AI coding keeps getting better and better. Cursor came first and it was basically VS Code with AI features added on top. Then we got AI in our terminals. Now Claude Code is one of the best coding agents out there. But what if I told you that today I found something truly amazing? Something that makes Claude Code work all by itself. It creates a whole team. And this team doesn't work one thing at a time. They all work together at the same time. The best part, you don't need to monitor anything because it's a truly autonomous system that just works. Before I show you the workflow, there are two key concepts you need to understand since many people won't be familiar with them. By understanding these, you'll see why this workflow is so amazing and how it defines the way AI agents may work in the future. The first is T-Mox, which stands for terminal multiplexer. It's a program that lets you create multiple terminal sessions inside one window. What this means is that when an agent like Claude Code is running in a terminal, it can spawn additional terminals within the same session and assign tasks to them. One instance of Cloud Code can effectively control and manage as many Claude Code instances as it wants. The best part is that T-Max sessions persist in memory, so when you return to them, you're exactly where you left off. This is incredibly useful for maintaining context. The second concept is terminal scheduling. This allows you to assign tasks to agents with specific timing parameters. You can instruct them to perform certain actions at designated times, then automatically proceed to the next task. There's no need for constant monitoring. The agents simply follow the schedule you've created, completing everything automatically. That's the power of scheduling. It transforms your terminal into a self-running system. Over on the AI Labs Discord community, we're hosting our first ever hackathon from July 22nd to July 28th. Submit your most interesting builds and projects, and the top five submissions will be featured in one of our YouTube videos. You can join by clicking the link in the pinned comment below. And if you're enjoying the content so far, make sure to hit that subscribe button so you don't miss what's coming next. Before we start, navigate to the directory or folder where you want the system to be set up. Once you're there, go to the GitHub repository linked in the description below. This repository is called T-Max Orchestrator, and you'll need to copy the link from there. Go back to your terminal and type git clone followed by the link. This command will clone the repository to your local machine. After cloning, move into the new directory called T-Max Orchestrator. And once inside, you'll run the setup commands shown here. You don't need to copy these commands from the video. Later, I'll show you how to get them in a structured format for easy copy and paste. Running the setup executes scripts that enable the agentic workflow. This step makes all necessary files executable. Once the setup is complete, you need to start a new T-mok session. Since I already have one running, I'll name this one my second agent. This will initialize a fresh session for you. Before you start working with the system, you need to fix two things. Since this project is still in development and relatively new, not everything has been finalized. There are two errors in the repository that need correction. After fixing these, the system should run properly. Both fixes I'm about to explain require editing files inside the framework. This will make more sense when I show you the demo. These are minor changes that you can easily make yourself. First, when you're inside the T-Mox session, type out Claude to initialize Claude code. You'll then give it a specific prompt that explains how the orchestrator works. There are two checks to ensure everything is functioning correctly. The first check verifies that Claude can control different windows using T-Mox. The second check confirms the scheduling system is working. When I run this prompt, it passes successfully. However, when you run it, the check will fail because the author has hard-coded paths with their own username in the configuration files. Don't worry about this failure. Claude will automatically detect the issue and fix these paths to work with your system. You won't need to copy this prompt from the screen either. Moving on to the second fix. You need to ask Claude code to edit the files again. This time, tell it to look through all the commands and replace the claude command with the claude command using the dangerously skip permissions flag. This change is necessary because the original framework we cloned doesn't include this feature. As you can see, there are three terminals that one main agent is controlling. This is the key benefit of using the T-Mox system. It allows the main agent you just set up to automatically create new terminals within your main terminal window. This is my main terminal and it has created three sub terminals inside. While I'm showing you three, there are actually eight terminals running here. Normally, when you give Claude a command, it asks for your permission. Even with autoaccept edits enabled, Claude still requires approval for certain commands. With the dangerous skip permissions flag, Claude runs all commands automatically. I've used this flag many times without any system issues. You need this setting so all the different terminals can run Claude commands without requiring you to open each one and approve every action. I learned this the hard way. My first time using the system, I started it but noticed it was stuck. Even after 15 minutes with the scheduling system, everything remained frozen at the first phase. This happened because all the terminals were waiting for my approval. You need to enable this flag. So the system can execute all commands independently without waiting for your input. You've seen the multiple terminal setup, but there's another crucial part