How Non-Coders Are Vibe Coding $100K+ Businesses with AI | Amol Jain

非程序员如何借助 AI 氛围编程打造 10 万美元级业务 | Amol Jain Amol Jain 详细阐述了非技术背景的创始人如何利用行业专业知识、社区信任和 Replit 构建高收益业务，并列举了 Pep AI 首月创收 6 万美元，以及某企业评估工具在两个月内入账 18 万美元等实际案例。通过现场演示一款 AI 面试辅导应用，他概述了如何利用自动化安全威胁建模、自定义域名配置、原生 Stripe 变现以及搜索引擎和答复引擎优化，实现从粗糙原型到生产环境的关键跨越。对话还探讨了企业内部团队如何直接基于现有数据仓库，通过氛围编程打造定制化内部工具，积极取代昂贵的企业级 SaaS 套件。Jain 最终指出，随着前沿模型和单纯的代码语法走向同质化与商品化，真正具备防御壁垒的价值将属于那些拥有最后一公里基础设施、记录系统以及现实运营护城河的应用。 非技术背景的创始人正利用 Replit 将氛围编程的原型转化为六位数收益的业务，其中包括一名首月创收 60,000 美元的 22 岁学生，以及一位用三天开发取代了 10 万美元外包报价、并在两个月内实现 180,000 美元总收入的运营者。 (2:08) 将氛围编程原型推向生产环境需要加固“最后一公里”，这涵盖了零配置自定义域名、自动化漏洞扫描、即时 Stripe 接入以及答复引擎优化。 (14:01) 企业正通过赋能非技术类营收与数据运营团队，使其能够直接在核心数据仓库之上构建定制内部工具，从而系统性地淘汰成本高达六位数的 SaaS 订阅。 (21:35) 随着编写代码的成本逼近于零，经济价值正从日趋平庸化的基础模型和单纯的 UI 套壳中抽离，转移至记录系统、监管责任壁垒、线下实体运营以及分发渠道。 (28:22)

Amol Jain, Head of Product Engineering at Replit, argues that as artificial intelligence drives the cost of writing code toward zero, lasting commercial value shifts entirely from syntax generation to domain expertise, unique distribution, and end-to-end operational packaging. Founder Advantage Over Pure Programming Syntax (2:08) Generic consumer software, such as basic habit trackers or meal planners, faces rapid commoditization as users can easily prompt bespoke variations or deploy autonomous agents. Jain explains that builders achieving six-figure outcomes rely instead on an asymmetric "unfair advantage"—such as operational judgment, pre-existing community trust, or specialized subject matter expertise. In these cases, the generated application functions merely as an accessible vehicle that translates personal leverage into a cash-flowing business. Several non-technical builders illustrate this dynamic. Cedric, a 22-year-old student with zero coding experience, built Pep AI for the peptide and GLP-1 community, reaching $60,000 in subscription revenue in his first month. Similarly, repeat founder John avoided a $100,000 agency quote by building an enterprise AI proficiency certification platform on Replit in three days, generating over $180,000 in its first two months through his sales network. Professional gamer FaZe Apex co-created TryNearby, a local merchant-creator marketplace that exceeded $100,000 in annual recurring revenue and secured acceptance into Y Combinator's Summer 2026 batch. Hardening Vibe-Coded Prototypes for Commercial Production (11:44) Transforming a conversational prompt into a durable commercial product requires bridging the gap between raw interface code and production infrastructure. Novice builders often struggle with networking, domain management, and deployment pipelines. Modern platforms abstract this friction by automating database provisioning, container orchestration, and custom domain configuration directly within a single interface, allowing users to bind custom domains without external DNS overhead. Production hardening also demands automated security auditing and seamless monetization. Jain outlines three prevalent security risks in AI-generated codebases: exposed API credentials, malicious or hallucinated open-source dependencies, and improper data architectures that leak personally identifiable information. Rather than hiring costly manual penetration testers, automated security agents generate architecture-specific threat models before deployment. For commercial monetization, integrated platforms provision Stripe merchant sandboxes, test keys, subscription tiers, and webhook endpoints via natural language, eliminating technical roadblocks to collecting payments. Displacing Enterprise SaaS with Bespoke Internal Workflows (21:35) Collapsing software costs are enabling non-technical operators to replace six-figure SaaS subscriptions with custom in-house software. Jain notes that many legacy SaaS products are merely thin interfaces over standard data. To solve live AI demonstration latency and non-deterministic model behavior during sales calls, a non-technical Replit revenue operations employee built a bespoke Chrome extension and demo generator in two days, replacing commercial demo vendors that cost six figures annually while enabling instant feature customization. This unbundling extends to complex corporate infrastructure. Replit eliminated its third-party business intelligence subscription after a data scientist vibe-coded an internal analytics dashboard directly on top of Google BigQuery. Peter Yang notes this trend defines the "10x employee," who creates widespread organizational leverage by engineering custom software for entire internal teams. While core systems of record like Salesforce remain untouched, the surrounding software stack is increasingly replaced by purpose-built internal applications. Defensible Enterprise Moats and Multi-Model Economics (28:22) When code generation is free, software defensibility retreats into moats that natural language prompts cannot synthesize. Jain identifies these defensible assets as authoritative systems of record, regulatory compliance risk, physical human labor, and established network effects. For instance, Salesforce endures as an immutable data repository, Workday survives because enterprises pay it to absorb complex legal liability surrounding payroll compliance, and consumer delivery apps remain protected by their physical logistics networks. Consequently, traditional SaaS products are increasingly operating as headless backends navigated programmatically by AI agents and Model Context Protocol servers. At the platform level, commercial power is shifting away from frontier model labs toward the application layer. With open-weight and alternative models closing performance gaps, enterprises avoid single-vendor lock-in and vanity benchmark competitions, moving away from what Jain calls "token maxing." Architectural advantage now favors model-agnostic systems that pair a high-reasoning frontier model for system design with lower-cost sub-agents for routine execution, anchoring long-term defensibility in reliable deployment infrastructure and distribution. While AI code generators remove technical development barriers, sustainable software businesses depend entirely on assets external to the code itself: proprietary data, regulatory liability assumption, physical operations, or deep domain distribution channels.

