Replit CEO：为什么 SaaS 末日论有其道理，以及为什么编码模型正在触及平台期 | Amjad Masad
在 Masad 看来，编码模型的性能提升正在接近平台期，这可能重新打开一个 3-6 个月的窗口，让拥有专有数据的公司针对特定用例进行微调。 Replit 在 2023 年实现了优于 GPT-3.5 的编码表现，随后认为与 Sonnet 和 Opus 直接竞争是个糟糕策略，因为后两者正在缩小差距；如今更强的开源模型又让这笔下注重新可行。在企业比选中，哪怕只是短暂领先，也可能决定一笔大合同：“成本低于性能问题”（“Cost is secondary to the performance question.”）。
Replit 的护城河在于“agent 实验室”能力：编排、评估并持续替换模型，而不是对某一家供应商保持忠诚。 Anthropic 是连贯、可长期运行的核心循环主力，Gemini 则在代码搜索子 agent 等任务上提供最佳性价比。基础设施是可弃用的：Agent V2 让 Replit 删除了旧的护栏，而为 Agent 3 构建的自主能力后来原生进入了 Opus 4.6。Masad 本身并不反对中国模型，但对敏感企业数据存在安全顾虑；他认为开源对于避免 AI 形成寡头垄断至关重要。
AI 经济学目前仍更看重获客和能力，而非短期毛利率的纯粹性。 Masad 否认 Replit 每 100 美元收入中有 80 美元流向模型供应商，但承认这一占比很高；他以 Anthropic 为例，称后者每赚 100 美元就有 60 美元付给 Nvidia，同时维持 40% 的利润率。他认同 Jason Lemkin 关于免费推理可以获客的核心判断，并以 Claude Code 和 Codex 赠送 token 为例；Core 则是一种付费免费增值产品，企业版和 Pro 收入可以补贴它。Replit 去年已接近盈利，但随着公司先打造最佳产品、再进入优化阶段，利润率会随之变化。
SaaS 抛售在方向上有其合理性，但冲击是选择性的，并非全面推倒重来。 企业仍把 Salesforce 和 Workday 作为记录系统，并通过 API 和 MCP 构建；但部分客户会跳过应用层，直接在 Databricks 上构建。主持人认为，如果 20-30% 的客户或用户使用 Databricks 作为数据仓库，就可能重创 incumbent 的增长，Masad对此表示认同；垂直点解决方案还面临更大的替代风险，以及微型创业者带来的价格压力。
运营可能是近期更大的 agent 机会，因为它能把碎片化的 SaaS、电子表格和人工流程直接转化为可衡量的节省或销售增长。 Replit 用户正在构建报价配置器、deal desk 自动化和客服工具，其中 75% 的用户不是工程师。Masad 举例称，如果能节省 10,000 美元的 SaaS 支出和 200,000 美元的人力成本，只需额外花 1,000 美元做安全投入就是“不需要思考的选择”，ROI 可以达到“100倍”。
Masad 称，“从实际意义上说，IDE 已经死了”，但可审查的安全关键型代码仍会保留一个持久的细分市场。 AI 已经吞噬了 IntelliSense、自动补全、点击跳转符号等功能；NASA、SpaceX、自动驾驶和飞机软件仍然需要验证。主持人提到 Cursor 据报有 20亿美元收入，Masad 则认为，凭借广阔市场、黏性的企业客户，以及仍想逐行检查代码的开发者，Cursor 即使面对 Twitter 的“讣告”也能持续存在。
AI 不会带来单一的人员规模结果：一些创始人会把效率收益存下来，进取型经营者则会向更大的机会加码招聘。 一位桌游卖家在 Replit 带来收入增长、降低 SaaS 支出后增加了 8 名员工；年收入 6000万美元的 Firecrown Media 也招聘了更多 vibe coder。Replit 自身希望减少支持岗位，增加构建者和销售人员，而销售的工作正转向教育和转型咨询。
最尖锐的公司特定风险来自平台权力：Apple 在 Replit 应用已运行 4 年、此前通过 100 多次审核后，仍将一次更新卡在应用审核中 3 个月。 Masad 反复表示“我不知道为什么”，并指出 Apple 接受用 Replit 制作的应用，且该品类的其他应用也受到影响。他只能推测 Apple 可能正在确立对 vibe coding 的立场，或防范绕过 App Store，而 Replit 并不支持后者。他的应对方式是继续创新，在竞争对手“都会照搬 Replit 的做法”时，保持领先“1、2、3、4、10步”。
1. Replit 的十亿开发者论，最终需要的是创作者而非程序员
Masad 将 Replit 的起点追溯到约旦：15 岁时，他靠一家小生意赚了几百美元，足以请全班去安曼新开的一家 McDonald's。软件看起来可以改变“财富创造、财富分配和创业”，这份信念后来在 Code Academy 工作期间进一步加深。
Replit 于 2016 年上线时，目标不是当时约 2000 万名程序员，而是 10亿名开发者。公司逐步消除了开发环境、托管、包管理、维护、版本控制和多人协作的摩擦，却最终撞上一个顽固瓶颈：“人们不想学编程。”
他在 2025 年 3 月的转向——“我不再认为你应该学编程”——之所以激怒很多人，是因为他此前一直是“学编程”路线的倡导者。独立创始人打造数百万美元规模公司的案例证明了两者的区别：人们需要学会创造和构建，但不一定要学会编程。2021-2022 年的 GPT-3 还不是 agent 化的突破口，真正的突破来自 2024 年能够在长时间跨度上自主行动的 AI。
Masad 回看一家产品能否达到风险投资级别的产品市场匹配，标准不是拥有几个客户，而是产品被用户“从手里抢走”，需求增长速度快于供给能力。创始人可以保留愿景，但必须持续转向，直到遇到这种爆发式需求。
2. 应用护城河是可弃用的多模型 agent 栈
Masad 将产品开发形容为与模型能力“共舞”。Replit 于 2024 年 9 月加入大量基础设施和护栏，2025 年 3 月为 Agent V2 删除其中大部分，随后在 2025 年 9 月围绕 Agent 3 数小时级的自主能力重建。自 Opus 4.6 起，这种自主能力有很大一部分已经内置于模型。
Anthropic 仍是连贯、可长期运行的核心循环主力，但 Gemini 在代码搜索子 agent 等任务上提供了更好的性价比。Replit 2022 年提出的“模型社会”（society of models）如今意味着使用所有供应商；有一段时间，尽管 Anthropic 驱动着核心 agent，Replit 发给 Google 的 token 数反而更多。
这种不成文的评测知识本身就是护城河：Replit 的 AI 工程师会花数天测试新模型的边界，再把专有基准与 A/B 测试结合起来。Masad 称，Replit 用 Gemini 做出的设计可以优于 Google 自家产品，因为 Replit 知道如何评估和使用这个模型。
主持人质疑，为了仅仅 3 个月的领先而训练专有模型，是否是在浪费资源。Replit 于 2023 年在编码上击败 GPT-3.5，随后认为直接与后来的 Sonnet 和 Opus 竞争并非好策略，因为差距正在缩小。更强的开源模型和编码能力可能触及平台期，让微调重新成为选项；模型评估能力与特定用例数据，决定这段暂时领先能否在企业比选中胜出。
