# I Built a Company Where Every Employee Is AI

Channel: AI Growth Lab
Video: https://www.youtube.com/watch?v=zaupeVQxKkM
Duration: 13 min
Language: English
Words: 2278
Transcript page: https://viewrankai.com/tools/youtube-transcript/zaupeVQxKkM

---

[0:00] Last weekend, I gave my AI company a project and then I basically stopped working. My AI CEO planned the project. My AI developer wrote the code. Another AI reviewed it, found the bugs, sent it back for fixes, and repeated the process until everything [music] was clean. About 30 minutes later, the project was finished and ready to deploy. And here's the crazy part. I didn't write a single line of code. I've basically built a oneperson company where I'm the only human and every employee is AI. I have an AI CEO, developers, code reviewers, a chief of staff, content editors, and even graphic designers, all working together like an actual team. And I

[0:43] [music] didn't build some complicated custom AI system to make this happen. Most of the setup was done simply by talking to AI and telling it how I wanted my company to work. So, in this video, I'm going to show you exactly how I built this AI company from scratch, how all these AI agents work together,

[0:59] and how you can build your own version. You don't need to know how to code or things like that, because I think this is where AI gets really interesting. Instead of using AI as another tool you have to manage, you can start building a team that manages the work for you. Before we jump into the video, I spend a lot of time researching and figuring out these workflows. If you appreciate that, please hit the like button and subscribe

[1:21] for more AI and productivity tips. Right now, my whole AI company runs in chat GPT. That was a deliberate choice. A lot of tutorials and case studies I see use a stack of different apps. They use ChatgPT, Claude, OpenClaw, Hermise, and more. One task gets routed to this app, another task goes to a different app, and before long, you are managing the tools more than the work. I tried setups like that. Some of them were powerful, but the benefit I got was not worth the constant switching, fixing issues, etc. And honestly, isn't AI supposed to make our lives easier? So, I simplified everything. Chat GPT became

[2:02] the front door. I use it to capture ideas, think through decisions, and start the work. When a task needs access to my files, code or a proper review process, I use codecs and let separate agents handle the different roles. Instead of spreading myself across several tools, I built one team around a single workflow. The company framework I built for this team looks like this. At the very top is me, the chairperson responsible for setting the general direction, proposing ideas, and making the final decisions. Below the chairperson is chat GPT. I usually use the mobile app. It acts as my chief of staff. It captures my scattered

[2:39] thoughts, [music] turns them into clear briefs, and helps me decide what should be sent to the CEO. The CEO in the middle is played by codeex. It is responsible for overall company planning, taking my ideas to analyze and validate them and finally arranging their implementation. Under the CEO, there are two project departments. One

[2:58] is the vibe coding project department. It writes code and builds products. The other is the media content project department. Sounds quite fancy, doesn't it? But building it is actually very simple. So let's do it step by step. First I discussed this idea with chat GPT. I opened the desktop app, opened codecs,

[3:27] and shared all of my ideas there. Here's what I said. I want to build an AI agent team to manage my entire oneperson company. Currently, I have two business lines. One is doing vibe coding to develop some small projects and products. The other line is content creation for social media. Now, here comes the key part. I told it not to rush [music] into action. I said, "Search online first to see how others

[3:53] build such AI agent oneperson companies. Don't write any code or make any modifications yet. Just tell me what you find. Codex quickly found several existing case studies and gave me the links. I went and checked them out myself. They were pretty interesting. So, don't make your AI start from scratch. I will share the prompt in the video description. [music] I must give a quick reminder. This is not something you should copy verbatim. Everyone's business and habits are different. When building it, just have your AI adapt to

[4:23] your own ituation and you're good to go. For example, I'm doing [music] vibe coding and media content here, but you might be working on other projects, so modify it according to your real situation. You can also give chatgpt/claw this video link and say you want to set up something like this. It will start working. The frameworks from the case studies I read looked something like this. At the very top is a CEO. Below the CEO are coders, reviewers, as well as product managers and operations. Here I made a particularly critical design choice. I used two different AI models for coding and reviewing. Many of them also sent coding to one app and review

[5:01] to another. I understood the logic. A reviewer should be independent and critical. But I did not want to bounce between apps. I don't want to subscribe to that many services. So I kept the separation of roles but moved the handoff inside codeex. One agent writes the code. A separate reviewer agent gets a fresh brief, [music] checks the changes, runs tests, and sends back a list of problems. The coder fixes them and the reviewer checks again. This gives me the independent review I wanted without making me manage a chain of

[5:30] different apps. Once the framework is clear, the next step [music] is building my own team. You can't swallow an elephant in one bite. So I started with the product creation route. [music] I first set up my vibe coding project department. I established three roles. The first is CEO handled by codeex. It is mainly responsible for brainstorming [music] with me, pointing out loopholes or areas that need fixing in my ideas and then delegating the tasks. The second is the coder. It is also powered by codecs in the background. It is responsible for actual development and implementation following the CEO's [music] instructions. The third role is the reviewer. It is a separate codeex agent