that requires your attention. This is where you'll spend most of your time, and it's what makes the entire system worthwhile. I've opened the project directory and cursor to visually explain the structure. You can see our T-Mox orchestrator folder that we cloned. Outside of that, I have an app folder containing two subfolders, spec and task manager. The spec folder contains the documentation that the T-Max Orchestrator uses as its blueprint. The task manager folder is where the actual app gets built. The spec folder is critical. It contains the main spec, integration spec, and separate front-end and backend specs. These files provide guidelines that tell the orchestrator exactly what app to build. The main spec includes timing for each development phase. Everything is timed and Claude Code must complete each phase within its allocated time frame. I've prepared a template you can use to create your specs step by step. It includes default requirements for a full stack web app, which you can modify based on your needs. If you're building something different, make sure to review and adjust the specifications accordingly. Once your spec is ready, you'll use a specific prompt to start cla working on the project. After this point, everything runs autonomously. Here's what you need to know about the prompt. You must provide the full path to your spec folder starting from the home directory. Don't use relative paths. Next, specify what teams to create. There's a front-end team and a back-end team, each with a project manager and developer. You can add more teams using the same format. For example, to add authentication, you type create an O team and follow the same pattern. The system builds according to your spec and checks on each team every 15 minutes to ensure tasks are completed on schedule. It also commits changes regularly, creating restore points in case anything goes wrong. Version control is built into the workflow. Let me demonstrate my setup. I've reached my Claude usage limit, so I can't show live coding, but I can walk through how I started the workflow. After completing the two fixes and creating the T-Max session, initialize Claude again using the dangerous permissions flag. This launches Claude with the necessary permissions. Then provide your prompt and the system will begin building everything automatically. Let me show you the terminals it opens and how it manages the coding process across them. This is the start of the project. I initialized it exactly as I showed you. To be safe, I ran the check command mentioned at the beginning. Since I had already fixed the hard-coded paths, it confirmed the setup was successful. It recognized the first agent window and verified the scheduling was correct. I provided the full stack app specification and you can see it adding what I wanted to build just as I demonstrated earlier. After that, it proceeded automatically. I also included a UI implementation reference in the front-end spec. It knows to check the UI reference folder. Inside I have a design image to replicate and a UI implementation.m MD file. This implementation plan provides exact steps using Shad CN components with the Shad CN MCP. It's an amazing workflow that I'll link in the description if you're interested. The system understood my project requirements. It needed to deploy the front-end team, then the backend team, set up their development phases, and schedule 15minute checkups. It completed all these tasks. You can see it briefing both teams using the send claude message script. It sent instructions to the front-end team's project manager, then provided them with the project specifications from the spec folder. It repeated this process for the backend team, giving them their schedule and updating the to-do list. It marked completed tasks and moved to the next phase. The system informed me that both teams were actively working. It provided a status update and scheduled the next check detailing what each team was doing and their current progress. Both teams were successfully deployed and working on phase 1. After 15 minutes, the orchestrated check-in report showed both teams were ahead of schedule. Phase 1 was complete with bonus features added. The 15-minute update confirmed phase 1 completion with several components ready. It updated the to-do list and asked permission to proceed with phase 2. You could configure the maind file to proceed through all phases automatically without asking. However, I prefer step-by-step progression for review at each checkpoint. Looking at the actual setup, you can see our first agent at the top. Below that is task front end with three windows and task backend also with three windows. Both teams work simultaneously. Each has one window for the project manager, another for the developer, and a third for the server that runs tests and checks for errors. Everything operates in a closed loop. These agents work and develop independently. It's like having multiple engineers working simultaneously, each in their own terminal. This demonstrates the true power that Demox brings to your development workflow. At the bottom, you can see the agent we created at the start, which I named the second agent. To understand the workflow and get all the commands, you can head over to the GitHub repository. However, it can be confusing and I've made several modifications myself. Instead, I recommend using git ingest. Simply take the GitHub URL and replace github.com with gitingest.com. This will summarize the entire repository into AI readable text. Copy that summary and paste it into Claude or chat GPT. From my experience, Claude provides better results for this type of task. Ask it to explain the workflow and installation process step by step, and it will guide you through everything, including all the necessary commands. If the commands don't appear exactly as expected, don't worry. I'll include all of them in the description below for reference. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next one.