--- author: The MAD Podcast with Matt Turck date: '2026-03-31' guest: '' layout: post.njk source: https://www.youtube.com/watch?v=rX3CxlGm-00 speaker: The MAD Podcast with Matt Turck tags: - ai-agents - autonomous-development - startup-scaling - future-of-work title: AI 驱动的单人百亿公司：General Intelligence 公司 CTO 详解 summary: Andrew，General Intelligence 公司 CEO，介绍了如何利用 AI Agent 构建‘一人百亿公司’的愿景。重点展示了革命性的‘Co-founder CTO’产品，它能实现全自动化的软件工程流程，从需求到部署、测试和监控，极大地提升开发效率，并预示着未来公司运营模式的深刻变革。 insight: '' draft: true series: '' category: ai-ml area: tech-engineering project: [] people: [] companies_orgs: - General Intelligence Company products_models: - Co-founder CTO - Co-founder media_books: [] status: evergreen --- ### AI 赋能：从助理到百亿公司的飞跃 Andrew 介绍了他的背景，作为 **General Intelligence Company** 的 CEO，他怀揣着一个宏大的愿景：赋能**一人百亿公司（one-person billion-dollar company）**，即创建能够完全由 AI 自我运行的组织。他分享了公司创立的初衷和任务——实现这一愿景。 在深入探讨未来之前，Andrew 回顾了他早年的经历，包括 18 岁成为 CEO 并辍学创业。他指出，在不久的将来，**AI Agent**（AI 代理）将变得极其强大，能够执行几乎任何任务，这将深刻改变公司的运作模式。他设想了一个未来：一个公司可以由一个人来运营，而大量的垂直性 **AI Agent**（如客服、财务、营销）则会独立运作，由“协调者”或“管理者”Agent 统一调度。General Intelligence 公司将聚焦于这一**管理者 Agent**（manager agent）和人机交互界面。 他们推出的首款产品是 **Co-founder**，一个基于云的 AI 助手，拥有先进的记忆能力，能像用户一样起草邮件，并能跨计算机执行 Agent 任务。Co-founder 曾帮助超过 8000 家初创公司自动化运营，并在处理复杂 Agent 任务方面积累了宝贵经验。然而，Andrew 强调，一个强大的助手**并非一家公司**。由此，他们开始探索“什么是公司？”这一核心问题。他们提出了一个构成公司的基础框架：**创建产品（create a product）、销售产品（sell it）、支持产品（support it），最后是运营（operate）**。简单来说，能够创造软件产品，将其卖给用户并持续获得收入，从而得以生存，这就是一个完整的公司。这四个环节构成了 Andrew 作为创始人日常工作的核心。
Original English My name is Andrew. Uh, and I'm going to be talking about how we go from here to the oneperson billion-dollar company. Now, a little bit about ourselves. We founded the general intelligence company in New York about a year ago. I'm going to call it general intelligence because I only have so many words I can say. Our mission is to enable the oneperson billion dollar company. We want to build uh organizations that run themselves with AI. There's a number of ways we're going to get there. I'm going to talk about them, but first a little bit about myself. Um, this is my second company. Uh, I've been CEO since I was 18. I dropped out of college. About the world. Now, in a couple years, we're going to start seeing all of this stuff um come together. So, agents finally work. They can do basically anything you want them to do. What happens next? Well, we're going to start seeing companies change a little bit. And eventually we'll have one person that can run a company that's very wide and very I guess large in operations run by just one person and then below that will be a bunch of agents. Um vertical agents like customer support, finance, marketing and things like that will be run on on their own and then there's going to be coordinators or like managers above them and uh that's where we're focusing is the manager agents and the interface in which the people are going to use them. So think about us like the like the the middle management layer and then we want to use other people's agents for things on the bottom. We don't want to make every agent for everything. That wouldn't be possible. But let's talk about co-founder. So co-founder was our first product. We created it back in uh September. We launched on September 8th and uh it was like an assistant where you can go and you can run agent across a computer in the cloud. It's got state-of-the-art memory so it knows about you. It can draft emails. It really read like you. And so people were using it for all sorts of stuff. Um they were using it to make their uh email spam for their sales thing work. They were using it to look up candidates on LinkedIn. Some people just started giving it API keys which we told them not to do but they keep trying to do it. It was able to like automate different pieces uh of a company and people really liked it. So it was pretty successful. We've had over 8,000 startups use co-founder to automate their operations which uh was was pretty fun. Um, we learned a lot in that and we were able to just workbench on this state-of-the-art agent and get it to do really difficult things. An assistant is not a company, right? I got up here and told you we're going to make one person billion dollar companies. Well, something that writes your emails better is not a company. What is a company? This is the question we started asking ourselves. And we wanted to figure out how to go from co-founder to our vision of, you know, this this one person billion dollar company. And so we came up with this framework. Um this is for product based companies or services companies which are different but you create a product you sell it you support it and then you operate. So you know in very simple terms this is an entire company. If you can create something like a piece of software and then you can go and sell it to people that give you money and then you can make sure that they keep giving you money and then you can exist. That's basically what I do all day long as a founder is is these three things. These four things. Then we moved on to like how do we do what's next which is step one and that's creating a product and that's what I'm going to show you guys today is kind of some of our approaches there. Back in the summertime we used co-founder to make a version of fully automated feature development. This was back like two generations of models ago but we were able to set up these automations where let's say uh an email comes in with a bug report or a feature. We can pull in some data. We can make a report on that and create a linear ticket to to track the feature. Then we can use Slack to kick off Devon to go code a fix and then update the ticket when it's done. And then we were able to hack together this thing where we would generate a QA test list to see how how do you test this thing that you just got this report on and then actually use a browser agent to see if we could get it to work. And so this was a way of going from, hey, here's a bug report in an email all the way to it's now coded and merged and you have a browser that's using it. When we started looking at this, we started thinking about, okay, well, you you kind of need a management layer. So, you need a manager that looks over these things like your product agent or your your email agent or your your your coding agent to be able to actually work on these things, which is like, how do I make the product better? How do I get it to a system where there's no bugs in it and it just runs itself? And this wasn't really possible in the summertime. It's become more possible now. You can scale this all the way up. So, you know, if you have one agent that can manage another agent, you can have an agent on top that manages all of that. And this is kind of what we we call super optimization. And so here's how it would work today with a system that was able to have a manager agent as well as a couple of lower level agents doing things like code generation and browser testing. So you take claude to write your code, use browser base to run a browser and then you deploy it with something like Verscell and then you manage that all together to be able to develop products autonomously. This is brand new. You guys are the first people seeing this actually. Uh, our alpha comes out today. This is the first fully autonomous engineering department. It's all in the cloud and it allows you to build features, test them, deploy them, manage infrastructure, and run a software company autonomously. This is our new product called co-founder CTO. We've got a couple of other things coming out later that we'll play on this word, but you know, I have a co-founder and CTO. You might notice this doesn't really look like an assistant, right? This is kind of like a canvas. And you know what it does look like? It looks like a video game. And that's what we've developed it based on. We looked at couple games like Starcraft and Civilization. This has become a very popular thing to talk about over the last couple of weeks, but uh we were doing this back in December. And you can spin up an agent that will code a feature and then it will go through the entire life cycle of that feature and then uh merge it automatically. So, I'm going to take this one and I'm going to copy it and we're going to show it from scratch. So, you create a new task. You've got your repository that you can assign it to. So, I'm going to select super optimizers, which is our our current repository. And I'm going to say make the blur on the canvas when a node is selected a bit darker. This means white overlay instead of instead of white. We have a couple different ways of doing this. I'm not very good at specifying my tickets, but like my co-founder Obby, who's much smarter than I am, will write very long ones and it'll have many of these going all day. This is going to take a minute. It's agents running. It's going to go and it's going to start looking for repositories and then it's going to delegate to a coding agent and then that coding agent is going to actually code this feature. Now, it's run this coding agent. I believe it's going to run claude code. Um, and all this is in the cloud. So you don't have to have a computer locally. This is one of the things