[6:12] with fresh context and a clear review checklist. Its job is to audit the code, run tests, and find bugs, vulnerabilities, or weak assumptions. The workflow basically goes like this. After the coder finishes, the reviewer agent checks the code, runs a security review, and looks for bugs or weak assumptions. If it finds anything, it sends the work back with specific [music] fixes. Then the coder revises

[6:36] it, and the reviewer checks it again. They repeat that loop until [music] the reviewer is satisfied and the test pass. Only then does the result go back to the CEO. [music] Then the CEO reports back to me to plan the next steps. Actually, I could also add a product manager or operations role in between [music] to decide whether it can be released and deployed. But currently, I want to keep control over the quality of the product I'm making. So, I haven't handed the final release permission over to AI. I want to review it once more with my own human eyes. So by this point, my coding

[7:08] project department is all set up. Throughout the entire process, I didn't create a single new folder or write a single line of code. Once the setup was done, we moved to step three, verification. I needed to see whether this whole setup could actually run. The verification method is also very simple. Open Codeex [music] and then call out my CEO. I said, "We're starting work." And as soon as it came online, it did something I found remarkably smart. Before I even told him what projects I currently had on hand, it had already scanned the files on my computer [music] by itself. It found several vibe coding projects I had

[7:47] modified in the past 3 months. Then it asked me if any of them should enter the production pipeline. I chose a project. The CEO first went through the pros, cons, and areas needing improvement of this small project with me, then handed it over to the coder to write the code. Once the code was finished, it was

[8:06] handed to the separate reviewer agent. [music] The coder and reviewer went through several rounds of fixes and checks. When the reviewer was satisfied and the test passed, the result went back to the CEO who told me it was ready for my [music] final review. Throughout the whole time, I was just sitting in my chair doing nothing. After about half an hour, they had completed the entire project and all the bugs were ironed out. When I checked the app again, the window had just finished writing the final summary. [music] I reviewed it, saw no issues, and could then tell it to

[8:36] deploy. Step four was setting up a content creation department. [music] Following the exact same steps as before, I built a content creation project department. The CEO simply opened an additional content department under him. In this department, I divided up several roles. First is the editor, mainly responsible for drafting graphic posts and editing video scripts. [music] Next is the graphic designer, mainly responsible for illustrations, creating mini animations or charts for my videos, as well as designing my video thumbnails. With that, my oneperson

[9:12] company was basically fully established. Now, let's do a quick recap of this entire framework. At the very top is myself as president setting directions and frameworks. Next is chat GPT app acting as my chief of staff. It is where I dump scattered ideas, shape them into useful briefs and decide what needs attention [music] with the right

[9:34] connected services and scheduled tasks. It can also help me keep track of meetings, messages, and recurring updates. Then there is the CEO in codeex. When I sit down to work, I call it in to help set the direction for the day. It manages two departments. the vibe coding project department and the content creation department. [music] In theory, my company can run 24/7. I can rest while my entire company and creative team don't need to sleep at all. If there are unlimited tokens, the CEO can continuously assign tasks

[10:04] [music] and keep the employees working. Of course, I don't really have that much work or token budget right now. At most, when I'm busy with other tasks, I have them modify code for me. So the framework is in place, but how well each agent can actually perform still required some finetuning. For example, for the CEO, I directly hooked it up to a CEO skill I created from

[10:27] distilled biographies of some tech CEOs. So my current CEO has a brain installed that is knowledgeable about how top CEOs [music] perform. When it discusses product directions with me, it truly brings in first principles thinking, the five-step algorithm, [music] and a whole array of thinking frameworks to brainstorm with me. As another example, I have also created and installed various custom skills for my editor-inchief. Whenever the editor-inchief needs them, it can call them directly. I also have a graphic designer connected directly to a video cover creation skill. I will make these skills public very soon. Do you see the brilliance here? Before I built this team, I had to manually invoke each of

[11:09] these skills myself to do all these fragmented tasks. [music] The workload wasn't huge, but it was extremely fragmented. I had to stay glued to the screen watching it. Now, I hand all of this over to my AI company. As president, I just drop a single sentence. Help me revise this draft. My CEO then controls the quality, automatically delegating to different employees who invoke different skills to

[11:34] finish the job. Let me share a personal observation. The AI on your computer is actually a bit like slime mold. At first, everyone's looks the same, [music] just a tiny dot. But as you continuously use it, install different skills, and connect various MCPs, it gradually grows. Since everyone's focus is different, each person's AI will eventually grow into something completely unique. So, the team you build using my method today will also be one of a kind and understand [music] you best. Of course, to be completely honest, as of right now, my company is at a break even point. I have made some sales which went towards paying for tokens. I haven't

[12:17] made any profit yet. So, moving forward, my company's number one priority is just two words, [music] revenue generation. But even so, I still strongly recommend that everyone build their own AI agent team. Even if you don't want to start a one-person company, building a team to help manage your daily affairs is completely viable. That way, you don't need to use your human brain to memorize so many skills. You just hand the work to your company whose sole mission is to serve you well. On a broader level, every individual is essentially a company. Everything you do can be institutionalized. Someone manages, someone executes, and you only review

[12:54] the results. I think this might very well be a worthwhile direction for the future, liberating us from tedious

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt