Transcription
My dear friends, behold and marvel, this is my AI company. Specifically, as employees, we have the CEO, who runs on Codex, a Content Pipeline Engineer also on Codex, a LinkedIn Editor via Crow Code, and Amadeus, an OpenCA Agent. And this is not the same Open Craw Agent as from my last videos, but the very same. Complete with memory and all already set up workflows. And the software I'm showing here is called Paperclip. The new Kit on the Block. That's right. We have another new, highly hyped AI tool again. Barely a month old and again an open-source project. Kind of great, isn't it? Thanks to the AI revolution, we are simultaneously experiencing an open-source renaissance. New community software solutions are literally sprouting from the ground because a team of developers is suddenly no longer a luxury good. And Paperclip showcases this particularly well. Open Source Orchestration for Zero Human Companies, it says here. Simply hire AI employees, automate everything, and the business takes care of itself. Big promises, these are. Fabulous reviews are here. Many see it as the next logical step. OpenC was the AI employee. Paperclip is the AI company. The idea is, you define the goals that should be achieved, and a team of AI handles the rest. What kind of AI? Doesn't matter. Claud, Codex, Curser, your existing Open Claw Agent, or another custom solution here. That's nice. If it can receive a heartbeat, it's hired. And Paperclip then promises to be the glue that holds everything together. Projects, goals, tasks and tickets, overviews of which agent causes how much cost, and much more. And I must admit, these are indeed problems that resonate with me. My Open Claw Agent, for example, can theoretically do quite a lot. In the last video, I demonstrated how it independently programmed a new platform that is already generating initial revenue. It takes a lot of real work off my hands, but I also have many different projects running via a Telegram chat, and I myself hardly know anymore what my AI agent is even working on or where my input is needed again. Paperclip, at least in its initial approach, is therefore quite interesting. I would say, folks, let's dive in. Let's take a look at this, my dears. Let's install Paperclip on a server. It would be a shame if our great AI company went to sleep every time I closed my laptop. Well, which provider should we use? Of course, the best provider. Let's even look. Google. Paperclip Hostinger, they are already on board, of course. Landing pages, YouTube videos, entire tutorial articles, I will link everything for you in the description. They are simply very good at it. You have to give them that, riding this open-source wave and making the installation of the tools simply particularly easy. You can get a server via my link on which Paperclip will then be automatically deployed. Of course, as always, voucher code Niklas for 10% discount, or if you already have a server at Hostinger, just go into the catalog, search for Paperclip, and select it once. Admin email address. We don't need API keys, we can handle that better, and deploy. That's it. Paperclip is already installed on our server. What's missing, of course, is the actual AI. Paperclip, as I said, is just the orchestration layer. We still need an agent, like a Cloud Code or a Codex. But you know what? I have a certain suspicion. Let's go to the terminal. On this beautiful Ubuntu server, Codex is probably not installed. Or is it? Hostinger has created its own Docker container for Paperclip. We find the container name here. Copy it once and then into the container. And here we look again. Codex is indeed already up and running. Just like Clord. Yes, probably too. Great, greetings go out to Hostinger. They have already set up Codex and Cloud for us in the Paperclip container. We just need to log into our existing ChatGPT account, for example. To do this, simply install Codex locally on your own computer, log in there, and then, watch out, we copy our local OTH file to this server directly into the Docker container. Folks, what am I doing here? Root, it should be called. There we have it. But Root is also an important keyword. Because if we look at the server again, the ORD file suddenly belongs to the Root user. We will change that again. Wonderful. With that, we are ready to create our first AI company. So, we go into the Docker Manager in the projects, click on Papercript, open it. We log in with the email and password that we set in Hostinger earlier, and now comes the fun part. We think of a company name. Productivity Labs. Does that sound too much like Ali Abdal? We'll leave it as is. And we are a productivity software company that builds the tools of tomorrow. What is that supposed to mean? Next. The first employee is the CEO. He runs on Codex, and the first task is already pre-filled here. We could change it. I'll leave it as is. Sounds reasonable. Hire a first software developer and make a plan. Let's go. Productivity Labs, ready to launch. And here we go, folks. Here's the CEO. Immediately to work. Not a second is lost. Hire your first engineer and create a hiring plan. Let's take a look around. We see we have an agent, the CEO, who is currently working on something. There is a task list, and we have the first task. We can also look at the instructions, and this is very Open Cla inspired. Agents, Heardbeats, Soul, and Tools, such Markdown files are quite familiar to us. Only that here it's no longer one agent, but many. And speaking of which, I see a CTO has just joined. This agent is pending board approval. We are no longer the CEO, we are the Board. We are the board of directors of the company. Here in our inbox, we also have a TAS accordingly. We can approve it once, and then the CEO will continue. In the orchestrator, we see we have hired a new employee. The AI models for these employees are still the same Cloud, Codex, and the like. What's new is that we don't chat with the AI, but everything happens via tickets and comments. I'm getting a lot of notifications like in a real software company. So I could create new projects here or create a new issue in this impressive task list at any time. I can also directly assign who should receive this task. Quite an interesting approach. I've already tried it for my own business and thought, maybe these AI agents can help me recycle my YouTube videos into shorts and LinkedIn posts. And because I've already worked a lot with an Open Claw Agent named Amandus regarding my YouTube channel, I thought it would be cool to bring exactly this agent with all his existing knowledge and his tools that I've already set up into my new team. That was actually relatively easy. I created a new Docker network on my virtual server, integrated the network into Paperclip and OpenC, and then I tasked the CEO with inviting Amadeus to the team. This is generally how Paperclip is configured. I have to get used to it a bit first. You don't go into settings, but you simply tell the CEO everything you want, or you create an issue. That's also somewhat unusual that there's no chat with the CEO. Although, watch out, Papercit, I saw something there. Yes, CEO Chat also seems to be on the roadmap. It's really still a very, very young tool. I will test it myself in the coming weeks and then report back to you whether it has really made me more productive. That's a bit of the catch with these hyped AI tools. My first impression of the UI is very positive, and it's simply fun. I sit there and marvel and watch what my AI employees are up to. But that, of course, carries the great danger of pseudo-productivity, because the question at the end is not how cool it looked or how much fun it was, but what was actually achieved with it. What is the real added value that was generated, and was it really easier and faster with Paperclip than, for example, with a Crowd Code agent? It's a big question in general whether it's efficient to link multiple AIs in such a company structure, as we do with humans. Skeptics will probably say that such a hierarchy, CEO, CTO, manager, employee, is made for humans and might not be necessary for AI at all. Perhaps a single LM that does everything would work just as well. My first impression, however, is that I, being a human, keep a better overview with Paperclip. With OpenCW, I had many different projects in a Telegram chat. With PayperGP, I have everything clearly structured by goals, projects, and issues. This has also actually reduced the threshold for me to finally tackle long-postponed projects, such as cutting short videos from my longform videos. Whether Paperclip is ideal for exactly that is another question. I imagined it like this: I hire an AI video editor who cuts the videos, and a social media manager posts them, and so on. But watch out, I'll show you. In fact, the task has landed in my inbox again with an NN workflow. My wise CEO has decided that it actually makes much more sense if he gives me a Nat Workflow that talks to Opus AI, and I should please just copy and paste it. Yes, thanks for the tip. After all, NLN is also a great tool, for which I have already made several tutorials, and which you can also install at any time with one click via the catalog on the Hostinger VPS. Well, that's probably the better approach, and Paperclip is probably particularly well suited for software development. The examples on the website, they always talk about AI Software Engineers. Makes sense, LMs can program incredibly well, especially if they are repeatedly triggered by an agentic loop. See my last video with OpenClow and Codex. Paperclip goes a step further there. Well, maybe soon there will be the first productivity app from my new AI software company available in the app store, or there will be an update video saying it didn't work out so well after all. Stay tuned, I will report. Interesting new open-source tool, definitely. Try it out yourself now, preferably with the Hostinger VPS, link in the description and voucher code Niklas. Here you will find my last video. There I show how I programmed an app with OpenC, Codex, and Telegram that generates real revenue. An app whose further development will perhaps soon happen in Paperclip. Let's see. If you want me to test more tools for you, write it in the comments.