对于中国模型，Masad 本身没有道德层面的意见，但认为敏感企业数据存在安全问题，因此 Replit 目前尚未采取这一步，同时保留未来采用的可能。他认为开源对于防止 AI 形成寡头垄断、协同定价，以及避免企业控制 AI 的使用方式至关重要，甚至提出由美国政府支持的国家级开源模型联盟。
3. 在 S 曲线变平之前，性能高于利润率
Masad 的经营规则明确且不留余地：在能力持续提升期间，以牺牲性能为代价优化成本会输。只有当一个领域达到“S 曲线的渐近平台期”后，毛利率才变得重要；他表示 Replit 去年已接近盈利，但利润率会随公司先打造最佳产品、再进入优化阶段而变化。
他否认模型供应商每拿走 Replit 每 1 美元收入中的 80 美分，但承认这项支出仍然很高。他对上游的比较是：Anthropic 每赚 100 美元，就有 60 美元付给 Nvidia；Anthropic 公开披露的利润率为 40%，但仍在补贴服务。
智能的价格已经大幅下降，但单 token 价格并没有像预期那样下降：与 GPT-4 时代相比，人们在新模型上的使用支出高得多，同时获得了更多智能和生产力。Masad 认为，只要真正的前沿基本还在 GPT-5 与 Claude 之间，价格压力就有限；Gemini 和开源模型可能改变这一格局。他认为成本持续存在，部分原因是芯片昂贵且竞争有限，并以 Nvidia 约 80% 的利润率为例。
Masad 认同 Jason Lemkin 关于免费推理可以充当获客手段的核心判断。他认为，2025 年末和 2026 年初 Claude Code 与 Codex 引发的热潮，部分来自反复赠送 token 和放宽速率限制；他把 agent 化开发称为一种潜在健康的“创造性成瘾”（creative addiction）。留存仍是一个未解问题。
4. Agent 首先冲击掌握工作流的运营部门
Masad 预计，产品组织将纳入“产品构建者”，让技术、设计和产品取向的人共同决定下一步要构建什么。工程师仍会负责基础设施、AI/ML、嵌入式系统和其他底层工作。
运营是尚未被充分服务的切入口：团队被困在彼此隔离的 SaaS 数据、失败的自动化、Excel 表格和人工流程之间。Replit 客户正在构建销售报价配置器、自动化 deal desk，并重做客服运营，由此带来的效率和收入提升，可能与缩短产品开发周期相当，甚至更高。
Salesforce 和 Workday 并没有被整体替换；企业使用它们的 API，Replit 则创建 MCP 并接入 HubSpot 等平台。但部分 Databricks 客户会绕过应用软件，直接在数据仓库上构建。当主持人认为，如果 20-30% 的客户或用户把 Databricks 作为数据仓库，就可能显著损害上市 SaaS 公司的增长时，Masad 表示认同。
垂直 SaaS 面临更大风险：Masad 认为调查软件正在被整体替代，而微型创业者则在 Replit 上推出更便宜的点解决方案。被问及 Wix 时，他只表示其广告投放行为看起来像是在“把一切都押向 Base44”，而不是说 Wix 已经真的变成了 Base44。
5. 维护与安全把 token 支出变成护城河
Replit 对每次改动都进行代码审查，并称维护软件消耗的 token 与创建软件一样多。开启高级功能后，agent 会写代码、启动浏览器、测试应用、审查改动，再把失败结果回传给编码 agent。用户喜欢这种直截了当的审查员：“这看起来就是 AI 生成的垃圾。”
生产环境中的安全 agent 已经在监控企业部署、软件包、活动以及供应链攻击。Masad 的递归式判断是：“AI 制造的任何问题，都有更多 AI 可以用来解决。”
分层经济学决定定价。工程师有更多替代选择，也更在意价格；而一名节省了 10,000 美元 SaaS 支出和 200,000 美元人力成本的运营经理，很容易再花 1,000 美元购买安全能力。新创业者则需要 $20–$30 的入门价位，因此 Core 是一种付费免费增值形态，企业版和 Pro 收入可以越来越多地补贴它。
6. IDE 广泛而言已经过时，但在安全边界上仍不可或缺
主持人将 Twitter 声称 Cursor 已死的判断，与其据报 20亿美元收入放在一起比较。Masad 的回答是，软件生成并不存在一个既有的 TAM，市场规模可能与互联网相当，而且它是在增强劳动力，而不只是替代劳动力。Cursor 还拥有黏性的企业客户，以及希望逐行检查代码的开发者；要被替代，它必须先落后到很远，或许要像 GitHub Copilot 那样。
真正死掉的是 IDE 过去的价值主张：IntelliSense、自动补全、代码导航和“代码智能”已经被 AI 吞噬。但负责验证自动驾驶、NASA、SpaceX、飞机或其他生死攸关系统的工程师，仍会继续使用类似 IDE 的环境。
Masad 以 JavaScript 作类比：飞机软件没有采用这种无类型、易出错的灵活性，而 Web 开发者接受它，是因为 Gmail 宕机不会导致人员死亡。同样，vibe coding 会先在风险较低的软件中扩散。计算机科学对于真正有内在驱动力的学生，以及算法和数据结构等持久基础仍然重要，但不再是拿到高薪 Google 工作的自动通道。
7. 公司规模是选择，平台依赖不是
AI 带来的效率会引出相反的人员规模决策。Jason Lemkin 偏好极度精简的运营，而一位桌游卖家则利用更高的收入和更低的 SaaS 成本招聘了 8 人。年收入 6000万美元的 Firecrown Media 在营销自动化成功后，也招聘了更多能够进行 vibe coding 的人。
Replit 希望减少支持岗位，增加构建者和销售人员。销售正变成教育和转型咨询，这可能让 10,000-15,000 美元的合同在有限人工陪伴下变得可行。Masad 过去 12 个月最大的观念变化，是放弃过度追求销售效率：如今的需求足以支撑“能招多少销售就招多少销售”。
Masad 表示，尽管 Replit 应用自 2022 年起一直在运行，并通过了 100 多次审核，Apple 仍将一次更新卡在应用审核中 3 个月。他指出 Apple 接受用 Replit 构建的应用，且该品类的其他应用也受到影响，因此自己不知道 Apple 的动机。Apple 可能正在确立对 vibe coding 的立场，或防范绕过 App Store，但 Masad 表示 Replit 并不支持这种做法。他把这场冲突视为 Replit 已经变得举足轻重的证据。
We’re approaching a certain plateau in how good coding models could get. The cost question is secondary to the performance question. When you focus on cost, you reach a certain asymptotic plateau in the S-curve. I think, for all intents and purposes, IDEs are dead. The name of the game is just staying 1, 2, 3, 4, 10 steps ahead.
Amjad, dude, I’m so excited for this. I’ve wanted to make this one happen for a while, so thank you so much for joining me once again today.