[0:00] Peter Yang: make money while I sleep [9:42] Amol Jain: Users are... inherently lazy. [11:44] Peter Yang: trust-building or a grift [17:50] Amol Jain: Add a way to let my users buy subscriptions for $100 a month or per year. [21:35] Amol Jain: incredibly average and basic [21:35] Amol Jain: at the mercy of someone else’s roadmap [25:28] Peter Yang: 10x person is building a tool that the entire team or the entire company can use. [28:22] Amol Jain: If the cost of code goes to zero, what do you pay for? [34:02] Amol Jain: lighting a cigarette with a flamethrower

[0:00] He went to an agency that quoted him a [0:01] $100,000 plus. [music] So, he jumped on [0:04] and essentially within 3 days was able [0:06] to build the whole thing end-to-end. [0:08] Within the first 2 months, he was [0:09] already at $180,000 plus in revenue. If [0:13] the cost of code goes to zero, what do [0:15] you pay for? It's this unique expertise [0:17] that they have. It's the unique judgment [0:19] or taste. It's the security. It's the [0:21] distribution. It's all of that extra [0:23] packaging that meets the user where they [0:25] are, and users pay for that. [0:30] >> Hey everyone, today I'm really excited [0:31] to welcome Amjad as Replit's head of [0:34] engineering. He has a first-hand look at [0:36] what people have built with AI that [0:38] actually make money and become real [0:39] businesses. And we're also going to talk [0:41] about, you know, how AI's impacting SaaS [0:44] and how Replit and other app layer [0:45] companies can differentiate from the [0:47] frontier labs. So, welcome, sir. [0:49] >> Thank you, Peter. Great to be here. [0:51] >> So, now that I'm kind of independent, [0:53] you know, we were just talking about [0:54] this, I would love to just make an app [0:56] that makes money while I sleep. [0:58] And you have a broader view of this than [1:00] anyone, right? At Replit. So, can you [1:02] start by sharing some examples of maybe [1:04] Replit apps that actually make money, [1:06] people have actually built real [1:07] businesses with this stuff? [1:09] >> Absolutely. Yeah, I [1:11] you know, one of the key reasons I [1:12] joined Replit was was was exactly that. [1:15] The the [1:16] Replit really lets people, you know, [1:20] it's the entrepreneurial dream of of of [1:22] helping tons of people create businesses [1:25] that make money. And and that's exactly [1:27] what we're seeing is every day we see [1:29] tons of new businesses started start [1:31] that make real money. And these don't [1:34] have to be with massive TAM. We see [1:36] businesses that, you know, serve 100 few [1:38] hundred customers, but these are making [1:39] real money and helping people, you know, [1:42] have that upward up economic mobility. [1:44] And so, um essentially, I have I have a [1:47] few examples that that I that I pulled [1:49] out. Um [1:51] in some ways, you know, [1:53] what Replit does is takes a person's [1:56] advantage. That advantage might be their [1:59] their distribution network, their [2:01] community, their their expertise, their [2:03] judgement, and turn that into into a [2:05] business. Um so, let me let's let's [2:08] start with [2:10] with one that's called um Pep AI. [2:14] And [2:15] Cedric Cedric is someone who lives in in [2:18] in Oregon. He goes to University of [2:20] Oregon. He's 22 years old. He has never [2:23] coded in his life before. Um [2:25] his roommate and and Cedric decided that [2:28] they want to they'd heard about [2:29] peptides. They wanted to take peptides, [2:31] and they got really deep into this [2:33] community of peptides and GLP-1s. And [2:36] and as he did that, he sort of he sort [2:38] of realized that there was [2:40] um really an opportunity to help the [2:42] broader community and turn that into a [2:44] great business. He wanted to build [2:46] something that helps anyone track [2:49] uh track their their peptide and GLP-1 [2:51] intake, and also um uh learn more about [2:55] about these emerging sort of sort of [2:57] medications and such. And so, he jumped [3:00] on Replit. Again, no exposure to code [3:02] before. He was able to build a mobile [3:04] app. Um and here's his landing page. [3:09] He built it all in Replit. [3:11] >> Wow. [3:11] >> Um and and really tapped that need [3:14] because um [3:16] after a little bit of of of testing and [3:17] experimentation with distribution, he [3:19] was able to grow this product very [3:21] significantly. He was he was making [3:22] $60,000 in in his first month, and [3:26] and the app has grown [3:28] it now has tens of thousands of of [3:30] active users and growing every day. And [3:32] so, it's really amazing to see someone [3:34] who just had that entrepreneurial spirit [3:36] and that curiosity um [3:38] learn a lot about a certain topic and [3:40] then and then build a great business. [3:41] >> And and and he's making money from like [3:43] subscriptions or in-app purchases? [3:46] >> He's making money through subscriptions, [3:48] yes. [3:48] >> Okay. Yeah, that's super awesome. The [3:50] the website look looks great. Yeah. [3:52] >> Yeah. Um another example that that I'd [3:54] love to walk you through. This is a [3:57] a a wholly different persona. So, John, [3:59] he is actually a repeat founder. He had [4:01] started something in the past as well. [4:03] Um as AI was taking off, he sort of [4:06] realized that there was an opportunity [4:07] to um to uh [4:10] help people measure their AI proficiency [4:12] and and even get certified for it, [4:14] especially [4:15] enterprises and companies could do that. [4:17] And um he's non-technical, didn't know [4:20] how to build it, so he went to an agency [4:22] that quoted him a $100,000 plus to build [4:24] this platform for him. Um that didn't [4:28] you know, that didn't sit right with [4:30] him. He had heard of Raplit, so he [4:31] jumped on and essentially within 3 days [4:35] was able to build the whole thing end to [4:37] end. Uh something that that has this AI [4:40] proficiency [4:41] um AI proficiency sort of tool that can [4:44] measure a person's AI proficiency, way [4:46] to get uh certified, etc. And he built [4:50] and launched it and within within the [4:52] first 2 months he was already at [4:53] $180,000 plus in in in revenue um [4:57] purely because as as a repeat founder he [4:59] had this great product intuition. Um he [5:02] he knew how to distribute and sell, he [5:05] just didn't know how to build and and [5:06] and and and the fact uh tool like Raplit [5:09] really made it easy for him to do that. [5:11] >> Yeah, and now he has like a homepage for [5:13] people to come and check it out and take [5:14] the test and maybe get a lead and stuff [5:16] like that, right? Yeah. [5:17] >> Exactly. [5:18] >> Yeah, 100K man, maybe they're maybe that [5:20] agency using Fable tokens. I don't know [5:21] why it's on there. [5:22] >> [laughter] [5:23] >> That's very expensive, yeah. Okay, do [5:26] you do you have another one? [5:26] >> Yeah, let's talk one more. Let's see one [5:28] more. So, if you're familiar with gaming [5:31] at all, if you've ever [5:32] ever been a gamer or been in those [5:34] communities, you definitely know of [5:36] FaZe, the the clan that is incredibly [5:38] popular, one of the most successful [5:40] gaming groups out there. And FaZe Apex [5:42] Yusuf, he's he's a phenomenal, very [5:45] famous and and very successful gamer. [5:47] Um he had over, you know, years built [5:50] built an audience. He he's he's he's in [5:52] some ways a creator who's, you know, [5:53] streams his video games and such and and [5:55] and has had made money through that. Um [5:58] but recently he wanted to start a [5:59] different business and again, someone [6:01] very non-technical, just someone who [6:04] understands sort of audiences and [6:05] creators. Um and what he wanted to build [6:08] was a way for [6:10] local businesses to connect with local [6:13] creators to drive that word of mouth for [6:15] them. [6:16] >> Mhm. [6:16] >> Um it's it's it's it's a topic, you [6:18] know, he obviously knows a good bit [6:20] about. And so, he came to Replit, built [6:23] the whole thing here, and um he's been [6:27] incredibly successful. He They just [6:28] They're They're in YC summer 26 batch [6:31] now. [6:32] Um him and his co-founder and for for [6:34] this company called TryNearby, and [6:36] they're upwards of 100,000 ARR [6:39] uh when we when we last chatted with [6:40] them. [6:41] >> So so this is like uh [6:43] creators helping local business like [6:44] restaurants and stuff get the word out. [6:47] >> Yeah, it's almost a marketplace that's [6:48] matching those those local creators to [6:50] to these local businesses and to get [6:52] that word of mouth out, yeah. [6:54] >> That's amazing, yeah. Yeah, I have a lot [6:55] of restaurants that I love that I'd love [6:57] to promote, so we're going to check it [6:58] out. Yeah. [6:59] >> Yeah. [7:00] >> This episode is brought to you by [7:02] Riverside. I've used Riverside for years [7:04] to record my podcast because it records [7:07] in 4K resolution each person locally, so [7:10] the audio and video still comes through [7:12] clean even if a guest's Wi-Fi gets [7:14] shaky. But the reason I love it now more [7:16] than ever is what happens after we stop [7:19] recording. If I go in here, then I can [7:21] use these AI tools to remove pauses, [7:23] remove filler words, and just clean up [7:25] the recording. And I I can also edit the [7:27] transcript directly, and it will [7:29] automatically generate clips with [7:31] captions ready to publish to YouTube, [7:33] Spotify, and all types of social media [7:36] platforms all from one place. As a [7:38] one-person creator business, that [7:40] matters a lot. Riverside is the upgrade [7:42] your content workflow needs. Try it at [7:45] creators.riverside.com/peteryang [7:49] and use code Peteryang at checkout to [7:51] get 1 month completely free. That's [7:53] creators.riverside.com/peteryang. [7:57] Now, back to our episode. [7:59] Let me kind of stress this a little bit, [8:00] right? Cuz like I personally have also [8:02] built a bunch of apps using Replet and [8:05] other things like I built like a fitness [8:06] app and I built like an app to curate my [8:09] movie taste and all that kind of stuff. [8:10] But I feel like so so like I I guess my [8:12] question is [8:13] do you see any common themes across [8:15] these examples of apps that actually [8:17] make money? Like are they targeting a [8:18] specific very niche problem or is it [8:20] kind of like it's not just software but [8:22] it's also services? [8:23] Like is there do you have any tips [8:25] around some of this stuff to actually [8:27] build a real business out of this? [8:28] >> Yeah, so [8:30] I think I think [8:31] you're correct that a lot of sort of [8:34] software it's it's becoming rather easy [8:36] to build software. And so a lot of sort [8:38] of like generic use cases, meal planners [8:42] or or you know exercise trackers not to [8:45] not to not to diminish your idea [8:47] are getting rather commoditized. People [8:49] are building their own versions or or in [8:51] some ways I see in the future agents [8:53] just just kind of swallowing all of that [8:55] whole. Everyone has an agent that has a [8:57] skill that does this. Um I I so I think [9:00] the the the the thing that said the the [9:03] overall pie of businesses is still [9:05] expanding. If you look at Stripe data, [9:06] you can see the number of businesses [9:08] being created going up. Uh it's because [9:11] it's it's that unique sort of advantage [9:14] that that people have and especially the [9:17] sort of you know average Joes in the [9:18] past that have no exposure to [9:19] technology. It's their It's their sort [9:22] of like unique expertise that they have. [9:24] It's the unique sort of judgment or [9:27] taste or [9:28] their their community, their their [9:30] network. It's people tapping into those [9:34] to create the types of things that that [9:36] can be actually tangible businesses. And [9:39] these don't have to be have to be, you [9:41] know, massive businesses. Again, I think [9:42] I think I really like to emphasize that [9:44] that there is a lot of success you can [9:45] have with just a small business that [9:47] that thrives. And so, I think it's it's [9:49] about identifying that that what that [9:51] unique advantage is and really leaning [9:53] to that. So, we see, you know, [9:55] pool cleaners and and and and nurses and [9:59] and jewelers and such, all these people [10:01] with domain expertise that in the past [10:05] they wouldn't have been able to [10:06] businesses, now really bring those ideas [10:09] to life. So, I think I think the tip is [10:10] truly finding that that that specific [10:13] advantage. And I think [10:15] one thing to always remember, users are [10:18] are are [10:19] people in inherently are lazy. Not [10:21] everyone has that entrepreneurial spirit [10:22] and desire. Not everyone's going to [10:24] going to, you know, do everything [10:26] themselves. There's people, if you offer [10:28] them value, they will pay for it. [10:30] >> It's kind of like dumb for you. Okay, so [10:32] so let me let me kind of