere and basic SEO stuff. Comparison pages—us versus SendGrid, us versus Mailchimp. Although our page could be much better. Go to—
This isn’t there.
I’m sure you have it somewhere in a footer or something.
Just go to /migrate. I can give you some—
Ah, okay. By the way, for 404 pages, I always think we should guess what people meant. Take the slug and find the nearest neighbors.
Oh, yeah.
In case you ever break a URL, it just auto-fixes itself. Don’t just serve the 404.
My intention with those was to try something that wasn’t exactly a comparison page.
A comparison page.
Yeah. But I don’t think it was a great idea, because our competitors actually built comparison pages, and they get recommended more than ours. What we try to do, if you go to Postmark, for example, is demonstrate first: “Hey, if you want to migrate, here’s a code converter.”
Insert an example.
So you can just put any code block there.
Why? Why, dude? This is so extra.
That’s what we do.
There’s a cube on the landing page. Everything’s extra.
The cube has been there for 2 years. Adoption is something I pushed this last weekend. Now we have a graph that compares our adoption with every competitor. Again, one thing I learned from the pricing page is that JavaScript widgets won’t be executed, so I spell out the same data in text so LLMs can read it.
I think they’re getting somewhat better. I see a lot of agentic use: load the website, click on an image, interpret the image. It’s slower, but with computer use, they’re getting much better.
I agree.
If you’re fetching information and just comparing things, they don’t do that, right?
I think they’re starting to go more toward computer use.
I’ve been doing a lot of “compare this and this,” and browsing a website is pretty computer-use-heavy, I think.
Okay.
I noticed you don’t have another classic AEO tactic: “Here’s how to connect our thing with Claude Code. Here’s how to connect our thing with Codex.”
Oh, yeah. It’s all an AI page.
Okay.
We do it as a separate AI page.
Okay. Both, both MCP. There are some things that are interesting. If you go to the legal page, resend.com/legal, you can summarize or ask it to explain the complexity.
I see.
That’s just an external link.
Yeah.
Again, I’m not sure that matters.
You mentioned a lot about recency for AEO compared with traditional SEO.
Yeah. That was the thing I was really curious about in the beginning. There was no reason for any of the models to recommend us because we were so behind in SEO. We don’t have the typical glossary page or generic articles that are just SEO-heavy. I was trying to figure out what was going on. I could tell that, at least with Opus 4.5, there was a recency bias in terms of the technology and where it was fetching data. Recently, they said maybe Replit isn’t necessarily something people suggest anymore.
Newer tools started popping up, and Resend was one of those. Because there was a lot of organic content about the tool, we started getting recommended more. The advice here is to try to get as many people as possible, in as many places as possible, talking about your product, because it definitely has an influence.
I think that’s a class of its own. Did you ever engage SEO consultants or AEO consultants? Are they useful, or a waste of money?
We never really did that.
You seem like you’re doing everything yourself.
Yeah. I think you were the AEO king.
You had that 2024 tweet: “Let’s talk about AI SEO.” I think you were one of the first ones to pick it up.
Yeah.
The general state of things, since we did an AEO or GEO episode around this stuff, was that most people said, “Just do what normal SEO does, and don’t block AI chatbots.” Most tools, like Profound at the time, would track your mentions and let you run experiments. Those things are par for the course and normal. I’m trying to figure out what to do. People are pinging me because I’m also doing AEO experiments.
Hmm.
I have an unfair advantage because I’m a content business. I pump out content, whereas you actually have a product.
Yeah.
You’re doing very well, but how closely do you track how well your docs are used? How often you’re recommended compared to competitors?
I feel it’s so crucial. Before, if we were unsure which tool to use, we’d ask Google, and it would show us comparison pages or the first few results. Now you just get one response out of the gate. If you ask, “Which tool should I use?” people want to go with the most frictionless option.
Yeah.