My pleasure. Great to be here.
We were just chatting about the desert Replit’s been through and the amazing position you’re in today. Did it take a long time for the world to see what Replit was, or did technology take a long time to catch up to your vision of what it could be?
1. Replit Makes Building Accessible
It’s more the latter. The core insight that I had, even before we started the company—maybe at this point 20 years ago—was that software is much more transformative than we actually think it is. Mark Andreessen wrote, “Software is eating the world,” and all of that, but I thought it was transformative for people’s lives and for the prospects of wealth creation, wealth distribution, entrepreneurship, all of that.
It’s my story: the impact it had on my life as a kid growing up in Jordan, coming to the US, and now running a multibillion-dollar company. I’d seen that even earlier than that. When I was 15, I built a little business. I was making hundreds of dollars, and that was an amazing amount of money for me. I took my entire class to McDonald’s when it opened up in Amman, and that was the core insight.
Initially, when I came to the US, I worked at Code Academy, and the goal was, “Let’s teach everyone how to code. Let’s make it as simple as possible.” I started seeing the stories back then, too. If a fitness guy learns a little bit of coding, launches an app on the App Store, and makes $1 million, it’s like, “Wow, this is amazing.”
When we started Replit in 2016, the goal was: How do you make programming more accessible? How do you get to a point where there are 1 billion developers, not just 20 million developers at the time? We started solving one problem at a time: the development environment, the hosting environment, package management, maintenance, iteration, version management, multiplayer.
But there’s one bottleneck, and that is that people don’t want to learn how to code. It took me a long time to really accept that because I became the learn-to-code guy. In March 2025, when I was in TPBN and said, “I no longer think you should learn how to code,” it went super viral. People were pissed. “How could you say that?”
It’s just the realization that there are people who are now successful—we were talking about Jason Lemkin from SaaStr earlier—who are building multimillion-dollar businesses solo with no developers. They don’t need to learn how to code. They need to learn how to create; they need to learn how to build.
The real unlock wasn’t just AI. It was AI that could take actions over a long horizon. We’ve had AI since 2021 and 2022 with GPT-3, but the unlock in 2024 was agentic AI. That was the first glimpse of it, and we had to build a lot of infrastructure around it to make it work.
2. Building Beyond The Models
Dick Coleman, how much of the magic is what you build versus the performance of models beneath you?
It is a dance, I would say. Think about Elon Musk when he did self-driving 1.0, before version 13 or whatever, when it became end-to-end. They had to write a lot of software. They had to write a lot of classical computer-vision-type software in order to make it work, but then they went to end-to-end learning.
