Replit Raises $400 Million to Shift Target From Software Developers

Replit融资4亿美元，将目标客群从软件开发者转移 未来的公司将只需要销售人员以及在高抽象层级操作的构建者，因为自主智能体将消除传统的编程岗位。 在本次讨论中，Replit首席执行官Amjad Masad探讨了该平台如何转型为AI原生实时编程环境，专为泛技术创作者与非传统开发者而非标准程序员打造。他详细介绍了用户如何利用Replit构建多元化的个人、创业及企业应用，并深入阐述了推动Agent 4多智能体架构开发的战略转变。展望未来，Masad描绘了一个由“后提示词”时代、专用的计算机操作测试智能体，以及围绕调度智能体的通才型创业者对未来商业组织进行全面重构所构成的图景。 Replit 以 90 亿美元估值完成 4 亿美元 D 轮融资，致力于让具备基础读写能力的用户都能通过自然语言直接构建、托管和部署真实且可扩展的软件。 (0:00) Replit 于 2023 年进行战略转型，放弃迎合传统开发者，转而全力服务产品经理、设计师以及垂直领域专家等新型“AI 原生”创作者。 (2:08) 随着 Agent 4 的发布，Replit 引入了画布设计交互与多智能体并行架构，使用户能够在后台智能体异步构建与跨平台部署的同时进行实时构思与规划。 (23:11) CEO Amjad Masad 预测未来企业核心仅剩“构建者”与“销售人员”两种角色，普通员工都将转型为指导智能体执行业务的通才型创业者。 (34:48)

According to Replit CEO Amjad Masad, the rapid rise of AI coding agents is shifting software creation from specialized engineering to broad human intent, leaving future companies staffed primarily by abstract builders and educator-evangelist salespeople. Following Replit's recent $400 million Series D at a $9 billion valuation, the company is orienting its platform toward non-technical creators who direct parallel AI agents rather than write manual code. Pivoting Toward Non-Technical and AI-Native Creators (2:08) Masad observed that developer tools historically accumulated excessive accidental complexity. While Replit originally aimed to empower traditional programmers, the company discovered that conventional software engineers often prefer manually configuring their environments. Consequently, in 2023, Replit made an explicit strategic pivot to target tech-adjacent professionals—such as product managers, designers, and entrepreneurs—who possessed domain expertise but were blocked by technical barriers. To eliminate the coding hurdle entirely, Replit transitioned in September 2024 into a "live coding" platform powered by natural language AI agents. Masad states that basic reading and writing literacy is now the only skill requirement to build, host, and deploy scalable applications, moving the user experience away from syntax toward visual and conversational interfaces. Real-World Applications Across Personal and Enterprise Domains (5:44) The abstraction of software engineering has enabled non-technical creators to build specialized applications across consumer, entrepreneurial, and enterprise domains. For example, a physical therapist and her husband avoided costly offshore agencies by using Replit to build a health-tech app featuring 3D body rendering and range-of-motion tracking. Other users have created tools ranging from family chore trackers to modern software replacements for legacy MS-DOS sports club systems. In enterprise environments, fitness technology company Whoop increased its product testing capacity tenfold, expanding from evaluating 5 out of 100 concepts to testing 50. Organizations also deploy Replit to build custom internal tooling and RevOps automations that consolidate data from Salesforce, HubSpot, and Gong. Replit supports these workflows through product-led growth, client hackathons, and Model Context Protocol (MCP) integrations. Parallel Agents and Asynchronous AI Workflows (23:11) Replit aligns its product releases with six-month AI model breakthroughs. While earlier releases like Agent 3 introduced long-horizon autonomy—enabling background containers to execute extended coding tasks—they often forced users into passive downtime while waiting for results. With Agent 4, Replit introduced a multi-agent architecture and parallel agents paired with a visual canvas, allowing users to design, chat, and iterate asynchronously while background agents handle code execution. Agent 4 also addresses collaborative engineering by automatically forking virtual machines when teammates join a session, using an orchestrator to subdivide tasks. The platform supports cross-platform contexts, enabling users to generate and deploy web applications and mobile apps to TestFlight or Android simultaneously from a single workspace. Unresolved Limits in Computer Use and Continual Learning (31:52) Despite rapid progress in agentic software generation, Masad notes two critical technological bottlenecks: reliable computer use models and continual learning. Computer use models continue to develop slowly because visual video feeds are significantly more difficult to compress into high-dimensional space than language. While coding agents act as a partial workaround by scripting spreadsheets or querying APIs directly, automated agents still struggle with "taste"—the qualitative assessment of whether an application's user interface is genuinely well designed. Furthermore, true on-the-job continual learning remains an unsolved industry problem. Currently, platforms must rely on static workarounds, such as having agents record acquired skills into flat markdown files like `skill.md`, rather than having agents autonomously and permanently adapt to an enterprise's bespoke environment. The Two Enduring Roles in Future Companies (34:48) As abstractions continue to absorb engineering tasks, Masad predicts that future corporate organizations will consolidate primarily into two human roles: builders and salespeople. Sales roles will remain defensible because human trust is essential; these professionals will function primarily as educators and evangelists guiding client transformation rather than delivering traditional pitches. Meanwhile, builders will shift from writing code to directing autonomous agents toward broad commercial goals. Masad envisions a corporate model where employees act as generalist entrepreneurs identifying organizational friction and deputizing agents to resolve it. Replit tests this internally through a "vibe coding in residency" team given open-ended operational mandates; the team has built custom Zendesk queue visualizations that raised customer satisfaction scores and deployed an internal HR platform to streamline employee onboarding. Masad anticipates a "post-prompting" future where founders simply assign high-level business goals to autonomous software agents. However, realizing fully autonomous enterprises remains limited by current model deficits in contextual adaptation, environmental learning, and qualitative design evaluation, ensuring that human oversight and domain expertise remain essential.