summarize. So, [10:34] I guess there's one aspect is like what [10:36] is your unique skill set? What is your [10:37] unique knowledge? And also like you [10:39] can't just like wipe out this thing and [10:40] and like put it on there and hope to get [10:42] business. You got to got to do a lot of [10:43] marketing, right? You got to like reach [10:44] the unique audience that you have [10:45] online. And then maybe there's like some [10:48] additional services or other things that [10:50] you can offer. [10:51] >> Absolutely. Yeah, and I think I think [10:53] more and more I I I see this as [10:55] you know, if you if you just zoom out [10:56] and think of businesses in general, the [10:58] types of businesses that survive are the [11:00] ones that have sort of [11:02] either this expertise that we've been [11:04] talking about or more on the B2B [11:06] enterprise side, it's it's you know, [11:08] businesses that offer trust or or [11:10] infrastructure or network or systems of [11:13] record, data, [11:15] even atoms, services as we were talking [11:17] about. I think those are those are [11:18] incredible businesses that are now [11:20] accessible to everyone. [11:22] >> That makes sense. That makes sense. [11:24] Um awesome. So, um [11:26] let's say I build like a app or like a [11:28] website on Replit or like another [11:31] provider. And how do I like I I hit a [11:33] button take a life, right? So, [11:35] is it as easy as that or should I do [11:37] some security scans, do some checks [11:39] before I do it? Like can you show us how [11:40] this What's the reason of vibe coded app [11:43] and app that's actually ready for [11:44] production? [11:44] >> Absolutely. Good great question. And and [11:47] I what I did was in preparation of this [11:49] actually come up with an example that [11:51] that we can walk through and and and [11:53] show the audience of of how to go from [11:55] that raw product, that purely vibe coded [11:58] first version to something that can be [12:00] production grade and also [12:03] also, you know, be something that grows [12:05] into a business. [12:06] >> Yeah. [12:08] >> let me [12:11] go to this screen. So, [12:12] you have you know, you have this amazing [12:15] podcast where you have [12:17] uh product experts uh come on here. You [12:20] have these product experts in your [12:21] network. And you know, one topic you [12:23] guys always discuss is is is product [12:25] market product management market has [12:27] changed. Like the the jobs are [12:28] different, the interviews are different. [12:30] >> Yeah. [12:30] >> What if since you were looking for a for [12:32] an idea of a business that can make you [12:34] money, what if you were to build a an [12:35] interview coach where you pair up [12:38] product experts with with applicants so [12:41] they can do mock interviewing and and [12:43] you know, rather than just repeating the [12:45] circles framework or or or such, [12:48] actually give real or you know, actually [12:50] can pass these interviews. So, [12:52] >> That's a great idea. [12:52] >> Um [12:54] I I figured we will will do a session [12:56] here and see how we would take this this [12:58] idea and turn it into a real business. [13:01] Um [13:02] So, first of all, let me let me sort of [13:03] walk you through how I got to even this [13:04] page. Uh essentially, I I dumped uh um a [13:08] very sort of [13:10] I I I use WhisperFlow a lot, so I just [13:12] WhisperFlow dictated what I wanted this [13:14] to be and used Droplet design first to [13:17] to sort of explore different directions [13:19] of of what what might feel right [13:22] for for a product like this. And I think [13:25] the I tried a different vari- bunch of [13:27] different variations and I think what I [13:29] really liked was the simplicity of your [13:30] landing page so I just took that as the [13:33] as the starting point and asked Droplet [13:35] to build it. And what it did was was [13:37] built this this whole landing page with [13:40] some basic functionality to accomplish [13:42] what we're looking for. [13:44] But this is what I would call that raw [13:46] product and now we can sort of focus on [13:47] how do we turn it into a real thing. [13:49] >> Yeah, I think having my face on there [13:51] can either be trust building or grip [13:53] thing I know but yeah [13:55] >> trustworthy face I would I would [13:57] purchase [13:58] >> Great great [14:00] >> Amazing so so [14:01] I think first thing is is even taking [14:04] this thing live you know you you always [14:06] see that those stories of people use a [14:09] cloud code and they can of course buy [14:10] code something but then [14:12] they're they're sharing local host links [14:14] to their friends that obviously is a [14:16] it's it's almost a meme now because how [14:18] often it happens and I think one one [14:20] great thing about Droplet is we have the [14:22] home road infrastructure that abstracts [14:24] away all of of hosting and databases and [14:28] and and and all of all of that so to to [14:30] an end user our users tend to be [14:32] non-technical it's a rather seamless [14:34] process and so I'll I'll let's let's [14:37] dive into this what we call it [14:38] publishing [14:39] and as a part of publishing I already [14:41] did publish it but I'll I'll show you a [14:43] couple of things one I think it's [14:45] important to have a great domain we're [14:46] building a business here so [14:48] we have a way for you to easily buy a [14:50] domain I bought coachpeter.net [14:52] coachpeter.com was a little too [14:54] expensive so so here we are [14:56] um [14:57] and then then we go on to publishing um [15:00] but before we publish one thing that's [15:02] important is making sure that your [15:07] site is actually secure it can actually [15:10] you know survive attackers because [15:12] there's a lot of bad actors out there [15:14] that are constantly attacking these and [15:15] by coded websites are are notoriously [15:19] notoriously exposed to those things and [15:22] what we've done is built this [15:24] what we call security center. It's [15:26] essentially a security scan of your app. [15:29] Um [15:30] in the past, you would have to have a [15:32] security team that you that you contract [15:34] with, [15:35] pay them a lot of money, spend they [15:37] spend 1 to 2 weeks doing this and give [15:39] you back a report which you then have to [15:41] go build it. We wanted to we essentially [15:43] built a product that runs the scan, [15:46] gives you the outputs, and and [15:49] just ensures that your app is actually [15:51] secure right before