A new video from The Mitchery, our own YouTube channel, puts an AI coding agent to a public test: make a working video editor from scratch, live, in one sitting. The video runs a long single session and was published in 2026.
The idea is simple. The person leading the session describes what the app should do. Claude Code, Anthropic’s AI coding agent, writes the code. No part of the process is arranged around a safe result.
What Gets Made
The aim is a real video editing app, not a small toy. The description names three parts the team works toward: timeline editing, trimming pieces of video, and getting finished video out of the app. These are the plain things anyone who has used editing software would expect.
There is no ready-made project sitting behind the scenes. There are no tricks used to smooth over rough patches. The description states plainly that whatever breaks, breaks on camera.
That matters. Rather than a finished, polished piece, the video is offered as an open test of whether Claude Code can make working software under real conditions, with the failures left in.
Why It Matters
The whole session answers a question that has been going around: is AI coding just talk, or can it truly deliver working software? The description calls the episode a direct test of that claim.
That question runs through the entire session. Nothing here is put together as a lesson or a smooth guided tour. It is a live attempt, with the result unknown to the audience.
Viewers watch the exchange between instruction and result as it happens. The person leading the session describes a part of the app. Claude Code writes code to match it. The two check together whether it works.
Anyone curious about AI coding tools has a clear reason to watch.
So does anyone who has wondered whether an agent like Claude Code can go beyond small pieces of code and truly put together an app with many working parts: a timeline, trimming, and a way to get the finished video out.
People weighing whether to bring AI coding into their own work may find the session useful as a true test done in the open, not a piece of advertising. The lack of any safety net is the point. Nothing is set up in advance to guarantee things go without trouble.
Viewers are also asked to help shape what comes next. The description asks people to drop a comment naming a feature they would want in a video editor, with the chance that it gets made live in a later episode.
The episode does not promise a pure success story. It promises an honest one. Whether the editor works, breaks, or lands somewhere between is left for the viewer to judge.
For a channel built around testing tools out in the open, this fits well: a plain idea, a clear list of parts, and a long, continued look at the process. The Clay Tribune will be watching to see whether Claude Code can turn a spoken list of features into a working app, and what that says about where AI coding tools stand now.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.