Replit 联合创始人兼首席执行官 Amjad Masad 认为，随着 AI 智能体将底层编码与系统配置彻底抽象化，未来企业的组织形态将被重塑为主要由“构建者”与“销售人员”组成，员工将普遍转型为调度智能体解决业务问题的通才型创业者。 消除编码门槛与聚焦非传统开发者 (0:00) Masad 回顾指出，传统开发工具伴随偶发复杂性而变得愈发繁琐。Replit 在走过十年历程并完成 4 亿美元 D 轮融资（估值 90 亿美元）后，系统性解决了开发环境配置与云端部署难题，并于 2024 年 9 月转向“实时编程”（Live Coding），利用后台编码智能体全面抽象代码，使用户仅凭基本读写能力即可构建可扩展应用。在 2023 年，Replit 实施了明确的战略转型，不再重点服务偏好自行配置环境的传统程序员，而是将目标转向产品经理、设计师和创业者等“AI 原生开发者”，践行其“创造 10 亿新开发者”的使命。 随着 Agent 4 引入拖放与标注等画布式交互，开发流程进一步向多模态协作演进。Masad 强调，平台真正的高价值用户是因工程瓶颈受阻的跨领域人员。通过完全封装底层代码，非技术人员得以从琐碎的配置中解放出来，将注意力集中在产品构想与业务逻辑的实现上。 赋能领域专家与产品驱动的增长路径 (5:44) 根据 Masad 的观察，Replit 上的软件构建涵盖个人应用、企业级工具和创业产品三大类。由于消除了外包沟通损耗，领域专家能够直接主导复杂系统开发，例如一对物理治疗师夫妇在平台上独立构建了包含身体扫描、运动幅度追踪和 3D 身体渲染的医疗科技应用；其他案例还包括取代旧版 MS-DOS 系统的体育俱乐部软件，以及个人用于罕见病护理和家务管理的轻量应用。在企业内部，智能穿戴公司 Whoop 将产品测试比例从 5% 提高至 50%；企业亦利用平台自建报价配置（CPQ）与营收运营工具，大幅削减了孤立 SaaS 的采购开支。 在商业化拓展方面，Replit 采取了类似 Stripe 的产品驱动增长（PLG）模式，并辅以销售团队的内部布道。员工常在周末尝试个人项目，体验到技术赋能后便将工具带入工作场所。Replit 进而协助企业内部支持者举办黑客松，并通过安全与合规保障促成顶层采购。此外，生态中还涌现出新型服务商，如冰岛某代理机构通过“氛围编码”（Vibe Coding）使客户交付成本降低了 60% 至 70%。 多智能体架构与目标驱动的自主工作流 (23:11) Replit 将产品路线与 AI 模型的阶跃式提升深度绑定。从 2024 年中 Claude 3.5 Sonnet 显著提升代码生成能力，到 Agent 3 支持后台数小时的长时程自主构建，平台持续重写后端以承载长时间运行的容器。针对等待智能体运行带来的停顿，Agent 4 转向多智能体并行架构：用户可以在视觉画布上进行异步设计与规划，后台智能体则同步执行编码与合并冲突处理；当团队协作加入时，系统会自动分叉虚拟机并分配任务，实现网页与移动端（如 TestFlight）在统一上下文中的同步部署。 Masad 预测行业正步入“后提示词时代”（Post-Prompting World），用户将无须细致编写提示词，而是向智能体下达高维度的商业目标（如优化获客漏斗或自动化构建运营 SaaS）。在这种范式下，核心竞争力不再是提示词工程，而是对技术可能性的敏感度、探索心态以及持续交付的敏捷度，创作者需要像多产的独立开发者一样快速迭代以应对产品生命周期的更迭。 未来组织构想：构建者与教育型销售 (34:48) 针对 AI 代理全面渗透后的企业形态，Masad 提出未来公司将主要由“构建者”与“销售人员”主导。销售岗位的本质将转向组织变革咨询与教育引导，因为客户始终需要基于人际信任的学习支持；而构建者的角色则向更高层级的抽象迁移，从早期的手工计算、操作计算机、编写代码，演进为指挥智能体系统的总指挥。企业通常仍需要能够统揽市场、客户需求和技术能力的商业通才来确立抽象愿景。 在这一演变中，员工将普遍具备“创始人思维”，专注于识别业务问题并指派智能体予以解决。Replit 内部推行的“氛围编码驻留”（Vibe Coding in Residency）项目印证了这一模式：该团队带着“巡视公司并使其变得更好”的开放式任务，先后为支持团队搭建了工单优先级可视化工具，直接提升了客户满意度（CSAT），并为人力资源团队自建了集中式入职与福利管理平台，展示了通才型创业者借助 AI 赋能的组织潜力。 Masad 强调当前技术仍面临显著瓶颈：计算机视觉操作模型（Computer Use）在理解交互界面与体验“品味”上依然受限，目前主要通过编码智能体编写脚本或调用 API 进行变通规避；同时，代理在组织内部自主积累经验的“持续学习”（Continual Learning）尚未实现，目前仍依赖写入技能文件等临时手段。未来高度自主化企业的全面落地，依然取决于这些底层能力的突破。

[2:08] Amjad Masad: create a billion new developers [2:08] Amjad Masad: like the pain [2:08] Amjad Masad: fire in them [2:08] Amjad Masad: AI-native developers [11:02] Amjad Masad: neurological shift [11:02] Amjad Masad: vibe coding [16:06] Amjad Masad: create different variations of this [23:11] Amjad Masad: Viper coding revolution [28:04] Amjad Masad: post-prompting world [28:04] Amjad Masad: optimize my marketing funnel [28:04] Amjad Masad: every day, build me a SaaS company, try to market it, and make me some revenue [28:04] Amjad Masad: an explosive thing [34:08] Amjad Masad: learning on the job [34:48] Andrew Miklas: agents all the way down [34:48] Amjad Masad: builders and sales people [37:17] Amjad Masad: almost everyone is a founder [37:17] Amjad Masad: Go around the company, make it better

[0:00] [music] [0:04] >> Today I'm joined by Amjad Masad, the CEO [0:07] and co-founder of Replit. Replit is the [0:10] leading no-code app builder uh for [0:13] consumers and enterprise. They just [0:15] recently raised their Series D, a $400 [0:18] million raise at a $9 billion valuation. [0:21] Amjad, thank you very much for joining [0:23] me and I'm excited to learn more about [0:25] Replit. [0:25] >> Thank you for having me. [0:26] >> Of course. Why don't you tell us a [0:28] little bit about uh what Replit is? [0:30] >> Yeah, so um you know, [0:33] our ambition is that [0:36] anyone, no matter what level of skill [0:39] they have, anyone who can read and [0:40] write, basically, that's the skill that [0:42] that you need, [0:43] can come in with an idea [0:45] and can leave with a an app that's [0:49] deployed, that's hosted, that's getting [0:52] traffic, that's can scale, and they [0:54] don't have to worry about any technical [0:57] aspect of the [0:59] uh of building that thing. [1:01] Um [1:02] and it's been it's been a mission we're [1:04] on for for 10 years right now. [1:06] So, initially we solved the development [1:08] environment. You know, that was the [1:09] hardest thing is like setting up the [1:10] development environment. Then solve the [1:12] dev- deployment environment. Then what [1:15] was left is like the coding part was [1:16] still very intimidating for people. And [1:19] September 2024, [1:21] Replit became the first like what's [1:22] called live coding product where uh we [1:25] abstracted away code entirely. So, [1:27] there's like a coding agent behind the [1:29] scene, but you're just interfacing with [1:31] AI using natural language. And more [1:34] recently with Agent 4, [1:36] uh also design interactions. So, you can [1:38] write things, you can comment on things, [1:40] you can [1:41] uh drag and drop things in a like a [1:43] canvas. And we're thinking about a lot [1:45] of different modalities for how people [1:46] want to interact with agents because I [1:48] don't think it's always going to be just [1:49] text. I think at some point it's going [1:51] to be multimodal, maybe video, maybe [1:52] audio. [1:54] But we want to create a natural place [1:56] where people can express their ideas. [1:58] And those ideas can turn almost [2:00] magically into software. That is real [2:02] software. That is it's not like a toy [2:03] software. Real software, secure [2:05] software, scalable software. [2:08] >> One of the things that's most [2:09] interesting to me about about your [2:10] company is that it really bridges the [2:12] the gap between a dev tool and and and [2:14] not. I I think it's the first [2:17] dev tool I've seen but that's not [2:19] marketed towards engineers. Like can you [2:21] tell me more about how you came to that [2:24] uh decision that you were going to do [2:25] that and and and who is your primary [2:27] user today? [2:27] >> Yeah, so [2:29] you know, I I started coding at a very [2:31] very young age, but um I was always [2:34] interested in in the act of creation. I [2:37] was interested in entrepreneurship. I [2:38] built like my first business when I was [2:40] 13, 14. [2:41] And I always thought that the developer [2:43] tools were getting in the way. [2:46] It actually gotten worse over time. So, [2:49] you know, as I started you know, coding [2:50] on basic and you just start the basic [2:52] command line interpreter and you can [2:53] just like type a little of basic and [2:54] that's good. [2:55] >> Yeah. [2:55] >> By the time I graduated from college, [2:57] like setting up, you know, a web app was [3:01] like a nightmare, right? Um and so it [3:04] created this desire to just like build [3:07] tools that are [3:08] that are more joyful, more enriching, [3:12] kind of focus on the act of creation as [3:14] opposed to [3:15] the accidental complexity of of of [3:18] developer tools. And so uh I started [3:21] building a lot of tools for myself. Uh [3:23] and eventually I you know, I built what [3:26] would become the kind of the first [3:28] in-browser IDE. Initially, it was like [3:30] an open source project. Later on, I [3:33] would also worked on React and React [3:35] Native. And I approached every time I I [3:38] built a developer tool, I've approached [3:39] it with the same thing. Can you apply [3:42] design sensibilities in the same way [3:44] that we you would work on a a consumer [3:46] app? [3:47] Uh and that's been successful in in many [3:51] ways. [3:52] But ultimately, when we started Replit, [3:53] the goal was make programming [3:55] accessible. Later, we updated our [3:56] mission to create a billion billion new [3:58] developers. [4:00] As we progress in our in our mission and [4:03] in solving every part of the software [4:05] development life cycle, [4:07] what we've noticed is a lot of [4:09] developers actually like the pain. [4:12] They they like setting they they you [4:15] know setting up things and they like [4:17] configuring every aspect of it. It's [4:20] It's sort of like There's no knock [4:22] against that, but like it's sort of like [4:23] a craftsperson kind of liking to build [4:26] their own tools. [4:27] And what we've noticed is the people [4:28] that are getting the most value out of [4:30] our product tend to be the more [4:31] tech-adjacent ones. Maybe people who [4:33] product managers have written code many [4:35] years ago, but don't want to [4:37] worry about the development environment [4:38] setup, don't want to worry about the [4:40] deployment setup. [4:42] Um then designers. Designers who have [4:45] ideas, but are often [4:47] blocked or bottlenecked by engineers. [4:50] And they want to be able to build their [4:52] own ideas. [4:53] Um and later on as we layered on AI and [4:56] the product got better, entrepreneurs. [4:58] And when when I talk to these people, [5:00] it um [5:02] it reminded me of myself. Those are [5:04] people with with ideas, with passion, [5:06] with fire in them, [5:08] but they but they but they're getting [5:09] hamstrung by by just like the need to [5:11] learn to become technical. [5:14] So, at some point in 2023, [5:18] um we just made it an explicit goal of [5:20] like we're not going after developers. [5:22] They're still developers. Like if you [5:24] walk around Replit today, it looks like [5:26] a dev tool company. We're like building [5:28] for creators, right? [5:30] But they're not the traditional type of [5:32] developers. They said like a new [5:33] generation of developers that are coming [5:35] up right now because of AI. They're [5:37] AI-native developers that are creating [5:40] software without having to worry about [5:42] every component in the system. [5:44] >> It's so interesting hearing you say this [5:46] because it it sort of reminds me I [5:48] learned to program on VB6 [5:50] and [5:51] >> [clears throat] [5:51] >> and that's [5:52] to me that's a real [5:54] you know this idea of the tools opening [5:57] up and making making it possible for [5:59] people that couldn't code I couldn't [6:00] code I learned to code in VB6 and and I [6:03] see sort of a very similar thing that [6:05] you're exploring with this. [6:06] >> Yeah VB6 was better than setting up [6:10] React and Webpack. [6:11] >> 100% [laughter] I remember the [6:13] transition yes. [6:14] >> It's it's it got worse over time which [6:17] is like not many things get worse in [6:18] life like programming got worse and I [6:20] wanted to bring it back [6:22] make programming great again. [6:24] >> [laughter] [6:26] >> Yeah that's so interesting. Tell tell me [6:29] about what people are building today in [6:31] Replit. [6:31] >> Yeah I mean there's a few different [6:34] categories there's personal software [6:36] there's enterprise software and then [6:38] there's like entrepreneurs building [6:39] building products. [6:40] One of the apps that I really liked [6:43] recently I was talking to [6:45] a physical therapist [6:47] and her husband and she is physical [6:51] therapist that's [6:52] built up so much deep knowledge in a [6:56] subsection of physical therapy that's [6:57] like around fascia and fascia release [7:00] and she has all these different [7:01] methodologies and ways of tracking [7:03] the progress for her clients and she [7:06] wanted to build an app that is very [7:08] sophisticated [7:10] like being able to take scans of your [7:12] body being able to to scan your range of [7:14] motion and to track that on a 3D [7:17] representation of your body and they [7:19] spent hundreds of thousands of dollars [7:22] offshoring to developers around the [7:24] world and eventually they kind of were [7:27] frustrated by the entire process took [7:29] matters into their own hands built it on [7:31] Replit and I when I saw that app it was [7:32] like one of the best [7:34] like health tech apps I've ever seen and [7:36] and they're not [7:37] >> Yeah because the domain experts can now [7:39] actually build the products. [7:40] >> Yeah people who are closest to the [7:42] problem can build up the the the [7:44] products they need. You know, I talked [7:47] to a founder the other day that's [7:48] building [7:49] a SaaS solution for [7:52] um [7:53] for people who maintain pools. Like and [7:56] he grew up in a in a household where [7:59] their where their family business is is [8:02] a pool business. And so it's like [8:05] there's so much software to build in the [8:06] world. Also met a founder yesterday [8:10] who's building uh software for sports [8:13] clubs. And he was showing me pictures of [8:15] the software they're using today, which [8:17] is MS-DOS-based software. [8:19] >> Oh, wow. [8:20] >> And so here in Silicon Valley, we look [8:22] around us and we're like, "Oh, well, [8:24] what else is left to build?" [8:25] But but there's so many walks of life, [8:27] there's so many [8:29] um things that are kind of a blind spot [8:31] for us. And so suddenly when [8:33] anyone can make software, a lot of parts [8:36] of the economy is just going to improve [8:38] and that's like I think a beautiful [8:39] thing and a lot of wealth creation is [8:42] going to happen, productivity gain as [8:43] well. [8:44] Personal software is also really cool. [8:46] Like, you know, talked to a lot of [8:47] families that built like you know, [8:50] healthcare software. Like there's like a [8:51] mom in Korea that built software for [8:54] that helped manage a very rare condition [8:56] for her kid. Um [8:59] a lot of people build like personal like [9:02] you know, healthcare software like [9:04] tracking [9:06] physical activity or ingesting data from [9:08] all their wearables and and doing [9:11] something with it. [9:12] A lot of software for families. Like a [9:14] mom built like a chore hero software on [9:16] iPad that's like on the wall that shows [9:19] the kids how they're ranking of their [9:21] chores. [9:23] And then there's the enterprise use [9:24] case. [9:25] And in enterprise, it tends to be two [9:27] different use cases. One is [9:29] product development. So companies want [9:32] to move a lot faster in product [9:33] development. Everyone now is feeling the [9:35] pressure of of AI. We need to move [9:37] faster, faster, faster. And they're [9:39] realizing that it doesn't fall just on [9:41] their engineers. Their product people [9:43] can now build software. Their designers [9:45] can build software. We hear from clients [9:49] like one of my favorite stories like [9:50] Whoop is a is a client of ours and they [9:53] told us like the amount of ideas they [9:55] can try has grown by an order of [9:57] magnitude. Right? Like they used to, you [10:00] know, get get like 100 ideas but are [10:02] able to only try five of them. [10:04] But now they can try to 50, right? [10:07] Um so [10:09] you know, companies become a lot more [10:10] prolific. They can release a lot new [10:12] features, create new new business lines, [10:14] new products. [10:16] And then there's like internal tools and [10:18] you know, line of business applications. [10:21] Um there's a lot of sales automations [10:23] that are that that is happening at these [10:24] company. Any any role inside a company [10:28] that's dealing with a lot of data flow, [10:29] for example, think of RevOps. They're at [10:31] the nexus of a lot of different data [10:32] flow from from their CRM, from their [10:35] data people, from Gong, from [10:37] And they want to be able to do things [10:38] with this data. And typically they're [10:40] kind of bottlenecked by either [10:41] engineering resources or the different [10:44] SaaS tools that they would want to buy. [10:45] The problem with SaaS is you onboard a [10:47] new SaaS tool that creates a new silo of [10:48] data that you can't really program. So [10:51] now they're taking matters into their [10:52] own hands and building things like, you [10:54] know, quote configurators and [10:57] and being able to save the company [10:59] hundreds of thousands of dollars if not [11:00] millions of dollars on SaaS tools and [11:01] things like that. [11:02] >> How how are you finding these users? [11:05] Like how do you market to them? How do [11:07] you get them to to try it out? And then [11:09] I I assume for enterprises it's got to [11:10] be a very complicated sales cycle like, [11:13] you know, getting getting into the [11:14] company. Like how does all that work? [11:16] >> Well, the cool thing about the the world [11:18] we're in right now, it is very similar [11:19] to developers. You know, I think the [11:21] insight that a lot of YC dev tool [11:22] companies has had like, you know, [11:25] Stripe, you know, PagerDuty, [11:28] you you uh [11:31] Developers were empowered to be able to [11:33] make decisions to bring software into [11:35] the company at some point like in the [11:37] last couple decades, right? Um [11:40] that that shift is happening outside of [11:42] development right now where the product [11:44] manager, the designer, the operations [11:45] manager is empowered to bring software. [11:48] So, [11:49] the consumer use case, the personal use [11:51] case is often overlapping with the work [11:54] use case in that [11:56] people are playing with these tools on [11:57] the weekend. They're building their own [11:59] personal apps. And they're like, "Oh, [12:01] but the moment you understand that you [12:03] can solve a problem with code, [12:06] it changes your mind like almost there's [12:08] like a neurological shift where you [12:10] start looking at the world differently. [12:12] Where you go around and you're like, [12:14] "Oh, I can solve this. I can fix this." [12:16] >> I remember that when I learned to [12:17] program. [12:17] >> Yeah, exactly. [laughter] [12:18] And now that shift is happening in a [12:19] huge parts of the population. [12:22] And so, it's you know, PLG play I think [12:25] still the gold standard. Just make a [12:27] product that's really good that people [12:28] want to recommend their to their [12:29] friends. And make it easy to refer [12:32] others, you know, build a referral [12:34] program, do all of that stuff. [12:36] And then on the sales side, a lot of it [12:38] is very kind of sales assisted. So, [12:41] someone brings it to work, we talk to [12:42] them. [12:43] They're the champion, we help them it's [12:45] like, "Okay, what do you need to [12:46] convince your boss in order to bring [12:47] this to work? Let's work with them to do [12:49] a hackathon to bring create more [12:51] champions inside the company. [12:53] Uh let's work with you let's teach your [12:55] leadership about AI, what So, a lot of [12:58] what our sales people are doing is [13:00] evangelism, is education. So, it's it's [13:03] a different kind of sales motion I think [13:05] than the previous one. We still have [13:06] enterprise sales that's more top-down [13:09] where a company comes to us we're like, [13:11] "Hey, we're trying all the different [13:12] live coding tools. We heard it's good [13:13] for our business. Can you help us kind [13:15] of evaluate Replit?" We go into that and [13:17] um and there we're often winning because [13:20] Replit has built um [13:22] uh a history of just like being uh super [13:25] trusted on security, on compliance, all [13:28] that basic enterprise stuff you still [13:30] you still have to do. [13:31] >> Let's talk about sort of what what the [13:34] limits are today of what you can do with [13:37] a product like Replit. Like what what [13:39] kind of systems can you build in Replit [13:41] and what do you still need to have sort [13:43] of traditional [13:44] uh software engineering approaches to? [13:48] >> I can confidently say to any [13:50] entrepreneur out there that you can [13:51] build a SaaS product, a consumer [13:54] product, um like an automation product [13:58] on Replit comfortably. [13:59] >> Mhm. [14:00] >> If you want to build like a new car [14:02] platform or you want to build like a new [14:04] machine learning system, that's not [14:07] exactly what we're focused on today. [14:08] Some people still figure it out. Replit [14:10] is a versatile tool. We give you a [14:11] virtual machine. We give you a general [14:12] purpose agent. If you have some [14:14] technical knowledge, you can bring that [14:16] and you can build sophisticated things. [14:18] But if you want to build things entirely [14:20] by coded without worrying about the [14:21] technical details, we have so many [14:23] examples on our website and what we talk [14:26] about [14:26] in our marketing. [14:28] Uh we you know that there's so many [14:30] success stories now that I can [14:32] confidently tell an entrepreneur with no [14:33] technical knowledge that they can build [14:36] >> Right. [14:36] >> uh a piece of software. So there's a lot [14:38] of consumer apps, there are a lot of [14:41] there there are a lot of [14:43] you know vertical SaaS products. There [14:45] are people that are starting [14:47] like Replit native agencies. I was [14:48] talking to [14:50] uh someone from Iceland yesterday and he [14:52] was telling me they're getting so much [14:54] business because they're 60 to 70% [14:57] cheaper and more effective than [14:59] traditional agencies and they're all [15:01] vibe coding on Replit. So they they get [15:03] a client that that client wants an [15:05] application. Not everyone knows that [15:07] they can build the application [15:08] themselves. So they're going in it's [15:10] probably a period of time and they're [15:12] able to make a lot of money that that [15:14] way. [15:15] You know, the internal tools, [15:17] automations, those tend to be kind of [15:19] fairly simple uh [15:20] programs like you're building an [15:22] internal org chart, you're building, [15:26] um, a CPQ system, like a quote [15:28] configurator, where it's like pulling [15:30] from HubSpot or Salesforce. We have a [15:33] lot of MCP and integrations that allows, [15:35] uh, to do that. Um, [15:37] >> And how how do those integrations