publishing, but then [15:53] also on on a recurring basis [15:57] as you build more functionality into it. [15:59] And so, to see how this works, let's say [16:01] we're just going to hit one button which [16:04] opens up a new task [16:06] and [16:07] starts running a security scan. [16:09] What's happening here is [16:11] the [16:12] the agent Droplet agent is security [16:15] agent specifically is building a a [16:16] threat model of [16:18] for this specific thing, what what are [16:20] all the ways things could go wrong, and [16:22] then it it's running a deep scan with [16:24] the different scenarios and testing to [16:26] make sure that that [16:28] that it can detect all of those problems [16:30] and then surface them to us. And so, [16:32] this is going to take a little bit of [16:34] time, but but I just want to kind of [16:35] show what what happens on the security [16:37] side of things. [16:38] >> Okay. So, while this is running, let me [16:40] ask you this. Like, you know, as an [16:42] engineering leader, [16:43] what are like the top two or three [16:45] things that could go wrong? Like, [16:46] obviously, if I put my API key in the [16:48] code or something, right? That's like [16:49] one thing that could go wrong. But like, [16:50] is there something else that this thing [16:51] looks for? [16:52] >> There is um so much that could go wrong. [16:55] There's there's you could agent could [16:57] pick up a package that, you know, [16:59] there's so many malicious packages out [17:01] there. Could install a malicious package [17:02] that that steals stuff. Um secret [17:05] scanning that you would talk that you [17:07] talked about. Um [17:09] There are there are very clowny things [17:11] that could often happen as well where [17:13] agent, for example, [17:15] um [17:16] could be leaking PII because it's not [17:18] stored in the right way. [17:20] >> Mhm. Yeah. [17:20] >> [clears throat] [17:21] >> Stuff like that could happen. And so, [17:23] this essentially the Replit Agent that's [17:25] building it already does a pretty [17:27] pretty great job of making sure that [17:29] doesn't happen, but it makes it it it's [17:31] important to just tie the knot and make [17:33] sure that everything is actually [17:35] actually [17:36] um you know, secure. [17:38] >> Makes sense. [17:38] >> while this is running, uh maybe we can [17:41] we can talk about a few other few other [17:43] interesting things that that turn into [17:45] uh turn this again raw product into a [17:48] into into something that's production [17:49] grade. [17:50] >> Sure. Yeah. [17:50] >> Um [17:51] Yeah, another thing that I would love to [17:53] highlight. So, we'll go back to main [17:54] while our security our security scan [17:56] scan is working here in parallel and and [17:58] it's scanning. But, let's talk about a [18:00] few other things. Um another thing, you [18:02] know, we would want to monetize a [18:05] at the end of the day our business is [18:07] something that will actually make money. [18:09] So, you can just ask Replit Agent to add [18:12] a way to [18:14] to uh let my users buy sub- [18:18] 100 [18:23] dollars 100 a month or [18:27] per year. [18:28] Um [18:29] and it's as simple as that. We [18:31] essentially have this um this deep [18:33] partnership with with Stripe where we've [18:35] abstracted away away again all of the [18:37] complexity. So, user doesn't have to go [18:40] sign up for Stripe, uh you know, do the [18:42] KYC up front, [18:44] figure out what a plan says, what prices [18:46] are, what subscription is, what is what [18:48] one-time is. [18:49] >> Mhm. [18:49] >> Um all of that is taken care of. We are [18:52] in the background setting you up a [18:53] Stripe sandbox, [18:55] have setting up the key, setting up the [18:57] plans and prices, and and making the [18:59] whole thing work so you can test. And [19:01] once you feel good, that's when we give [19:03] you the live live Stripe to go and just [19:07] swap out the last last pieces to take it [19:09] live. Um and I think we we see so many [19:13] people who've never monetized anything [19:15] in the past able to add this [19:17] functionality now and and turn things [19:19] into real businesses. [19:20] >> Yeah. Yeah, that that that's super easy. [19:22] Do you have to make a Stripe account [19:23] first or is it just kind of works? [19:25] >> It makes the Stripe account for you and [19:27] all you have to do is just go ahead and [19:28] claim it. [19:29] >> Nice. Nice. [19:31] >> Another thing I will I'll touch on. We [19:32] were just talking about distribution. At [19:35] the end of the day, so much of the game [19:36] is distribution if everyone can build. [19:38] And so, we're investing a lot in in in [19:42] making that in in giving you the tools [19:44] that any entrepreneur can [19:47] try to distribute and and get audience [19:48] for their product. And there's a lot [19:51] more that's coming very very soon, but [19:53] but one thing that I would love to show [19:54] you is [19:55] um this this whole idea of an SEO scan, [19:58] you know, [19:59] especially for the tool that we're [20:00] building here, which is a a product [20:02] management interview coach. That's a [20:05] very hot term on Google and and and and [20:08] if with answer engines. So, you can run [20:11] a run a SEO agent as well that that does [20:14] all the work to make sure that your app [20:16] is discoverable by by both, you know, [20:19] search and and answer engines like like [20:22] ChatGPT. Um, so it really helps you [20:25] solve one piece of growth. There's a lot [20:27] more that is that that's coming in terms [20:29] of ads and and and [20:31] um [20:32] giving people a way to market their [20:34] product through emails and such, but [20:35] this is our first bet on this space cuz [20:37] we really want to help people not just [20:39] build great businesses, but also grow [20:41] those businesses. [20:42] >> Yeah, this is a really good smart [20:44] investment. I think I'm lucky I built [20:45] like a big Twitter audience and and I [20:48] feel like audience is like the durable [20:50] asset these days. Like if you have a [20:51] good audience, as long as you're not [20:53] grifting, you can like launch multiple [20:54] products and hopefully one of them will [20:57] work, you know? So. [20:59] Yeah, really important. [21:00] >> Yeah, so, you know, to to sort of [21:01] summarize here, while while these agents [21:03] are are running, they'll it'll a while, [21:05] but essentially we took this we took [21:08] this this uh design that we had built, [21:12] turned it into a real product, and