I think a free tier is very important. Michael Grinich from WorkOS is trying to promote this idea of authless usage. I don’t know if you’ve seen this. It’s very similar to Netlify, where you don’t need an account to do the basic action. It’s unauthenticated usage of your product.
Yeah.
It’s obviously very dangerous because it’s prone to abuse, but it’s the easiest thing: no sign-up.
That’s the thing I’m trying to figure out. I’m debating whether to do that or not. Do we need a provisioning API?
Provisioning and claims, right?
Yeah. But then there’s the free tier.
Yeah. I don’t know if it’s something about our industry, but there’s just a lot of abuse.
Yeah, you just have abuse.
Yeah, a lot of abuse.
I think those are the main tactics. The other thing I’d tell listeners is to hook your clanker up to your analytics and your codebase. You need to fully complete the loop and basically automate research.
What I have now is my cycle. I have both a Latent Space CMO agent and an AI Engineer CMO agent. They spawn other agents. Their job is to research ideas and keep track of things we’ve tried. Then I sign off on the things we want to experiment with. We have an experiment window, and we run the experiments. Other agents monitor those experiments.
That’s the auto-research loop for doing this stuff. You also need to give it tools. I signed up for Ahrefs data for SEO. Most recently, I made my own AEO tracker, which you can make. Then you just want to make the number go up.
You don’t know; you’re throwing things at the wall. Does the thing on the legal page matter? I have no idea, but I’ll keep throwing things at the wall and seeing what works. I think Reddit doesn’t matter. Hacker News used to be big for you.
Yeah.
Now, not as much. Twitter doesn’t matter as much. It’s weird.
It’s so crazy. Do you have a tracker?
No. We use all of those Profound-like tools. We tried them all. People ask me, “Should I use Gauge or Profound or this other one?” I’m like, “Yeah, we have 4 subscriptions.” That’s what it takes now. You have to see which one works best.
What I miss from those tools is, “Give me the action.” I think the problem is they don’t tell you what to do because they’re also trying to figure out what needs to happen. “Write an article on Wikipedia.” Yeah.
The old-school SEO tools like Moz would have more opinions about these things. Right now, the AEO tools, from what I hear—and I’m not an expert; we should probably talk to Profound—are mostly reporting tools. They’re analytics reporting tools, and that’s fine.
It’s a big business.
Exactly. Ahrefs is a Singapore business, and apparently they do, I don’t know, $300 million.
Yeah, totally.
The other parts of your business—we’re going to talk about the parenting side and the personal side. You seem very chill. You have 50 people working remotely all around the world. You’re worrying about MCP and AEO. That can’t be how you’re closing all these big customers, right? Where’s your fancy sales guy with the steak dinner and the golf course?
We’re just starting that motion. We’re very lucky that we’ve been able to go this far without having it. I’ve worked at an enterprise-only company in the past, so I knew what that takes: the steak dinners and all the apparatus around the sales organization and enablement. There are all these things that are needed. I knew that when the time was right, we’d have to turn that key.
But I’ve been very cautious about doing that at Resend, in different layers, instead of doing everything at once. Maybe it’s because we’re remote, but I think we’re very protective of the culture. Let’s make sure we don’t damage the culture all at once. Let’s try to figure out who fits. We just have one AE now, and it’s been helping. It’s been going super well.
Good enough. Once you double your ARR because of a couple of deals, you’re like, “Okay, this is a good sign.” People want to pay you more for things like having an agreement with you and having your software on their premises or in their VPC.
For sure. What I’m learning, too, is that you’re building a different product, so that’s how you need to see it. It’s not only about building a different machine internally as a company. As I’m designing the company, I’m designing which department, who sits where, and how each person is better equipped to do their job.
For enterprise, it’s actually a different product. What they care about is very different from what $20 users care about. If I just add WorkOS, SSO, SCIM, and audit logs, it’s not going to be enough. It’s not just putting those items on the pricing page. It’s a journey, for sure.
Any other things you want to reflect on in the journey before we go into the personal side?
No, I think we covered a lot.