work? [15:39] Like have you gone through and built [15:40] integrations to all Okay. [15:42] >> Yes, yes. We spent a lot of times [15:43] partnering with companies, building [15:45] those integrations, and now there's the [15:48] what's happening with the skills call [15:50] revolution, where you know, companies [15:53] are putting out skills and MCPs, and we [15:56] just vet them and integrating into into [15:58] the product. And so as you're talking to [16:00] a Replit and you're like, [16:02] "I want to integrate Stripe." And so we [16:04] already built a set of skills and [16:06] sometimes code and things like that. [16:07] We'll search a database, we'll bring all [16:09] that in, and now it's in contacts and [16:10] now the agent is like, It's sort of like [16:13] Neo in the Matrix, where like you you [16:15] download a new skill and you're like, [16:16] "Oh, I know how to fly a helicopter [16:17] now." So it's so cool to see the agent [16:19] is like, "Oh, let me search this set of [16:21] skills that I have." And suddenly it [16:23] knows how to do that. But we spend a lot [16:25] of time just making sure they're secure [16:26] and they're safe. [16:27] >> That makes sense, yeah. Having built a [16:29] dev tool community, like how how is that [16:31] different? How do you build the sort of [16:33] like groundswell of support to pull [16:36] yourself into those organizations? Is it [16:37] different than how it how you would have [16:39] done it traditionally? [16:41] >> It's it's a little different in that you [16:44] need to show what's possible. I think [16:48] with with developers, with more [16:50] traditional, you know, CS trained [16:53] developers, [16:54] they know what's possible. They [16:56] read your docs, they'll figure out [16:58] what's possible, they're on Hacker News [17:00] all the time, and they'll generally kind [17:02] of [17:03] are much more resourceful in knowing [17:05] what's possible. With Replit, there's a [17:07] lot of education that needs to be done. [17:09] So we have like a devrel, [17:13] uh, team, but they're not the [17:14] traditional devrel. They're they're more [17:17] like educators. [17:18] >> Yeah. [17:18] >> They're like, you know, going and [17:19] spending a lot of times just teaching [17:21] people what's what's possible. Our [17:23] documentation, I think, is a lot simpler [17:25] than the your typical dev tool [17:26] documentation. So, you have to be really [17:29] good on content. You know, Stripe got, [17:31] you know, popular for this you know, [17:33] being really good on documentation. [17:35] With if you're building a dev tools for [17:38] non-developers, you have to go above and [17:41] beyond on content. You have to produce a [17:42] lot of video content as well. [17:45] Um also, the agent itself needs to be [17:48] imbued with characteristics that is able [17:50] to talk to non-developers and tell them [17:53] what's possible and be a brainstorming [17:54] partner. Also, this is why we also moved [17:57] from purely just prompting to also [17:59] uh a canvas and more visual interface so [18:02] that people can explore things uh more [18:05] easily and we have buttons where it says [18:07] like, create different variations of [18:09] this. And so, we're adding a lot of [18:11] different shortcuts just to show people [18:12] what's possible. On the enterprise side, [18:14] when we go and we talk to leaders, we're [18:17] like, reserve any judgment. [18:20] Don't pay for Replit. We're going to [18:21] come in, bring the group that's most [18:23] excited about AI, and we're going to do [18:24] a hackathon. [18:25] >> How do you when you see someone, how are [18:26] you like, yes, this is going to be the [18:27] person who's going to be so excited to [18:29] use this? [18:29] >> We think a lot about that because you [18:32] know, typically when you think about [18:34] ICPs, you're thinking about, oh, that's [18:37] someone who's went to school for this [18:39] degree or someone who's a product [18:41] manager or something. [18:42] Which is still true, you can do that. [18:45] But oftentimes because we're able to [18:47] sell the sales department marketing [18:48] product design, [18:51] the set of traits that we think makes [18:53] that like the champion, the person who's [18:54] most excited about Replit, who's going [18:55] to spread it inside the organization, [18:57] who's going to build, who's going to do [18:59] the education evangelism, they tend to [19:01] be very entrepreneurial. [19:02] >> Yeah. [19:03] >> They're the kind of person that could [19:04] start their own company. [19:06] But they're able to kind of be very [19:08] influential internally. They're kind of [19:10] like what YC talks about, like PG&E [19:13] essays, like resourceful. Someone who's [19:15] not going to get blocked, who's going to [19:17] figure out what other AI tool I need to [19:18] integrate, what can I like go learn or [19:21] to figure it out. So that [19:22] entrepreneurial kind of founder mindset [19:24] I think is very important. [19:26] >> Got it. [19:26] I mean, while we're on the topic of YC, [19:28] you know, I'd love to hear about your [19:29] your experience here. Like how how did [19:32] YC influence, you know, what how Repl it [19:34] started and what it became? [19:36] >> The main realization from YC is how much [19:39] you can get done in 3 months. So when we [19:41] when we came in and Sam stood [19:45] when he was still running it, Sam stood [19:46] in front of the batch and said, [19:49] "For the next 3 months, tell your [19:50] friends you're going to be missing. [19:52] You're not going to be able to help them [19:53] move. You're not going to be able to [19:55] You'll come back into their lives later, [19:57] but for the next 3 months, you need to [19:59] be hyper focused on this company and [20:01] you're going to be able to achieve great [20:02] things if you really go intensely into [20:04] it." When we did YC, we had the we had a [20:07] whiteboard and a countdown to to demo [20:10] day. [20:12] Uh and the number of like a just a very [20:14] simple list of things we need to [20:16] achieve. And every day we wake up, we [20:18] erase that number and we change it to [20:20] the countdown. Um and and when Repl it [20:24] got got into YC, it was still a Repl, [20:26] that's the name, right? It was still [20:27] just like a command line with like you [20:29] just type a little bit of code and run [20:30] it. [20:31] >> Yeah, yeah. [20:31] >> We exited YC, we already had like um [20:34] web development, we had like the initial [20:36] aspect of hosting, we had code [20:38] intelligence, we had like so many sort [20:40] of IDE features is like got a lot more [20:42] feature complete just in 3 months. [20:45] And that that is an empowering thing, [20:47] sort of what we talked about with [20:48] programming. [20:50] Uh [20:50] just the fact that like, "Oh, you can be [20:52] so intensely focused and you can work so [20:54] hard and achieve so much in 3 months." [20:57] And now [20:58] every agent release, [21:00] there is it's not 3 months, it's more [21:02] like 4 weeks where we bring everyone cuz [21:06] some people are still kind of remote or [21:07] other offices, bring everyone to the [21:09] office. We provide breakfast, lunch, and [21:12] dinner, coffee 24/7. And we're like, [21:14] we're just going to hit this like very [21:16] very [21:17] ambitious goal. And that's like the YC [21:20] mentality. Another thing is the compound [21:23] growth. [21:24] Especially when you're spinning up a new [21:26] product or a new initiative, [21:27] this idea of like, I don't know where PG [21:30] came up with the number seven, but like [21:31] 7% week-over-week growth [21:33] >> Mhm. [21:34] >> is a like a very good way to bootstrap a [21:36] new new product line. So, we're [21:38] constantly doing sort of we're [21:40] constantly going back to the YC basics [21:43] and doing them and all all of that we [21:44] learned here. [21:45] >> Tell me about how how did YC change your [21:48] life? [21:49] >> There are a few ways. One is before we [21:52] got into YC, we couldn't raise all that [21:54] much money. We raised like maybe [21:55] $500,000. [21:57] And you know, VCs did not want to meet [22:00] us. It was like [22:01] the doors were not really open for us. [22:04] Uh after YC, you immediately, first of [22:07] all, Demo Day, you get to introduce a [22:08] lot of VCs. [22:10] Um and then we were lucky because [22:14] we got rejected from YC four three or [22:15] four times, and then we got invited to [22:18] do YC because Paul and Sam saw us on [22:20] Hacker News. [22:21] >> Yeah. [22:22] >> Uh and so we got a direct relationship [22:24] with Paul and Sam. And then at the end I [22:26] was like, I want an intro to Marc [22:27] Andreessen. [22:28] And you can still ask the partners here. [22:30] Like they'll they'll they'll try a lot [22:32] to to get you intros. [22:34] But I was like old enough to ask for [22:35] that. And so they got me an intro to [22:37] him. I went and had breakfast at his [22:38] house and pitched Repl.it, and A16Z [22:41] ended up leading our seed round. [22:42] >> Nice. [22:43] >> Uh and it and [22:45] it like it my network just expanded [22:47] tremendously after getting into YC. [22:50] And I don't think we would have been [22:52] as successful without YC. Perhaps we [22:54] would have quit if we couldn't fundraise [22:56] or so it gave it gave our our company [23:00] life. [23:01] >> Yeah. Yeah, for us it was very similar. [23:03] We were complete outsiders and I think [23:05] having having, you know, somebody sort [23:07] of welcome you in made made an enormous [23:09] difference. [23:09] >> Yeah, 100%. [23:11] >> So, you guys just announced Agent 4. So, [23:13] what is that like the fourth major [23:14] revision of your of your product? [23:17] >> Yeah, so we kind of an act of trying to [23:19] predict the future and base in what [23:22] we've seen at the time starting in kind [23:24] of 2024, we thought that [23:26] broadly [23:28] the AI capabilities have massive step [23:32] changes uh twice a year. [23:34] >> Mhm. [23:35] >> Uh so, if you think about 2025 [23:38] um there was the like Viper coding [23:40] revolution that happened in earlier 2025 [23:43] and then there was the the you know, the [23:45] autonomous revolution that happened in [23:48] late 2025 with Opus 4.6 that led to open [23:51] claw and and things like that. Uh but I [23:53] can tell you the same thing happened in [23:54] 2024, mid-2024 Claude came out and um [23:59] Claude Sonnet for the first time we [24:00] could generate like a lot of code as [24:02] opposed to like GPT-3.5 which has that [24:04] that laziness component. Um and then [24:08] late uh 2024 we started seeing initial [24:11] signs of the labs kind of um [24:14] doing long horizon reasoning. And so, [24:18] so that observation we're like, okay, we [24:20] want to align our road map to AI [24:22] capabilities. Um and so, every 6 months [24:26] we release a new uh a new agent version. [24:29] Um [24:31] and it's an act of [24:33] predicting what's possible. It's also [24:34] pushing the edge of what's possible. So, [24:36] for example, Agent 3 was the most [24:38] autonomous agent on the market. We We [24:40] knew that autonomy was coming [24:42] and we're like, okay, we want to be able [24:44] to run the agent for like [24:46] 2 3 4 hours. People should be able to [24:48] put put in a big prompt, go to lunch, [24:49] come back and and see the software fully [24:51] fully built. And so, okay, what do we [24:54] need to do to update our platform? For [24:56] example, like we had we had to rewrite [24:58] everything on the back end to have these [25:00] long-running containers in the [25:01] background doing work while the user is [25:03] not in front of the computer. [25:05] And so we did all of that work and [25:08] although autonomy didn't arrive until [25:09] like maybe [25:11] like like true autonomy until like [25:14] uh [25:16] November and and December, Replit Agent [25:18] had demonstrated where the world is [25:20] headed by by September. [25:21] >> Right. [25:22] >> With Agent 4, [25:25] >> there were a few things. One, parallel [25:26] agents. We thought that was finally [25:28] potentially possible to do parallel [25:30] agents. [25:31] >> The the the The thing that sucks about [25:33] autonomy is that you can put in a big [25:35] prompt and you're kind of sitting back [25:37] and just like watching it work. Like [25:39] what do you do next? [25:40] >> Mhm. [25:41] >> Okay, so that question, what do you do [25:43] next? Should be able to design. Should [25:45] be able to kick off other type of work. [25:48] Um should be able to chat with your [25:51] agent and and plan for other things. And [25:54] so we wanted a more asynchronous nature [25:57] to the product. [25:59] Uh so we started building towards a [26:01] multi sort of agent architecture and [26:03] parallel agent architecture. We had to [26:05] solve merge conflicts and so many things [26:07] to make that work. [26:09] Also, as as the agent is building, we [26:11] wanted also unblock you from designing. [26:14] So we also had a kind of a more [26:15] asynchronous design agent and we're [26:17] like, "Okay, what what is the best [26:18] interface for that?" So we designed this [26:20] canvas. [26:21] >> Yeah. [26:22] >> So now we have a built-in design [26:24] capability inside Replit to um [26:27] to be able to explore the next page [26:29] you're going to build or the next [26:30] feature we're going to build while the [26:32] agent's building. And then once you're [26:33] ready, you'll kick that off into another [26:35] thread [26:36] >> seeing. [26:36] >> and it starts. So people are now sitting [26:38] in front of Replit and just experiencing [26:40] the state of flow. [26:41] >> Yeah, yeah. [26:41] >> Cuz agents are slow, that's fine, but [26:43] they can work in the background. [26:45] And finally, uh teamwork. [26:47] Once you solve parallel agents, you've [26:49] also solved teamwork [26:51] >> Mhm. [26:51] >> because every time someone jumps into [26:53] the session, you can start an entirely [26:54] new you can fork entirely new VM for [26:57] them and they can work in parallel to [26:59] you. And also because the orchestrator [27:01] knows how to subdivide tasks, you can [27:04] also prompt in the same chat window and [27:06] we'll figure out how to make all that [27:07] work. And because we had the canvas, it [27:10] was is it's such a joy to just see other [27:13] cursors and people the the product just [27:15] becomes a lot a lot more live. So all [27:18] these components became [27:20] >> for Oh, final one I forgot. [27:22] We wanted it so that [27:25] when you make a mobile app, when you [27:27] make a website, when you make uh a deck, [27:30] when you make a video, all of that [27:32] should have the context of your project. [27:34] Previously on Replit and now every other [27:36] tool, [27:37] it's basically you need different tool [27:38] for a website, different tool for a [27:40] mobile app. So now with Replit, like you [27:42] built a really cool web app. So you go, [27:44] maybe I maybe I need an app. You can [27:46] just say make a mobile app. [27:47] Yeah, we'll like generate a mobile app, [27:48] we'll lay it out on the canvas. When you [27:50] hit deploy, [27:51] it deploys your web app to the web. It [27:53] deploys your app to TestFlight or the uh [27:56] or [27:57] uh or or your Android, whatever it is. [28:00] So so now you can run your entire [28:03] company on Replit. [28:04] >> What [28:05] What kind of skills do people need to [28:07] develop to to be able to make the most [28:10] of of products like yours? Like what [28:12] what is the you know, what do what do [28:14] people need to practice at? Like in your [28:16] mind, is it is it people have to get [28:17] good at prompting or is it more that the [28:19] system is going to get better at just [28:20] understanding whatever it is you're [28:21] asking it to do? [28:23] >> I actually think we're headed to like a [28:26] post-prompting world. [28:28] Um and you can you can sense that in [28:30] those sort of open claw and all the [28:32] derivatives. And the way people are [28:34] using it, um they're more giving it [28:36] high-level goals. Like optimize my [28:38] marketing funnel, you know, you things [28:39] like that. So [28:40] prompting will be there. It's like a [28:42] layers of skills. Prompting will be [28:44] there when you want to do more [28:46] interactive type work, but also you [28:48] should be able to give your agent [28:50] high-level commands. So, I think perhaps [28:52] agent five or maybe sooner, you should [28:54] be able to tell Replit "Every day, build [28:56] me a SaaS company and like try to market [29:00] it and see what works and make me some [29:02] revenue." Like she would be able to give [29:03] it like [29:04] >> high-level. [29:04] >> [laughter] [29:06] >> I think we're almost there [29:08] in order to be able to do that. So, what [29:10] kind of skills do you need? [29:11] Understanding what is possible is going [29:14] to be important. So, playing with these [29:16] tools a lot, having this like playful [29:18] mindset [29:19] of adopting these tools, playing around [29:22] with it, being plugged in [29:24] >> Mhm. [29:24] >> is super important. At some point I [29:26] think it was kind of a [29:28] drawback or a flaw to be someone who's [29:31] like constantly reading news or being [29:33] online, but actually now it's very [29:35] important to know what's happening, [29:37] what's coming down the line. [29:38] >> Yeah. [29:39] >> Um [29:40] I think um [29:42] not giving up is important. Like the [29:44] thing that you try to do today, the day [29:47] I cannot do, try it again in a month. [29:49] >> Yeah. [29:50] >> And I tell some Replit users if if [29:52] whatever you're trying to build Replit [29:53] cannot build today, try it in a couple [29:55] weeks. It might be able to build it. [29:57] Um and so that that mindset of like I'm [30:00] going to keep trying is is is is [30:02] important. Idea generation is still [30:04] going to be important. Obviously, we're [30:05] going to get to a point where where AI [30:07] is going to be really good at helping [30:09] you [30:09] generate ideas. Um but being generative, [30:13] just constantly thinking about problems [30:15] you want to solve, um [30:17] and just being creative and like [30:20] figuring out what what what what what [30:23] does the world need right now? [30:25] And um [30:27] and being generative is important [30:28] because let's say you're a small-time [30:29] entrepreneur, [30:31] um like you know, Peter Levels type of [30:34] entrepreneur or like this guy's very [30:36] famous on Twitter who built these [30:37] products. Oftentimes the products go [30:40] through cycles. Like we built like a [30:41] product that is really good for that [30:43] moment and you can generate like a [30:44] couple of million dollars, but then that [30:45] product is no longer relevant. So, you [30:48] need to be generative and creative in [30:49] order to continuously do that and you [30:52] can make millions of dollars doing that. [30:53] >> were starting RepliT today, [30:56] what would you do differently? [30:58] >> I mean, I made a lot of mistakes along [30:59] the way. [31:01] Uh so, if I was starting RepliT today, [31:03] I'd probably hope not to make as many [31:04] mistakes. So, [31:06] you know, culture is very important. Uh [31:09] you know, at some point we screwed that [31:11] up and we had to do we had to do a reset [31:13] and lay off and all of that stuff. [31:15] Um [31:17] I think [31:18] being really honest with yourself about [31:20] like product market fit is very [31:21] important. It's very easy to delude [31:23] yourself. [31:25] Getting any kind of user is an amazing [31:26] achievement. Getting any [31:28] any kind of money from users is also an [31:30] amazing achievement and you should [31:31] celebrate every one of those moments. [31:34] But true product market fit [31:36] is entirely different. It's like an [31:38] explosive thing. [31:39] >> Yes. [31:40] >> And so, being honest about that because [31:43] we've had periods where we're like, "Oh, [31:44] it looks like that's successful. Maybe [31:47] it's working." [31:48] And then like you keep going down that [31:50] path, [31:52] but in reality you should have changed [31:53] directions a little earlier. [31:55] >> Yeah. When it works, it really works and [31:57] you know it. In terms of AI development, [32:00] AI technology, what is something that [32:01] you are waiting for where you're like, [32:03] "This is going to unlock so much?" [32:05] >> You know, I've been waiting for computer [32:07] use models to to get better. They get [32:10] there slowly getting better. They're [32:11] like one of the things that is actually [32:13] kind of disappointing. [32:14] >> It's surprising to me that it's so hard [32:15] to build them cuz you'd think that'd be [32:16] the easiest thing in the world to get [32:17] data for. And do you do you have a sense [32:19] of why that is? [32:21] >> Well, language is a is a lot easier. Uh [32:24] language is also like much more [32:27] can be compressed down into [32:29] like [32:31] some kind of high-dimensional space a [32:32] lot easier than like [32:34] like a video feed, right? [32:37] >> But still like if we can make progress [32:39] on self-driving, surely we can make [32:40] progress on moving the mouse and [32:41] clicking on things. So, I think it's a [32:43] little bit of a of a mystery. That being [32:45] said, [32:47] coding in coding agents turned to be a [32:51] workaround and hack because [32:54] a lot of things that you could do in [32:55] front of the computer, [32:57] you could do with code, including, you [33:00] know, scripting an Excel sheet, right? [33:01] The reason Excel sheet agents got better [33:05] is because coding got better. The [33:07] reason, like, you know, some commerce [33:09] agents are starting to work because they [33:11] can call APIs and they can So, coding [33:13] turned out to be a bit of a hack or [33:15] workaround computer use agents. But [33:18] there's still a lot of software in the [33:19] world that's like very legacy software [33:21] that computer use agents would would be [33:23] really good at. [33:24] Also, when you're when you're creating a [33:27] vibe coding platform like like Replit, [33:30] you want to be able to test the apps [33:31] that people are making. [33:33] And you want to be able to be discerning [33:35] as well, like, things that are [33:37] functioning right, but also UX is not [33:39] very well, you want to be able to give [33:40] the user feedback on that. [33:42] >> Mhm. [33:42] >> But the models are not very good at [33:44] that. So, we spend a lot of time [33:45] >> question, I take it? [33:46] >> Taste question. So, we spend a lot of [33:48] time prompting and augmenting [33:51] computer use models in order to to make [33:54] them good at testing because Replit has [33:56] like a testing testing agent. [33:59] Um [34:00] So, that's one that I I think we'll [34:02] unlock a lot of different capabilities [34:05] once we get to computer use agents. [34:08] Um [34:10] I think, like, [34:13] yeah, I think it's become a buzzword, [34:15] but like continual learning. [34:17] Like, the way we're hacking around it [34:19] right now is we're writing to files. [34:21] Like, the agent will learn some kind of [34:22] skill and like we'll write a skill MD [34:24] file or something like that. [34:27] But true learning on the job is like not [34:30] been unlocked yet. [34:31] >> Yeah, yeah. [34:32] >> Uh and so for us to be able to like [34:34] deploy an agent inside an organization [34:37] and for that agent just to continues to [34:39] get better [34:41] for that org itself is is is such a [34:43] powerful thing. [34:45] But but that just still seems far away. [34:48] >> You know, I I I think, you know, what [34:50] one of the common themes I'm hearing [34:51] here is is like we're hey, we're going [34:52] to have like or we already have today [34:54] with Replit like agents to build [34:56] software and and you talked about how [34:58] we're increasing the abstraction until [35:00] finally you can just say [35:02] hey, I I want a company that just makes [35:04] money. How how is this going to work? [35:06] Like how what does that leave left for [35:09] people? And and what does the company of [35:10] the future look like if it's all just [35:12] agents all the way down? [35:14] >> Um I think I think the company of the [35:16] future is made of builders [35:20] and sales people broadly. [35:23] And but sales will change and sales will [35:25] be more like [35:27] let's help [35:29] companies transform. Let's use the [35:31] technology that we're building here to [35:33] help companies transform. So when sales [35:36] people are [35:37] more like I mentioned it earlier, like [35:39] more like evangelists, more like [35:40] educators. [35:42] But I don't think that part is going to [35:43] go away because a lot of [35:45] other companies will want to talk to [35:48] someone, will want to learn from [35:49] someone, right? That's how a lot of [35:51] people learn. [35:52] Um and they want to you they they trust [35:55] other humans. And so [35:57] the sales probably part is like one of [35:59] the more defensible jobs. [36:00] >> Interesting. [36:01] >> Um and then builders uh there's always [36:05] more to automate. There's always I think [36:08] our our job will continuously like get [36:09] higher and higher level. I mean, that's [36:11] already the case that [36:14] with computers we became a lot more uh [36:16] higher levels. Like computers were [36:17] literally humans, right? Like there was [36:19] like a you know, bunch of people like [36:22] doing computations. Tables of numbers. [36:24] And then we're like would you take that [36:26] entire room and you put it in a box, [36:28] that's a computer. And now there's my [36:30] job as like the operator of the computer [36:32] to use a computer for productive use. [36:34] Uh and [36:37] there was a lot of still manual work in [36:38] order to make that computer do [36:40] interesting things. [36:41] Uh software. And now we have an agent [36:44] that's using that computer to use So, [36:46] you know, different layers of [36:48] abstractions. I don't know when you have [36:50] a fully autonomous company, maybe at [36:52] some point there's certain [36:54] uh [36:55] styles of companies that could be fully [36:56] autonomous. [36:58] But I think you need [37:00] business generalists that [37:03] understand customers, understand what [37:05] people need, understand the economy, [37:07] understand where the world is headed, [37:08] understand AI technology, what what have [37:12] some vision, and you want to give them [37:15] So, here's like the abstract vision of a [37:17] future company. [37:18] It's like almost everyone is a founder. [37:21] They wake up in the morning and they [37:22] think, "How can I make the company more [37:23] successful? How can I make the company [37:25] make more revenue?" [37:27] And then they go around the company [37:28] finding problems to solve and then [37:31] creating or deputizing agents in order [37:33] to go solve these these these problems. [37:36] I'm already seeing that. We actually [37:37] have a team, like a vibe coding in [37:39] residency model at Replit. [37:41] And they have a very vague mission, [37:44] unlike like a product team or sales team [37:46] or very clear. Like, "Go around the [37:48] company, make it better." [37:51] So, they went to the support team. It [37:52] was like, "Okay, you know, you're using [37:54] Zendesk and all these other tools. What [37:57] are the main problems?" And we're like, [37:58] "Well, [38:00] we don't have a good way to prioritize [38:02] our [38:03] uh support queue. [38:05] Um for example, like there are customers [38:08] that are paying much more than other [38:09] customers and there are customers that, [38:12] you know, have more urgent tickets. And [38:13] so, they built like a way to to [38:16] visualize that visualize that and and uh [38:19] a way to like create more priority [38:21] queues, [38:23] um [38:23] and they spend some time with with the [38:26] support team and the CSAT score started [38:28] going up. And then to go around, they go [38:31] around their HR team. What are the [38:32] problems? Well, onboarding is a problem. [38:34] There's no place for people to to know [38:37] all the benefits and everything that we [38:39] have. Well, let's build an HR internal [38:41] HR platform. So, they they do that. [38:44] And I think that kind of role [38:46] is more [38:49] it is more where the future is headed [38:51] and more more people inside a company [38:53] will be more generalist entrepreneurs [38:56] that are trying to make the business [38:57] successful. [38:57] >> Well, thank you very much for joining [38:58] me. Pleasure to have you here at YC. [39:01] >> Appreciate it. Thank you. [39:04] >> [music]