then [21:14] now we've added things like monetization [21:17] and security and and figured out a way [21:19] to even distribute it. And I think that [21:22] really encompasses the journey that we [21:23] see a lot of lot of Replit users go [21:25] through every day. [21:26] >> Awesome. All right. Well, let's talk [21:28] about another topic. So, I I think [21:30] Replit has been doing really well in the [21:31] enterprise market. [21:33] And what I'm seeing is that you know, [21:35] companies like they don't have to pay [21:37] for all these crazy SaaS products. Like [21:38] they can just build a bunch of internal [21:39] products to replace them. Like is is [21:41] this actually happening or is it [21:43] something I'm making up? [21:45] >> Absolutely happening. We see this every [21:46] day. It is [21:49] the the whole sort of past 10 years led [21:51] to a lot of SaaS startups that were just [21:54] essentially interface over some [21:56] interface over some data. And now that [21:59] the cost of code is going almost to [22:01] zero, [22:03] there is real reckoning and people want [22:05] to build It's not just about money. It's [22:06] not just about saving on those costs. [22:08] It's very much about building bespoke [22:10] solutions that that fit your workflow as [22:13] a as a company. We internally, you know, [22:15] this is obviously biased take, but we [22:17] internally have replaced so many tools [22:19] with with with [22:22] things that we built ourselves that work [22:24] better and are tailored to our use [22:25] cases. [22:26] >> Do you have some examples to show for [22:27] that? Like internally how Replit uses [22:29] tools? [22:30] >> Yeah, so let's see. I think maybe maybe [22:32] we can talk about [22:34] recent one that that I really like. Um [22:37] so, [22:38] you know, as we were just discussing, [22:40] Replit's now getting really big in the [22:41] enterprise. [22:43] On the backs of their success, our our [22:44] sales org has grown a ton. And [22:47] what what the sales org is doing is [22:49] demoing to prospects. It's it's it's a [22:51] product that you really see the power of [22:52] when you when you demo it to a prospect. [22:55] And um [22:56] >> [clears throat] [22:57] >> but but demoing a product like Replit in [22:59] general AI products is is of hard [23:00] because [23:01] A, as we just saw, we started a security [23:04] scan, but it's going to take 10 minutes [23:05] for it to complete. So, it's there's [23:07] there's that inherent dead time and lag. [23:09] And then secondly, they're [23:11] non-deterministic. Things could go [23:12] wrong. It might produce an output that [23:14] you're not not super keen on. Agent [23:16] might take different directions. And it [23:18] makes demoing really hard. So, what we [23:21] wanted was to create a compelling demo [23:22] that that really lets that's that's [23:24] predictable that lets someone say [23:26] you know, follow a script and and show [23:28] the right things to a to a customer. Um [23:31] there's tools for this. There's there's [23:32] tons of startups that do this. They are [23:35] the products are incredibly [23:37] average and basic and um and they're [23:41] expensive, you know, they can easily run [23:42] into six digits. And so, [23:44] someone on our RevOps team, again, never [23:46] written code in in in their lives, [23:49] took this idea and over two days built [23:51] something that that's now our sales [23:53] folks are using. So, I would I would [23:54] love to show [23:55] >> Yeah. [23:56] >> that here. Great. So, [23:58] as we were describing the So, just just [24:00] to just to kind of kind of kind of share [24:01] what this product how how it actually [24:03] works, there's a Chrome extension. And [24:06] a salesperson would install the Chrome [24:07] extension. Let's say they want to create [24:09] demo for a specific thing. They'll use [24:11] the Chrome extension to walk through the [24:14] product, record screens, and annotate [24:16] things. And then once they're ready, it [24:18] turns into this [24:19] into this artifact that that that is [24:22] demoable that you could present to [24:24] prospects, give it to your sales people, [24:25] and such. And so, let's let's look [24:27] through something a rapid build flow [24:29] that someone recently built. Very core. [24:31] This is the core interaction that [24:33] that you know, every prospect [24:36] we we walk every prospect through. [24:38] >> Yeah. [24:39] >> And [24:40] couple of other things I'll mention [24:41] here. Like I was saying, again, it saves [24:44] on say six figures of of SAS spend, but [24:47] more importantly, [24:48] it is [24:50] bespoke software where anytime we want [24:52] to add a feature, we just do it. We're [24:54] not at the mercy of someone else's road [24:56] map. And so, someone asked the creator [24:58] of this tool, "Hey, I would love to have [25:00] way to prospect to to create different [25:02] demos for different personas." So, he [25:03] added a persona feature. And so, now [25:05] there's for different personas, there's [25:07] different demos and such. I think that [25:09] is incredibly powerful and just really [25:11] lets us lets us you know, solve our own [25:14] problems and not pay someone money for [25:16] it. [25:17] >> Your point makes a lot of sense, too, [25:18] cuz as a SaaS provider, if I want to add [25:20] a feature, I have to think about all my [25:22] other customers, too, right? But if I [25:23] just have personal software that we're [25:25] using itself, it's so much faster and [25:27] easier to change. [25:28] >> Absolutely. Maybe another example that I [25:30] that I really like personally very [25:32] recently as as as a person who you know, [25:36] so much of my job is is building [25:38] products and being deep in the product [25:39] side of things, analytics is is so [25:40] important. And so, [25:42] we recently churned off of our our [25:47] analytics BI analytics provider and [25:50] built something in-house entirely. And [25:52] so, you know, the data was already ours. [25:55] It was already in our big query. [25:57] It didn't make sense for us to pay a [25:58] very hefty fee to a product that didn't [26:02] quite meet our needs in a lot of [26:03] different ways. And so, we this is this [26:05] is hot off the press as well. We [26:06] essentially built built something that [26:09] does exactly the same thing. I'm I'm [26:10] showing something that's not not [26:12] confidential. It's essentially an admin [26:14] portal tool and how many people are [26:16] opening it over time. Uh but you can see [26:19] it let it