Okay, then I’ll talk a little bit about my use case. I’m a content creator. The bulk of your business, I assume, is transactional email.
That’s right.
You have marketing emails and inbound emails. I’m sure that’s less outside of the business. I pay Substack something like $10,000 a year to send my emails, and that seems too high. Why is marketing email so different from transactional email?
It’s a great question.
Your business model is also different, right?
It is. We’re more expensive, too.
Then I was like, “I can’t do this.”
The industry has this pattern of charging per contact. That’s the first thing. They try to price around the value you create with each contact. But now it’s crazy. The AI companies we serve grow so fast. Mistral’s users recently—the amount of users they add is just insane. The business model of contact-based pricing is kind of broken, and we’re realizing that and trying to fix it too.
But that's how it had always been. There’s something about transactional email: you put a message in a queue and get it out of the queue. With marketing email, you’re building an engine of batches and intervals between batches, analyzing and increasing or decreasing the speed as the sending is going, maybe blocking.
So is there a maximum rate you should send emails at a time?
Yeah, for sure.
Per domain, specifically?
Yeah.
Okay. Let’s say I have 100,000 emails to send out. It should take me 10 minutes?
Yeah. For example, now, with the 4.3 million users that we have, it takes more than a day.
Oh, what? No, that’s unacceptable.
Yeah, for sure.
So then I pay you more to get in a different queue?
Yeah.
Okay. So you have to always be trying to figure out, “Okay, we need more at the door.” How does that work?
Yeah, yeah, yeah.
Oh, that’s interesting. It’s like rate limits for token limits.
Yeah, exactly.
Where’s my reset button?
Yeah, exactly.
You also went into YC and started this company as a young father. Part of the thing is that there are a lot of people without kids in San Francisco, and I think that’s part of the consideration. A lot of people listening might think, “I can’t be a founder. I have this great idea, but I have a young family.” How do you manage?
4. Choosing Family Over Growth
You’ve got to preface that with the fact that you’ve left San Francisco for your kids.
Yeah.
This is recent news: I’m spending more time in Brazil now because we’re expecting our second kid. At first, it was crazy. I came to San Francisco when my daughter was one year old, and every YC founder around me had no idea what it meant to have kids.
Yeah. What is this reproduction?
Exactly. You don’t even have Series B. What are you doing married?
Totally. Most of them weren’t married at all.
It’s crazy. I always try to look at those things. At first, you see them as disadvantages: all of these guys came from MIT and all these amazing schools. That wasn’t the case for me. They weren’t married, and that wasn’t the case for me. They didn’t have kids, and that wasn’t the case for me. But I still went for that program thinking, “I just want to build the best company in this batch—not number four or number five, but number one out of 250.”
Oh, yeah? Which cohort?
Winter 2023.
You realize in the middle of that there’s no winner in YC. There’s no such thing.
Yeah.
I’m pulling up Winter ’23 now just to see who I know.
Especially what you understand is that it’s an infinite game. Even if you had an amazing batch and raised the biggest round, it doesn’t matter because life goes on and continues. Many of the people I met there aren’t here anymore.
Yeah. Co-founder breakups, seven pivots, and they’re dead.
But we’re here. I was very much like, “No, no, no.” Even when we bought the domain resend.com, we were like, “It’s going to be an amazing website with an amazing .com. It’s not going to be a .io or whatever.”
Doing that with kids was so freaking hard, especially because I wanted to enjoy my first daughter. I really thought about this. There was one point during YC—this story translates to what I’m going to say. I came to an event, and PG was there. I thought, “Oh, man, Paul Graham, the father of everything. God is here.” Everyone was paying attention to PG, and then he said, “You know what? We always tell you not to do something as partners, and then you end up doing it. Months later, you come to us and say, ‘Oh, I regret doing this or not doing this.’”
So he said that, and then I went to my partners and asked, “What are the things I’m doing now that I shouldn’t be doing, that you think are wrong?” One partner—not my main partner, but another partner—told me, “You worry too much about design. You care too much about that. If you didn’t care about it, you’d be much further ahead.”
