The video showcases a project where an AI agent, powered by Claude Code, autonomously manages multiple coding terminals via tmux and streams its real-time coding process on Twitch, allowing viewers to suggest and queue projects through chat. The creator demonstrates the system’s workflow, highlights an Nvidia GTC 2026 giveaway, and invites viewers to interact with the AI agent live and participate in the event.
In this video, the creator demonstrates a project where an AI agent, powered by Claude Code, autonomously controls multiple nested coding terminals and streams the process live on Twitch. The main goal is to see if the AI can run tasks in parallel across two terminals—such as building a Snake game in one while researching pathfinding algorithms in the other. The setup uses tmux on a Mac Mini to manage multiple terminal sessions, allowing the AI to issue instructions to each terminal independently. The Twitch chat is integrated so viewers can suggest coding projects, which the AI then queues and executes.
The streaming component is handled using FFmpeg, bypassing the typical browser-based Twitch streaming setup. The creator adds custom overlays, background music, and visual effects like chromatic vibrations and scan lines to enhance the stream’s appearance. The chat is monitored in real time, and messages are fed into the AI agent, which parses them for project requests. These requests are managed in a queue, and the AI agent autonomously starts new projects or iterates on existing ones based on chat input.
A significant portion of the video is dedicated to showcasing the system’s workflow. The creator runs a dry test locally, demonstrating how the AI agent spins up multiple tmux terminals, receives project prompts, writes code, and switches between terminals for different tasks (such as coding and testing). The process is dynamic, with the AI agent closing and reopening browser windows as needed, iterating on code, and responding to chat suggestions. The system is designed to be interactive, allowing viewers to directly influence what the AI builds in real time.
The video also features a major giveaway in collaboration with Nvidia, promoting the upcoming Nvidia GTC 2026 conference. The creator highlights several sessions of interest at the conference, encourages viewers to register (virtually or in person), and explains how to enter the giveaway for an Nvidia GTX Spark. The giveaway is exclusive to virtual attendees who watch at least one session (excluding the keynote), and all relevant links are provided in the video description.
In the final demonstration, the creator streams the AI agent live on Twitch, showing how chat suggestions are processed and executed. For example, a viewer requests a Snake game in C++ with a GUI, and the AI agent begins building it while simultaneously researching pathfinding algorithms in another terminal. The system successfully runs both tasks in parallel, iterates on the code based on feedback, and displays the results live. The creator concludes by inviting viewers to join the live stream, interact with the AI agent, and participate in the Nvidia giveaway, emphasizing the project’s potential for further development and community engagement.