There’s this thing that happens in AI, and as a founder, it’s really important to understand this. At any given point, you have to plug in a lot of holes in order to build the most advanced thing that you can build. In September 2024, we had to build a lot of infrastructure and guardrails in order to make agents work. In March 2025, when we released Agent V2, we had to delete a lot of that code because the models got a lot better at staying consistent.
But then you upgrade your vision. In September 2025, we released Agent 3, and that was the most autonomous agent on the market. It was the first to run for hours on end. Again, we had to write a lot of software in order to make it stay on track.
Now, since Opus 4.6, autonomy is built into the model. You have to understand what the model capability is and how much infrastructure you have to build in order to make it perform. You have to stay ahead of that in order for your product to stay the most innovative on the market.
When you look at your model usage today, I’ve heard you say before, when I was listening to your prior shows, that you have a preference for Anthropic. What does the model usage look like across different providers today?
Anthropic has been the workhorse for over a year. It’s the core agent loop because it can run coherently for a long time. But a few things have changed. Google’s Gemini models are the best at price-performance, for example. Given their price, where do they sit on the Pareto frontier?
For tasks like code search, we might create a subagent that is cheaper and has good-enough performance, and we offload that from the main core loop. I wrote this thesis back in 2022. I called it “The Society of Models.” Now we use models from every provider. At some point, we were sending more tokens to Google than we were sending to Anthropic, despite Anthropic being the core workhorse.
There’s this concept of agent labs. We talk about AI labs, but there are agent labs—you know, us, Cursor, and some of these other companies.
Our goal is to start with the user problem: what are we trying to fix, what are we trying to build, and then walk back to the technology and use whatever model we need to use. In some cases, we build our own models.
That was going to be my question. Amjad, the reason why the show's become more and more successful is because I bluntly ask more and more direct questions. Cursor decided to build its own model. When we look at other players across different verticals, it's always been a mistake to build your own model. Was Cursor's decision to build its own model, and your decision to, in some cases, build your own model, a mistake, because equivalent model performance has just always come alongside you?
It ebbs and flows. Again, AI is such a changing landscape. The answer will change every 3 to 6 months.
When you're investing your team and your resources in that, you have to weigh whether it's worth that 3-month advantage, no?
Right. You do, and that's why it makes it so freaking hard and why you have to change your mind all the time, right? Because there's a lead-up time, but the most important thing is optionality. So, in 2023, when we were training models, we achieved better coding performance than the state-of-the-art models at the time, GPT-3.5, right? But then, since Sonnet came out, and later Opus, the gap has closed by a lot, and they were spending tens of billions of dollars, if not hundreds of billions of dollars, making agents work. That would've been a dumb strategy for us to go and try to compete on that.
But now I would say the opportunity has opened up again for other reasons. The open-source models are getting really good, and we're approaching a certain plateau in how good coding models could get. So you can use your data to fine-tune a model specifically for your use case. I don't know if you saw, but Intercom talked about its new model that is better at customer support than the frontier models. Maybe their model's going to be state of the art for 3 to 6 months, and maybe 6 months from now the models will zoom back ahead.
How do you analyze that? Is it better to be 3 months ahead and spend the money? How important is it to be ahead for 3 months?
It's very important. It's a matter of closing a large enterprise deal. We're constantly getting baked off against everything under the sun, right? We're winning most of our enterprise deals because our product is consistently ahead of the market.
The thing that I always hear is that we're using frontier models to basically set benchmarks. We're seeing where those benchmarks lie, and then we're switching to open source to get as close to them as possible with much more efficiency in terms of cost. Is that the future?
It depends, right? The cost question is secondary to the performance question, especially at a time when it's flush with capital. When you focus on cost is when you reach a certain asymptotic plateau in the S-curve. You don't foresee a massive improvement specifically in your domain.
Intercom might say, “We don't predict that models are going to get that much better at customer support for the foreseeable future. Therefore, we can focus on building our own, or there's a data flywheel that we can get, or there's a cost advantage that we get.” But if you focus on cost at the expense of performance, you're going to lose. It's similar to any era in tech, right? Whether it's cloud, mobile, or SaaS, there are moments in time where the goal is growth, performance, and being at the edge. Then, when things rationalize—Uber and Lyft, when things rationalize—you focus on your gross margins.
I think there's this idea, in all transparency, that your Replits, your Base44s, and your Lovables—and you might kill me for putting you in that bucket—but is it bluntly that, of every $100 that you make, $80 goes to the model providers?
For us, it's not 80. It's way less than that, but it is a significant portion, like anyone. Even Anthropic: every $100 they make, $60 goes to Nvidia. Their margins are public. It's 40%, right? They're also massively subsidizing their service.
Can I be blunt? What are your margins, and how have they changed over time?
They've changed quite a bit. I said last year that we were close to profitability. In programming, there's an old saying: “Premature optimization is the root of all evil.” As a programmer, if you think about performance and optimization early on, you're going to make the wrong decisions. So the reason margins keep moving around is that we're trying to build the best product. After we build the best product possible, we look around and say, “Okay, there are so many ways we can optimize.” Then we go into an optimization period.
When we think about price sensitivity, how do you think about essentially routing different behaviors, where some models are actually okay for certain things and some need the frontier? How do you think about intelligent model selection for different functions?
I would say that is the core competency and IP of an agent lab. If you think of yourself as an agent lab, you have this tacit knowledge of first evaluating models. I think about our AI engineers as psychologists in many ways. When a model comes out, the first thing they do is sit down with it for a day or two. They play around with it, plug it in, and say, “Okay, what are the limits? What can it do?”