that we decided we didn't want to do. So I was trying to figure configure cursor cloud agents and I wasn't able to um because getting local development to work on on their cloud agents is just very broken. Okay, we've got cloud code running here. Um and we decided that since local development was really hard, we would just get rid of it. CTO doesn't use any local development at all. It uses these things called preview environments. Every time a coding agent is spun up, it will make a PR and it will send it to your repository and then every PR is hosted like your production and then it's put on on the web. So if you have loginins for that, your agent can then use the login there. There's no local development at all. It all happens in the cloud. Okay, so now we got cloud code running. Now it's going it's looking at different things. It's looking at colors and when it's done with this, it'll spin up a browser agent. We have it so that every time it's spin up, it spins up a browser. And so I'll be able to show you that right here. And this browser session all happens automatically. So the agent goes in, it logs in with your credentials. And you can see this is like a little version of your app that's running in the cloud. And so we have this browser agent go and test your feature. And this one determined that it actually wasn't able to ask the the completion. It said it wasn't wasn't good enough. So, it went back up to the browser agent and it told it or it went back up to there, told it to code something again and then it did the browser agent again and then it worked. Now, it's tested it. It merges it successfully and that is now in your app. Now, I don't have that much time left so I'm going to show you some of the other stuff as well. We've also added infrastructure monitoring. All of your infrastructure can be monitored by CTO as well. So we also monitor infrastructure and whenever something happens, so we've got Verscell alerts and Versell deployment failures, uh it will spin up an agent automatically and fix them so that you don't have to have a uh a person monitoring this infrastructure. And so you can now go and ye features into your app and they get merged automatically and you have another agent that makes sure that your app doesn't break. that that means you can just send a bunch of features to your app and develop at a very fast speed. You can also monitor the deployments here. Um spin up agents below that. But yeah, so there's a there's a quick demo of CTO. We'll have a much slicker, cooler looking version in a couple of weeks. Yeah. So it it plans and delegates, uses coding agents, uses Versell preview environments, and then watches the deployment, then writes a testing plan, uses a browser agent to test, merges, and checks the infrastructure. Yeah, like I mentioned, our team is 400% faster compared to before we launched this. So, we use this internally. Um, this number is actually out of date. We have an average of like 25 PRs per engineer per day. We've started spending more on tokens than on on salaries depending on the day. Sometimes we'll spend like I don't know like I think today we spent 4 grand on tokens on Opus tokens. Some days it'll be less. But this shows that we're starting to shift our human capital to intelligence. I just showed you how to make products but that's not a whole company, right? There's a couple other pieces of a company that you need. Now, some people disagree with us on that, but we said you need to sell the product, support the product, and operate. And we're not just making an engineering department. We're making everything. So, this is what we're doing in the next 6 months. Um, you'll be able to try out co-founder CTO on it's I don't know if we're going to make March 1st. We'll make like March 6th. Then you'll have customer support so you can build the product and support the product. And then we'll have a full CRO. uh revenue and sales stack by the summer. Um so yeah, that's that's everything for me. If you want to sign up for the alpha, this QR code right there, it'll give you a Google form. We'll get you a login. We have to only do a couple people at a time, but you can get on by the end of the week if you're really lucky. Yeah. Thanks everyone. [music]