gives all the BI capabilities [26:21] of different types of charts, [26:23] comparisons, time frames. You can create [26:26] funnels or data tables or retention [26:28] charts and and and that's a different [26:30] things. It does all of that, but it's [26:32] ours and and we've we've been slowly [26:34] adding a lot of intelligence to it as [26:37] well to make it really really great and [26:39] and and solve for our use cases. [26:41] >> Yeah, I mean, those those analytics [26:42] platforms cost like millions of dollars, [26:44] right? [26:45] Expensive. [26:46] >> do. [26:47] >> Yeah. Wow. Okay, but that wasn't like [26:49] coded by a single person, right? Cuz [26:50] that that's that was pretty complicated. [26:52] Like is there is a team behind that or? [26:53] >> Believe it or not, it was live coded by [26:55] one person. In this case, it helps that [26:57] the person is a data scientist [26:58] themselves, so they understand a little [27:00] bit better, but they're not a coder, [27:02] they're not a builder, but they were [27:03] still able to do this. [27:05] >> That's amazing, dude. And I can see like [27:07] as that product improves, you can just [27:08] like ship it to customers too, right? [27:10] You can have like data analytics [27:11] platform for Retool. [27:13] >> We absolutely could, yeah. [27:14] >> Wow, okay. And do you guys encourage [27:16] employees to do this cuz they also have [27:17] their sales job and their, you know, [27:18] data job? [27:20] Is there a culture that encourages them [27:21] to do this? [27:22] >> Very much so. And and and it's not even [27:24] us having to push them. I think the [27:26] types of people that that Retool hires [27:28] have this sort of agency and that drive. [27:31] So, a lot of them [27:32] want to build the things themselves. [27:34] There's always that lens of of [27:37] how do we build something, A, it helps [27:39] us dog food, but then B, helps us [27:41] operate more efficiently as well. So, [27:43] our sales team, for example, has [27:45] replaced a lot of their sales stack. [27:48] Salesforce still is the system of [27:49] record, but but but outside of that, [27:51] it's all the tools are are are own. So, [27:54] yeah, it's it's rather common and it's [27:55] amazing to see because [27:57] as product builders, this is all the [27:58] feedback that we get today. [28:00] >> Yeah. [28:01] I I feel like this is how you become [28:02] like a 10x engineer or 10x sales person. [28:04] Like the 1x person is just thinking [28:05] about doing their core job, right? And [28:07] and like maybe like the the 2x person is [28:08] doing like building a personal tool to [28:10] do their job better, but a 10x person is [28:12] building a tool that the entire team or [28:14] the entire company can use. Then then [28:15] you're basically helping everyone else, [28:17] you know? [28:17] >> Exactly. You're creating leverage for [28:19] the whole company. That that is truly [28:20] the 10x definition, yeah. [28:22] >> Okay, so so let me ask you this. Um [28:24] so, what SaaS tools do you think can [28:26] actually survive inside companies then [28:28] if if you can live code every everything [28:30] like like is it like the Workdays and [28:32] the Jiras that are like [28:33] >> [laughter] [28:34] >> too complicated to remove or? [28:36] >> Yeah. Um [28:37] I think we were briefly touching on this [28:39] earlier, but essentially the the test I [28:42] think is [28:43] if the code [28:45] becomes, you know, the the cost of code [28:48] if the cost of code goes to zero, what [28:50] do you pay for? And and I think if you [28:53] apply that test, the few things that [28:55] stand that that that that test are [28:58] um [28:59] it's trust, it is data, it's [29:02] infrastructure, it's atoms, actual [29:04] physical things, it is labor, it is [29:09] network. So, I think [29:12] software that offer that, where it's not [29:15] just an interface over some data, [29:18] those are those are valuable and will [29:20] continue to exist. I think a few [29:21] canonical examples here, you mentioned [29:24] them, you know, Salesforce, [29:26] it is the system of record for your for [29:27] your sales uh side of things. And so, [29:31] it should stand even though more in a [29:33] headless capacity because everyone's [29:34] building their tooling on top of it. [29:37] Workday stands because it is [29:40] you know, paying people is a is a [29:42] regulated industry, you need to pay it [29:44] right, they're taking on that risk, [29:47] that's that's valuable. Um you know, on [29:50] the consumer side of things, a social [29:51] network is rather valuable, that's not [29:53] something you can buy with code and do [29:54] it yourself. So, [29:56] DoorDash is valuable, it is it is, you [29:58] know, [29:59] actual humans are bringing you your [30:00] food. So, I think [30:02] that that test has really stood the the [30:05] you know, it it it it it it's helped at [30:07] least me think really about well about [30:10] what software is going to survive in [30:12] this new age. [30:13] >> Yeah, so it's it's probably it's not [30:15] just the UI in the software, it's like [30:16] some sort of data or like services or [30:18] some some some other thing that's [30:19] included as part of it. [30:21] >> Exactly, yeah. [30:22] >> And and the good news is that like [30:24] cuz cuz you know, I don't want to talk [30:25] too much but like I don't think [30:26] anyone really enjoys using Workday the [30:28] UI, but [30:30] but I think all of these companies are [30:31] building, you know, APIs and, you know, [30:33] CLIs and MCPs. So, so you can just use [30:36] these tools to use like access all these [30:38] other tools without actually have to [30:39] click all the buttons. [30:41] >> [laughter] [30:42] >> So, I think that that helps a little [30:43] bit. It has been amazing to see that [30:44] that existing SaaS companies have leaned [30:47] into that. They're they're going [30:48] headless. And then well, now that agents [30:51] can operate browsers and and and [30:53] computers, [30:55] some of that restriction goes away as [30:56] well. They could they could reverse [30:57] engineer APIs or they could take actions [30:59] on your behalf as well. [31:01] Um [31:02] so yeah. Okay, cool. So so the last [31:04] topic I want to talk to you about is [31:05] just get some hot takes from you because [31:07] there's a lot there's a lot of chatter [31:08] right now about open source versus [31:10] closed source and app layer versus the [31:12] model layer and all that kind of stuff, [31:13] right? So I'll I'll tell you man, like [31:14] for a while I thought hey, you know, [31:16] like Anthropic and OpenAI, they're [31:18] building the model and the harness [31:19] together. They're going to they're going [31:20] to op- optimize the hell out of both of [31:22] both. And like as an app layer company, [31:23] you're kind of screwed because there [31:24] there's they just wanted to they want to [31:26] optimize themselves, right? They want to [31:27] take over the full stack. But but now [31:29] there's so many awesome models out [31:30] there. I I feel like app layer companies [31:32] are actually really, you know, strong [31:34] position. Do you have any thoughts on [31:35] this or cuz I I don't think Replit [31:37] actually lets you pick the model, right? [31:38] That that you use. [31:40] >> Yeah. Um well, I I love hot takes and I [31:42] I love this topic because I I I [31:44] you know, I'm I'm right there with you. [31:46] I think I think this is a great time to [31:48] be to be an app layer company. In a lot [31:50] of ways, the thesis that we've always [31:52] had and a lot of app layer companies [31:54] have had is coming true where models are [31:56] becoming commoditized. Um [31:58] and so more and more, you know, like if [32:01] you think about [32:03] about Replit and our sort of [32:04] differentiation from from these models [32:07] uh from these model labs [32:10] in some ways, yes, first of all, it's [32:12] the product differentiation of the of [32:14] the last mile that that that is super [32:16] important. It's you know, [32:18] cloud could giving you the code, but [32:19] hosting is where where where, you know, [32:21] and hosting in a reliable ways is is [32:23] what matters. It's the security, it's [32:25] the distribution, it's the [32:27] it's all of that extra packaging that [32:30] meets the user where they are and and [32:32] users pay for that. So I think well, [32:34] that's that's the that's the core thing [32:36] and that's what we're focused on a lot [32:37] too. The being model agnostic is a is a [32:40] feature itself because, you know, [32:43] like we were talking about with the in [32:45] the last 2 weeks with with Gwen and [32:47] Kimmy, the discussion has has fully come [32:49] to come to life of [32:52] the latest and greatest changes all the [32:54] time. And [32:57] And if you're an enterprise or even a [32:58] consumer, [32:59] I don't know if you want to be tied to [33:01] one put all your eggs in one basket and [33:03] and, you know, be tied to one model app. [33:06] Um it's it's it's the part about latest [33:07] and greatest. It's the part about cost [33:10] obviously as well, you know, token [33:11] maxing is coming to an end. And so, [33:13] do you really want to be beholden to one [33:15] company and their pricing? Really not. [33:17] Maybe different tasks. If Kimmy can do [33:19] it for 1/10, you let it do that. And [33:21] then lastly, resilience as well. It's uh [33:25] a single provider can go down. We've [33:26] seen that. Pricing changes all the time. [33:28] Policy changes. It's just hard to even [33:30] keep up and and I don't think [33:33] enterprises or consumers appreciate [33:34] that. It's it's almost great to have [33:36] that indirection layer, that app layer, [33:38] that that [33:40] is model agnostic and that helps you [33:42] orchestrate and run all your AI [33:44] workflows and and build yourself on top [33:46] of those. [33:47] >> From your point of view, is it true that [33:48] the like the Fables and the of of the [33:50] world are like do you use them more for [33:52] planning and and like more complicated [33:54] like I I I I guess re-architectural [33:56] work? And and and then for building [33:57] features or something, you just use like [33:59] a open-source or like how how do you [34:01] think about which model to use for what? [34:02] >> Yeah, I I think the the mental model of [34:05] of of a frontier model [34:07] like a like a fable, like a saw, like a [34:09] Kimmy 83 to do sort of the hard part and [34:12] then [34:14] an orchestrator that lets you delegate [34:16] to sub agents that are that are running [34:17] at our cheaper models [34:19] is definitely definitely the the the the [34:21] right approach. [34:22] >> And when you talk to enterprises, yeah, [34:23] so they they're not doing like token max [34:25] leaderboards anymore, right? They're [34:26] actually they're actually care about [34:27] cost cost now. [34:29] >> That is that is firmly firmly gone. [34:31] People care about it in the ROI, the R [34:34] part has become become very very [34:36] relevant. [34:37] >> Got it. That's actually really hard to [34:38] measure, dude. Cuz like ROI depends on [34:40] the function, depends on the team. [34:42] You know, cuz cuz yeah, this token cost [34:44] can easily go up. Yeah. [34:45] >> It's a funny meme you if you've seen of [34:47] like lighting a cigarette with a [34:48] flamethrower. [34:50] >> [laughter] [34:50] >> Yeah, yeah, yeah. I've seen that. I've [34:51] seen that. Yeah, yeah. All right. Well, [34:53] let's let's make some predictions. Let's [34:54] close with this really hot take. [34:56] >> It's great if you optimize for the [34:57] broader business ecosystem where [34:58] American businesses [35:00] get to use the latest and greatest at [35:02] good prices. That means AI advances. [35:05] Everyone gets to use AI. And and I think [35:08] I think that's going to do [35:10] well for us as a nation. [35:13] So I'm I'm very bullish of of the future [35:16] in that sense. [35:17] >> All right, well, well, thanks so much, [35:18] man. Like so where can people learn more [35:20] about Replit and about you? [35:22] >> Yeah, Replit [35:23] replit.com. Go try it out. It is [35:26] It's one of those products when I first [35:29] tried it, it was the aha moment that [35:30] took me out of [35:31] my parental leave that I was on to go [35:34] talk to Replit and go join him. [35:36] >> Nice. [35:37] >> moment is real. And and you know, [35:40] it's a great time to be an entrepreneur [35:42] or a creator. Go build something. For [35:44] me, you can find me on on Twitter, on [35:46] LinkedIn. Just my my first name and my [35:48] last name and [35:49] yeah. [35:50] >> Thanks to you and team for building a [35:51] great product. And yeah, I'm really [35:53] excited to build more stuff and turn [35:55] them into real businesses. Fingers [35:56] crossed. [35:56] >> Yeah, I I hope you make some money while [35:59] you're sleeping. And I think you're [36:00] You've got the audience down. Really [36:01] appreciate you having me, Peter. This [36:03] was This was a great conversation. [36:05] >> Thanks, Amjad.