That's more like the tacit aspect of it, which is very easy to underestimate. The reason Replit is able to build state-of-the-art performance when a new model comes out, sometimes even better than the lab itself, is that we're big partners with Google, and Gemini is one of the best models at design. I would say our products are better at design using Gemini than Google's products. It's because we know how to evaluate these models and get the best performance out of them.
Then we have a bunch of proprietary benchmarks that we use, and finally, we do a lot of A/B testing as well. That's really what you're doing as a company that is building on top of foundation models.
How concerned should we be about the proliferation of Chinese models in Silicon Valley companies?
I'm not sure about this question. Is their intention to actually innovate and play in the market fairly? Is their intention to destabilize? Is their intention to destroy the value of U.S. AI labs? I don't know. There are reasons to be careful, and there are reasons to be optimistic about the progress.
Would you have a moral issue with using a Chinese model?
I don't think there's a moral issue per se. I don't think they're using slave labor or anything like that. I would have a security issue, especially since we have enterprise customers that depend on us for sensitive data and things like that. So we haven't taken that step yet, but I wouldn't preclude taking it in the future.
I would love to see a U.S. corporation investing in open source. It looks like Nvidia is making moves in that regard. But open source is going to be very important for us to actually have a free market around AI, because if we end up in an oligopoly of AI companies, there's actually an economic theory of how they'll naturally collude on price, and prices will not go down as fast as possible.
They'll also control how we use these AIs. They'll also not provide everything through the API and keep some of the models for themselves. I think it would be bad if we're in a situation where AGI or AI is only controlled by a few corporations. Open source is going to be very, very important. I would venture to say that maybe the U.S. government should start a consortium of companies that are creating the best open-source national model, so that the market stays competitive.
We mentioned Jason Lemkin earlier. He said something brilliant to me, which is, “Inference is the new sales and marketing.” Obviously, free is a large part of your business and many others in the space. Do you agree with Jason that inference is the new sales and marketing, and how do we think about that?
If you think about the hype period that we got with Claude Code and Codex toward the end of 2025 and early 2026, a lot of it was driven by how many free tokens they were giving out. Every other day, they were saying, “Oh, it's 50% more tokens, or fewer rate limits,” and things like that. So it's clear that companies are using inference as a way to loop people in, and I think agentic development is addictive.
I think it's a better kind of addiction than social media or passive consumption. It's creative addiction, which is kind of good. Companies are realizing that a way to rope people in is through a lot of free tokens. Now, there is a question about retention, but I don't disagree fundamentally with the statement that you can use free tokens to acquire users.
3. Product Teams Become Builders
Enterprises will try things in certain respects. I was chatting to Jason Lemkin before, and he said, “Your number one ICP is product teams.” As product teams code or vibe more, what do they do okay in 2 to 3 years? How do they think about those functions?
There's a big question about how product teams will look in the future. If I were to make a prediction, I would say that we'll still have engineers inside the organizations.
Those engineers are responsible for more infrastructure, AI, ML, embedded systems, and low-level engineering. Then you have product organizations, and product organizations will have people who tilt a little more technical, people who tilt a little more toward design, and people who tilt a little more toward product, but you're not really calling them anything different. They're product builders, and their responsibility is to figure out what to build next.
I agree with Jason that a lot of the ICP right now is product, but one thing I'm really excited about is operations teams, and they're kind of underserved. Operations teams are sitting at the nexus of a lot of data flow. Typically, they'll buy a lot of SaaS software. They're typically not happy with it because there are all these SaaS tools siloing the data.
They try a lot of automation software that doesn't work very well. They have a lot of Excel sheets and a lot of manual work. We see a lot of our customers building quote configurators for their sales team, automating their deal desk, and automating support operations. When they use Replit, the return on investment is as good as, if not greater than, it is for product teams.
With product teams, you're cutting down the product development lifecycle, which is hugely valuable. With operations teams, you're actually being more efficient. You're selling more, reducing headcount, or needing fewer people.
4. The SaaS Apocalypse Is Justified
You said ops have a lot of tools that they're not happy with. That has led to the SaaS apocalypse and a lot of people believing in a deterioration of value in a lot of these public companies. Is that exaggerated, or is it just cause for concern that they've lost the market caps they have?
I'll tell you what we're seeing in enterprise. We're not seeing people rip out Salesforce or Workday, or really fundamental—I think the hype term is “systems of record”—type SaaS tools. Instead, they're building on their APIs, and we're creating MCPs and hooks into the HubSpots of the world, Salesforce, and all of that.
But there's also another side where we have this great partnership with Databricks, and people are skipping the SaaS tools entirely and building on top of their data warehouse. There's another view on this where the system of record is actually your data warehouse, and this is a way to be bullish on Databricks and companies in that space.
Well, that goes to a statement that Jason has made, though, which is that it is justified, because what you're seeing is the maiming of SaaS company growth. If you have 20–30% of customers or users using Databricks as their warehouse, and that maims that audience for the corp like SaaS public companies, that's enough to cause growth to decline significantly.
I would agree with that. A lot of vertical SaaS is in trouble as well. Maybe you don't think of it as a system of record; there's a lot of survey SaaS software, and we see that getting replaced wholesale with Replit.
On the other side, there are a lot of micro-entrepreneurs starting on Replit today, and they're undercutting the price of a lot of SaaS companies out there, especially vertical point solutions. There's another pricing pressure there as well.
What about the traditional incumbents in our space? I'm thinking of Squarespace and Wix. What happens to them?
If you look at the behavior of Wix, it looks like they're pouring everything into Base44.
Is Wix now Base44, basically?
I don't know much about their company, but it looks like they're spending a lot on ads, definitely behind Base44.
What did you think of the Super Bowl?
I haven't heard that many people talk about it, but I think it's good that people are getting vibe coding out there and more people are talking about it. I don't think it's in the public consciousness yet.
5. Maintenance Makes Agents Useful
How do you think about maintenance in this case? You have ops teams building tools. You have entrepreneurs building tools. You have to maintain these fuckers. It's hard enough running a business. I have to maintain this now?
This is where Replit shines. If you talk to Jason or some of our other customers, Replit goes way further than any other vibe-coding product in creating more maintainable software. Part of the reason Replit has been slightly more expensive than others is that we do a code review for every code change that we make. We spend as many tokens on maintenance as we spend on creating the software.
Replit also has a built-in tester. If you enable all the power features, whenever the agent writes code, it goes into a testing phase, spins up a browser, and tests everything in the app. It goes into a code review session, reviews that, kicks it back to the coding agent, and gives it feedback: “The test failed here. The code review is not good.”
People enjoy looking at the code review agent because it's kind of a dick. It will say, “This looks like AI-generated slop.” It will actually say that. Then it goes back.
We're also building agents that are sitting in production software. We already have security agents sitting in enterprise deployments, monitoring activity, monitoring packages, and monitoring for supply-chain attacks. The thing about AI is that any problem AI creates, there's more AI that you can build to solve that problem.
Have you been surprised by how price-sensitive people are around security and code reviews?
Not in our segment. I would say in the engineering segment—we have some engineers who use Replit—but 75% are non-engineers. Engineers are more price-sensitive because they have a lot of options. They can use a lot of different products on the market.
When you're an operations manager using Replit and you just saved $10,000 on SaaS software, you've saved another $200,000 on headcount, and you're spending an additional $1,000 just to make sure that the software is more secure, that's a no-brainer. The ROI has been a hundredfold for companies we work with.
On the consumer side, there's more price sensitivity, especially if I'm an entrepreneur just dipping my toes in, which is why we reduced the price on our Core plan. I think there's going to be different models for different use cases or different parts of your journey. If you're just starting out, you don't want to be hit with a $1,000 bill. You want to be able to play around with $20 or $30 before you commit.
Is Core a loss leader? Are you saying, “I'm willing to lose money on Core for the conversion to Pro”?
Yeah, it's sort of become a bit of the new freemium, because you can't totally run an amazing free tier when tokens are still very, very expensive. You want people to pay something to recoup some of the losses. But as a greater percentage of revenue comes from enterprise and Pro, we're willing to subsidize the cost of Core a little more.
Token costs today: how do they change over the next 3 years? Do they get 50% cheaper or 10 times cheaper?
It's an interesting question because there are different ways to slice it. You can look at the price of intelligence. The price of intelligence has gone down tremendously. It's really hard to quantify, but everyone's spending way more on opens than they were spending on GPT-4, and they're getting way more intelligence and way more productivity out of that.
Pure unit price is not going down as much as people expected it to, and I think part of it is that we're living in a world where the true frontier of models is actually just between GPT-5 and Claude, so there's not a lot of pricing pressure. Now we're seeing Gemini catch up, and we're seeing open-source models catch up, so maybe as there becomes more pricing pressure, we'll see token unit prices go down.
I also don't think these companies are all that profitable, partly because the underlying chips are not all that cheap, and there isn't a lot of competition. At the bottom, everyone's running on NVIDIA, and there's not much competition there. NVIDIA has amazing margins, especially for a hardware company. What is it, 80%?
Yeah, it's pretty phenomenal. It's why it's the world's most valuable company. I'm confused about something. I'm confused about Cursor. Can you help me? This is Jason's question, so you can blame him. All the hard questions—just blame Jason.
Twitter says Cursor is completely dead. I don't have a single portfolio company that uses Cursor, but they just hit $2 billion in revenue. Is Cursor dead, or is this narrative completely bullshit?
I'll tell you a few principles. One is that the market is so large. The market for software is not just the market for SaaS, although that's part of it. It is an expanding market.
There's no existing TAM. You can say, “This is the TAM we're going after.” Some people will say—I think Sequoia at some point said it's the labor TAM, right? The knowledge-worker labor TAM. That could be it, but it's not like we're displacing labor. We're actually supercharging labor, so companies are becoming more productive.
It's an ever-expanding TAM, and it's going to be perhaps as big as the internet, if not more. I'm talking about just software generation broadly.
It’s vibe coding, AI-powered coding, fully agentic coding generation in general. I think it’s a hugely valuable market, and the world is large, and a lot of different companies will prefer different products. There are people at Replit today who still prefer to review every piece of code, and something like Cursor is good for that because it’s still sitting in an IDE. So there’s a portion of the market that still really likes that.
The other thing is, I think Cursor has done well at enterprise sales, and enterprise is a very sticky customer. Once they adopt something, it has to really fall tremendously behind, perhaps like GitHub Copilot, in order for people to replace it. I don’t think Cursor fell that far behind, right? They still use the latest agents, and they have a good agent harness.
I think Twitter is a distortion machine, and Twitter is like the inside of inside baseball. It is the people at the edge of AI adoption. I think if you’re constantly on Twitter—which is why I think VCs don’t have the best information landscape, because a lot of them just live on Twitter—the world is much larger than that.
6. IDEs Are Dead
Are IDEs dead? Will we have IDEs in 2 years?
I think, for all intents and purposes, IDEs are dead. I think they’ll limp along because, again, some engineers just love that control. But there’s no future in them in that no one’s going to be asking for the latest feature of IntelliSense or whatever. One part of IDEs was code intelligence. We call it intelligence, but it wasn’t very intelligent.