### Co-founder CTO：全自动化工程部门的诞生 Andrew 接着指出，一个强大的助手**不是公司**，并引出了他们对“什么是公司”的探索。他们提出了一个由四个核心环节组成的框架：**创建产品（create a product）、销售产品（sell it）、支持产品（support it）**，以及**运营（operate）**。这个框架定义了一个完整的公司，而 **General Intelligence Company** 不仅限于构建工程部门，而是致力于实现完整的公司自动化。 他们最新推出的产品 **Co-founder CTO**，正是这一愿景的体现。它被描述为**首个全自动化的工程部门**，完全在云端运行，无需本地开发环境。Andrew 通过一个具体的演示展示了 **Co-founder CTO** 的强大能力：从一个简单的任务描述（“使画布上的节点被选中时变暗，意味着白色叠加而不是白色”），到选择存储库，**Co-founder CTO** 能够自主地分解任务，**委托给 coding agent（如 Claude Code）进行代码生成**。生成的代码通过**预览环境（preview environments）**自动创建 **Pull Request (PR)**，并部署到 Web 上，如同生产环境一样。然后，**Co-founder CTO** 会**生成测试计划**，并**使用 browser agent**（浏览器 Agent）模拟用户操作来测试新功能。一旦测试通过，代码会被**自动合并（merge）**。 此系统还集成了**基础设施监控**功能，能够自动检测并修复 Vercel 部署失败等问题。Andrew 强调，这种**自动化能力**极大地提升了开发速度（团队效率提升 400%），并促使人类资本向更智能、更具战略性的工作转移，甚至公司开始花费比薪资更多的费用在 AI Token 上。**Co-founder CTO** 被设计得像一个视频游戏界面，灵感来源于《星际争霸》和《文明》等游戏，它能够规划、委托、执行编码、测试、部署和监控整个软件生命周期，实现**自主产品开发**。
Original English And then we moved on to like how do we do what's next which is step one and that's creating a product and that's what I'm going to show you guys today is kind of some of our approaches there. Back in the summertime we used co-founder to make a version of fully automated feature development. This was back like two generations of models ago but we were able to set up these automations where let's say uh an email comes in with a bug report or a feature. We can pull in some data. We can make a report on that and create a linear ticket to to track the feature. Then we can use Slack to kick off Devon to go code a fix and then update the ticket when it's done. And then we were able to hack together this thing where we would generate a QA test list to see how how do you test this thing that you just got this report on and then actually use a browser agent to see if we could get it to work. And so this was a way of going from, hey, here's a bug report in an email all the way to it's now coded and merged and you have a browser that's using it. When we started looking at this, we started thinking about, okay, well, you you kind of need a management layer. So, you need a manager that looks over these things like your product agent or your your email agent or your your your coding agent to be able to actually work on these things, which is like, how do I make the product better? How do I get it to a system where there's no bugs in it and it just runs itself? And this wasn't really possible in the summertime. It's become more possible now. You can scale this all the way up. So, you know, if you have one agent that can manage another agent, you can have an agent on top that manages all of that. And this is kind of what we we call super optimization. And so here's how it would work today with a system that was able to have a manager agent as well as a couple of lower level agents doing things like code generation and browser testing. So you take claude to write your code, use browser base to run a browser and then you deploy it with something like Verscell and then you manage that all together to be able to develop products autonomously. This is brand new. You guys are the first people seeing this actually. Uh, our alpha comes out today. This is the first fully autonomous engineering department. It's all in the cloud and it allows you to build features, test them, deploy them, manage infrastructure, and run a software company autonomously. This is our new product called co-founder CTO. We've got a couple of other things coming out later that we'll play on this word, but you know, I have a co-founder and CTO. You might notice this doesn't really look like an assistant, right? This is kind of like a canvas. And you know what it does look like? It looks like a video game. And that's what we've developed it based on. We looked at couple games like Starcraft and Civilization. This has become a very popular thing to talk about over the last couple of weeks, but uh we were doing this back in December. And you can spin up an agent that will code a feature and then it will go through the entire life cycle of that feature and then uh merge it automatically. So, I'm going to take this one and I'm going to copy it and we're going to show it from scratch. So, you create a new task. You've got your repository that you can assign it to. So, I'm going to select super optimizers, which is our our current repository. And I'm going to say make the blur on the canvas when a node is selected a bit darker. This means white overlay instead of instead of white. We have a couple different ways of doing this. I'm not very good at specifying my tickets, but like my co-founder Obby, who's much smarter than I am, will write very long ones and it'll have many of these going all day. This is going to take a minute. It's agents running. It's going to go and it's going to start looking for repositories and then it's going to delegate to a coding agent and then that coding agent is going to actually code this