That hurt me so badly because it was true. My slides when I was pitching were beautiful, and he was like, “Doesn’t matter.” I was like, “Yeah, now I know it doesn’t matter.” That really hurt, and a few days later I reflected on it. I thought, “If I were to build a product with this identity that only 1,000 people used, I’d prefer that to having a product with 10,000 people that I don’t want to use myself.”
So, you know what? I’m going to follow what I believe is right. In this industry, everyone had terrible products. There wasn’t enough design infused into them, so I felt that was a differentiator, and we needed a differentiator in such a crowded market.
I go back to that with my kid, too. Ten years from now, if I’m with my daughter, she wants to be with me so much, we play all the time, and we’re a $10 million ARR company, I’d rather live in that reality than in one where we’re a $100 million company and my daughter doesn’t want to play with me, isn’t interested in me, and we’re not that close.
That’s the decision I made. Maybe it’s right, maybe it’s wrong. I have no idea. I go extremely hard. I have a lot of ambition, and that’s not going to stop me or anything. But setting those priorities and figuring out for yourself what works for you is important. More than ever, she’s the most important thing for me, and I prioritize her a lot.
How do you balance that with your ambition? I have this in a smaller way. I don’t have kids, but I have a lot of ambitious things and a family I should spend time with. When I spend time with them, I’m not present. I’m looking at my phone, saying, “Oh, I should be reviewing Astra”—you know, GPT-6. It’s like AGI. It’s not, right? But I really feel like I should be working, and at the same time, I feel guilty. I’m happy in both areas.
Totally. That’s how I feel. I feel like I’m a terrible dad, a terrible husband, a terrible CEO, and like I’m failing in every single area of my life. I feel exactly like that. But every day I try to be a little bit more present.
Especially when you’re working from home, you leave your office, and now you’re a dad, and then you’re not anymore. Now you’ve got to be a husband and pay attention to your wife, while your mind is thinking, “Oh, man, how do I do this enterprise? I’ve got to close this deal.” I feel like I cycle all the time, too.
Yeah, we’re on the same boat.
Yeah.
Well, it’s never easy. I think just naming the thing for your category is important, and people don’t think about category creation enough. This is something Netlify weirdly tried to do and failed with Jamstack. I think more people should try to create a category, because when they’re successful, they own the whole thing.
So for this, Resend is the next generation SendGrid, let’s call it, or something like “AI-native SendGrid.”
Yep.
I think—and I’m probably being too honest—that’s not category creation yet.
Yeah.
But it is category dominance. You own your category. That’s good.
Mm-hmm.
But it’s not clear enough yet what’s different. My classic example is another guest we’ve had, Dharmesh Shah from HubSpot.
Uh-huh.
Anyone who wants to think about category creation should read about inbound marketing versus outbound marketing and how they created the category of inbound. The ideal phrase for category creation is that once you’ve heard the category—the two words your company represents—you don’t want anything else. There’s no long essay, manifesto, or talk. It’s just two words, maybe some examples. That’s it. Your audience can take it from there.
Mm-hmm. It’s very hard. I’ve seen friends trying to do that, like Svix. It’s webhooks as a service.
Yeah. They took my name.
Yeah. It’s just so freaking hard.
I think they’re doing okay, right?
Yeah.
I don’t know if they’re still around. There was a webhooks company, and then there was an anti-webhooks company where you’d say, “We don’t do webhooks. We just help you publish a JSON file that you pull,” or something like that. It kind of reverses the push versus pull data flow. This was, again, the COVID-era dev tools. That was the most exciting thing at the time.
Yeah. Yep. Cool. Well, always a pleasure to catch up, and congrats on all the success. I’m still very curious how much you’ve dominated your category. I think you chose it very well, executed it very well, and you’re managing your family as well, so lots of respect.
Thanks so much. Yeah, it was great chatting with you guys. Thanks a lot.