The autocomplete, the click-to-symbol—all of that stuff is irrelevant. So, in that sense, IDEs are dead because AI has eaten all of that. But in terms of people who want to see the code, I think there’s still a population of users who want that. I think there are still engineers working with software that they want to actually verify works.
For example, life-or-death software: if I’m writing mission-critical software for self-driving cars or for a NASA or SpaceX mission, I think there’s always going to be a need for some kind of IDE.
I’d be pretty pissed if the autopilot on my plane had problems because of vibe coding with no IDE.
Yeah, that would be pretty sad. Even in planes, those kinds of software never adopted JavaScript, and the reason they never adopted JavaScript is because JavaScript is the original sort of vibe language. JavaScript didn’t have types, so you could run into errors.
But the reason why we adopted JavaScript is because web software was not going to kill anyone. If someone pushes a bug to Gmail, your Gmail isn’t going to die. Maybe it’ll be down for an hour, and that’s fine. So there are different risk appetites, and I think vibe coding follows the same thing, where it’s on the less risky type of software.
So if you’re a student listening to this, should you not study engineering at university? Should you not study computer science? How does this inform how you think about advising young people?
Before 2005, let’s say—which is when I went to school—people who went into computer science were very intrinsically motivated to understand computer science and exactly how computers work. They were hackers and really interested in programming.
After that, computer science became a hyped-up field because it was the easiest place to make money. We had the boot camps, and we had this whole thing, and computer science departments exploded because of that. Now, if you’re not into computer science, if you don’t feel like you’re drawn to it like a fly drawn to a light, then don’t go into it because someone told you you’re going to make a boatload of money working for Google. That’s gone.
It’s pretty dumb to tell people to go into computer science if they’re not really intrinsically interested in it. If you’re interested in it, I think there are still ways to contribute. You could get into ML and AI and go work at the big labs or a company like ours.
But if you’re thinking about university, the curriculums are not able to move at the pace of model progression. What would you advise me as a student?
I will say that the field of computer science, where you’re learning about data structures and algorithms, isn’t going to change. There’s always going to be a need for people to understand the underpinnings of computer science because we still need those people.
And do you think university is the best place to learn that?
It depends on the person. I think there are people who are autodidacts. I would consider myself someone who is very good at teaching myself, so it depends on you. If you’re really good, you can learn on the job, open the textbooks, and have the discipline to do that, you don’t have to go to a university.
But if you’re someone who likes the structure, who likes to meet other students and work with them, and likes the discipline that the university forces on you, then I think there’s still a place for a university.
Are companies going to be so much smaller in the future when you look at the capabilities of individual people? Do you buy the ideological Silicon Valley view that we’re just going to do more and be so much more capable? Or are you actually saying, “No, we will have dramatically smaller engineering teams”?
I see both. Jason is someone who wants to work with a very lean team. He’s doing better than when he had people on staff. Yesterday, I met an entrepreneur at a conference in DC who’s using Replit. I think he sells board games online, and he says, “It’s been so transformative for my business. We’ve saved so much money on SaaS. We’re selling more, so I decided to use the increased revenue and efficiency to hire more people.” He hired 8 more people.
There’s a customer case study we actually published on our site: Firecrown Media, a $60 million media company that owns magazines and different properties. They’ve been so successful using Replit for marketing automations and all sorts of things like that, they decided to hire more people who know how to do vibe coding in order to sell more, do more, and build more.
So it depends really on the kind of company. We see companies that are like, “We want to get leaner, and we want fewer people.” I think it’ll come down to the entrepreneur, the level of ambition, and how they want to run their company.
For us at Replit, I think what we want to try to eliminate or reduce is a lot of supporting roles. We want builders, right? We want builders, and we want salespeople because I think salespeople are still important. People really want to talk to someone to sell them software and teach them how to use it.
The sales role is also changing in that salespeople are becoming more like educators and transformation consultants. They go into companies, tell them how to use the product, and explain the best way to leverage this technology.
With the removal of other parts, or with fewer engineers, or with the removal of other parts of the organization and reduced headcount costs in those areas, are you able to justify sales reps moving down ACV categories or ranges because you don’t have the cost elsewhere in the business?
There are a lot of companies that now have self-serve enterprise, especially in AI, because there’s so much demand that you can justify that. But traditional SMBs still need some hand-holding, and typically the cost of sale is higher because you have so much supporting staff.
If you’re willing to spend an hour or 2 talking to a customer and onboarding them, and perhaps a few hours for the rest of the year supporting them, and the contract size is $10,000 or $15,000, I think you’re right. It’s possible that becomes more of a thing.
What worries you today? Are you positive about the future? Are you negative? And not just about software development—about life. I know it sounds weird and macro.
7. Apple Blocks Replit
Actually, let me tell you about Replit first and what worries me there. We’ve gone through this: first they laugh at you, and then they—I forgot the quote exactly—but we’re at the point where then they attack you, right?
It’s now been publicly reported that Apple is sort of blocking the Replit app. Replit has been on the App Store since 2022, doing exactly the same thing: allowing people to generate or write code and run it in a browser window. Suddenly, they’re saying that we’re not complying with their guidelines, and we’ve been stuck in app review for now 3 months. We haven’t been able to push an update.
Again, we’ve been on there for 4 years. We’ve passed 100 Apple App Store reviews, and suddenly Apple has decided that they don’t want this category. Other apps in the category are also getting affected.
I’m sorry. What is the reasoning behind that? They think that your apps are not as high quality, that they’re AI slop, and that they’re making discovery harder in the App Store?