feature. Now, it's run this coding agent. I believe it's going to run claude code. Um, and all this is in the cloud. So you don't have to have a computer locally. This is one of the things that we decided we didn't want to do. So I was trying to figure configure cursor cloud agents and I wasn't able to um because getting local development to work on on their cloud agents is just very broken. Okay, we've got cloud code running here. Um and we decided that since local development was really hard, we would just get rid of it. CTO doesn't use any local development at all. It uses these things called preview environments. Every time a coding agent is spun up, it will make a PR and it will send it to your repository and then every PR is hosted like your production and then it's put on on the web. So if you have loginins for that, your agent can then use the login there. There's no local development at all. It all happens in the cloud. Okay, so now we got cloud code running. Now it's going it's looking at different things. It's looking at colors and when it's done with this, it'll spin up a browser agent. We have it so that every time it's spin up, it spins up a browser. And so I'll be able to show you that right here. And this browser session all happens automatically. So the agent goes in, it logs in with your credentials. And you can see this is like a little version of your app that's running in the cloud. And so we have this browser agent go and test your feature. And this one determined that it actually wasn't able to ask the the completion. It said it wasn't wasn't good enough. So, it went back up to the browser agent and it told it or it went back up to there, told it to code something again and then it did the browser agent again and then it worked. Now, it's tested it. It merges it successfully and that is now in your app. Now, I don't have that much time left so I'm going to show you some of the other stuff as well. We've also added infrastructure monitoring. All of your infrastructure can be monitored by CTO as well. So we also monitor infrastructure and whenever something happens, so we've got Verscell alerts and Versell deployment failures, uh it will spin up an agent automatically and fix them so that you don't have to have a uh a person monitoring this infrastructure. And so you can now go and ye features into your app and they get merged automatically and you have another agent that makes sure that your app doesn't break. that that means you can just send a bunch of features to your app and develop at a very fast speed. You can also monitor the deployments here. Um spin up agents below that. But yeah, so there's a there's a quick demo of CTO. We'll have a much slicker, cooler looking version in a couple of weeks. Yeah. So it it plans and delegates, uses coding agents, uses Versell preview environments, and then watches the deployment, then writes a testing plan, uses a browser agent to test, merges, and checks the infrastructure. Yeah, like I mentioned, our team is 400% faster compared to before we launched this. So, we use this internally. Um, this number is actually out of date. We have an average of like 25 PRs per engineer per day. We've started spending more on tokens than on on salaries depending on the day. Sometimes we'll spend like I don't know like I think today we spent 4 grand on tokens on Opus tokens. Some days it'll be less. But this shows that we're starting to shift our human capital to intelligence. I just showed you how to make products but that's not a whole company, right? There's a couple other pieces of a company that you need. Now, some people disagree with us on that, but we said you need to sell the product, support the product, and operate. And we're not just making an engineering department. We're making everything. So, this is what we're doing in the next 6 months. Um, you'll be able to try out co-founder CTO on it's I don't know if we're going to make March 1st. We'll make like March 6th. Then you'll have customer support so you can build the product and support the product. And then we'll have a full CRO. uh revenue and sales stack by the summer. Um so yeah, that's that's everything for me. If you want to sign up for the alpha, this QR code right there, it'll give you a Google form. We'll get you a login. We have to only do a couple people at a time, but you can get on by the end of the week if you're really lucky. Yeah. Thanks everyone. [music]