They’re just telling us we’re not in compliance with their guidelines, and we’ve shown them repeatedly—
But peeling that back one layer more, it’s like they’re—
They're fearful that it's going to—
I don't know because they're accepting apps made on Replit. They're accepting them into the App Store. I tweeted about one yesterday. The acceptance rate is very high. If they think that what we're generating is slop, they could have said that. Second, they wouldn't have accepted the apps that are being made with Replit. So it doesn't seem like—
So then what is the reason, Amir? You're a smart dude. I'm a hopefully relatively smart dude. There is a reason behind that.
I want to give them the benefit of the doubt, and I want to say that perhaps they're trying to figure out what their posture is here, and perhaps the delay is just because they're trying to figure that out.
Posture with regard to what? How they—
With regard to vibe coding.
How many?
With regard to vibe coding and how people are going to be using that on the App Store. Look, maybe they're concerned that people are going to be circumventing the App Store rules. That's not what we do. Maybe there are other apps in the category that do that, and they looked at the category as a whole and were like, “Okay, let's put a pause on this.” But we're not doing that.
We are just making it possible for people to make apps. If their concern—and maybe that's their concern—is that people are going to be making apps and getting around the App Store, that is not what we do.
How detrimental is that to your business? When you think about Apple putting up that additional barrier or challenge, is that like, “Fuck,” or is that like, “Eh, we'll get over it. We've faced many before”?
I've been building this company for a decade, and the fact that we're such an important force in the world, in the culture, and in how the world is reacting to this technology is a good thing. It means that we are achieving our mission. We're important.
I never thought it was going to be easy, and I'm ready for the challenges to come. I actually like challenges. Life is much more interesting when there's a little bit of fight in it.
I think it's important to have a common enemy in a team. Amjad, you can shoot one: Claude Code, Lovable, or Base44. Which one do you shoot?
I don't really think that way. We—
If it's not a name, I'm going to tell Jason. The next word is a name, Amjad.
I'll tell you that they're all going to copy what Replit does. Replit is setting the roadmap for everyone. That's fine by me, but what I really like is innovation—building the next thing that can really change how people think about software.
I feel very proud that this agentic revolution was really kick-started by Replit Agents in 2024. The name of the game is staying 1, 2, 3, 4, 10 steps ahead.
Right, dude. We're going to do a quick-fire round. I'll say a short statement, and you give me your immediate thoughts. Does that sound okay?
Yep.
What have you changed your mind on most in the last 12 months?
How fast to scale our sales organization. I always wanted to build a hyper-efficient company, and Replit is extremely efficient for our size and valuation. But given how much interest we have in the market, I think we should hire as many salespeople as we can to go sell to these people.
Who would you most like to have on the board who you don't have on it yet?
I would like to have an operator who's 2 decades ahead of me or something like that—a CEO who's been in the weeds building a company from scratch, but went public and reached way past where I've gotten. We have a lot of VCs, and they're amazing, but no one who's actually an operator.
Tell me, when I say “high performance,” who do you think of first, and why them?
It's such a cliché, but obviously Elon. The reason is the amount of things that you can do at the same time—the number of things. Before that, Steve Jobs running Pixar and Apple. I'm dead by the end of the week running 1 company. How can you do 2? That is an enigma to me. How can you do 5?
What do you invest in for yourself that has been a game changer? I hope it's not too personal, but you've lost a lot of weight.
Yeah.
You look fantastic, dude.
Yeah.
Tell me about that.
Losing the weight has actually been 3 years. I always yo-yoed my weight depending on my stress level at the company, and I decided in 2023 that I'm going to take 3 years to get healthy.
Instead of sprinting and working out 5 days a week and then burning out, I'm going to work out 1 day a week. I started working out 1 day a week, and then my energy levels improved. I lost some weight. Okay, now I'm going to add another day a week. Now I'm going to add another habit. Let's add a walk. Let's add a bit of stretching.
It's so slow. I want to get healthy over the next 5 years, and that's been a major game changer. I'll just add another habit every once in a while. It's very simple stuff. Let's add a walk. Let's add a sauna once a week.
I really love sauna and cold—hot and cold. It's like a reset. It forces you—I'm someone who's constantly thinking, and I tried meditation and all that stuff, and it's really hard for me. When I go into a sauna and get burnt and then go into ice-cold water, you can't think. Your mind freezes, and that forces me into a meditative state for the next hour or so.
What is the biggest piece of advice to a new parent, given all you know now about parenting?
Don't stress out. I think it's the difference between the first kid and the second kid, and every parent will tell you, let alone with the third kid. The first kid is like, “Oh, it's nap time. We have to go do the nap thing. We have to do exactly that and measure this and that,” and you're measuring everything. You're doing everything.
Babies are quite resilient. Humans have been around for 100,000 years. We know instinctively how to raise a child, and I think taking it a little easy on yourself is probably the best advice.
Final one for you. What do you know now that you wish you'd known when you started Replit?
The feeling of product-market fit. You can deceive yourself into thinking you had product-market fit at different points. You'll get a few customers: “Oh, that's product-market fit.”
Real product-market fit—and people will say it, but you can't really internalize it—is the idea that the product is getting pulled out of your hand. You can't even provide it fast enough. I think if I understood that earlier on, perhaps I would have searched for it faster, harder, or things like that.
As an entrepreneur wanting to build not a lifestyle business or small business, but a venture-scale business, you have to find that moment. You have to keep pivoting and changing. Perhaps you don't change your vision, but keep searching until you find that explosive demand.
Dude, I've so enjoyed this. I so appreciate you letting me go wayward, but you've been fantastic, so thank you so much for joining me.
Appreciate it, man. Thank you.