### 未来展望：构建完整的自动化公司生态 Andrew 总结道，**Co-founder CTO** 代表了**全自动化工程部门**的实现，它能够处理从需求、编码、测试、部署到基础设施监控的整个软件生命周期。然而，他再次强调，这仅仅是构建一个完整公司的**一个重要环节**。公司由**创建产品（create a product）、销售产品（sell it）、支持产品（support it）**以及**运营（operate）**这四个核心部分组成。 展望未来六个月，General Intelligence Company 的计划是扩展其自动化能力，覆盖公司的其他关键领域。他们将推出**客户支持**功能，使得公司能够构建产品并同时获得支持。随后，他们还将构建一个完整的 **CRO（Chief Revenue Officer）**，即**营收和销售技术栈**，从而实现公司运营的全面自动化。Andrew 鼓励观众注册 **Co-founder CTO** 的 Alpha 测试，并提供了获取登录信息的二维码。他表示，尽管目前只能逐步开放给少数用户，但通过努力，用户有望在不久的将来体验到这一革命性的自动化工具。
Original English And then we decided we didn't want to do. So I was trying to figure configure cursor cloud agents and I wasn't able to um because getting local development to work on on their cloud agents is just very broken. Okay, we've got cloud code running here. Um and we decided that since local development was really hard, we would just get rid of it. CTO doesn't use any local development at all. It uses these things called preview environments. Every time a coding agent is spun up, it will make a PR and it will send it to your repository and then every PR is hosted like your production and then it's put on on the web. So if you have loginins for that, your agent can then use the login there. There's no local development at all. It all happens in the cloud. Okay, so now we got cloud code running. Now it's going it's looking at different things. It's looking at colors and when it's done with this, it'll spin up a browser agent. We have it so that every time it's spin up, it spins up a browser. And so I'll be able to show you that right here. And this browser session all happens automatically. So the agent goes in, it logs in with your credentials. And you can see this is like a little version of your app that's running in the cloud. And so we have this browser agent go and test your feature. And this one determined that it actually wasn't able to ask the the completion. It said it wasn't wasn't good enough. So, it went back up to the browser agent and it told it or it went back up to there, told it to code something again and then it did the browser agent again and then it worked. Now, it's tested it. It merges it successfully and that is now in your app. Now, I don't have that much time left so I'm going to show you some of the other stuff as well. We've also added infrastructure monitoring. All of your infrastructure can be monitored by CTO as well. So we also monitor infrastructure and whenever something happens, so we've got Verscell alerts and Versell deployment failures, uh it will spin up an agent automatically and fix them so that you don't have to have a uh a person monitoring this infrastructure. And so you can now go and ye features into your app and they get merged automatically and you have another agent that makes sure that your app doesn't break. that that means you can just send a bunch of features to your app and develop at a very fast speed. You can also monitor the deployments here. Um spin up agents below that. But yeah, so there's a there's a quick demo of CTO. We'll have a much slicker, cooler looking version in a couple of weeks. Yeah. So it it plans and delegates, uses coding agents, uses Versell preview environments, and then watches the deployment, then writes a testing plan, uses a browser agent to test, merges, and checks the infrastructure. Yeah, like I mentioned, our team is 400% faster compared to before we launched this. So, we use this internally. Um, this number is actually out of date. We have an average of like 25 PRs per engineer per day. We've started spending more on tokens than on on salaries depending on the day. Sometimes we'll spend like I don't know like I think today we spent 4 grand on tokens on Opus tokens. Some days it'll be less. But this shows that we're starting to shift our human capital to intelligence. I just showed you how to make products but that's not a whole company, right? There's a couple other pieces of a company that you need. Now, some people disagree with us on that, but we said you need to sell the product, support the product, and operate. And we're not just making an engineering department. We're making everything. So, this is what we're doing in the next 6 months. Um, you'll be able to try out co-founder CTO on it's I don't know if we're going to make March 1st. We'll make like March 6th. Then you'll have customer support so you can build the product and support the product. And then we'll have a full CRO. uh revenue and sales stack by the summer. Um so yeah, that's that's everything for me. If you want to sign up for the alpha, this QR code right there, it'll give you a Google form. We'll get you a login. We have to only do a couple people at a time, but you can get on by the end of the week if you're really lucky. Yeah. Thanks everyone. [music]