Turning long videos into YouTube Shorts is a real business. You post a long video, and the repurposing starts: a freelancer clips the best moments by hand, or you feed it to a web tool that auto-detects highlights and exports vertical clips with captions. Either way, you pay, and you keep paying.
So I asked the obvious question. Why not build the tool myself, for cents, with no programming knowledge required?
Cheap AI models are not just toys. They can build software that replaces a paid service, and this shorts clipper is the proof. It transcribes a long video, lets me mark a range with in and out points, and exports a vertical 9:16 Short with burned-in captions. It cost about $0.65.
Building the clipper
I started with a short prompt. Here it is, verbatim:
yup, this directory is now empty - let’s build the project. A wxPython app, using venv, that can read a video, transcribe its text, and generate a youtube short clip (vertical) from a selected range, with automatic subtitle rendering
That prompt was basically everything the agent needed. The rest was just fixing bugs.
It sorted out the dependencies and built a basic UI. But before it touched the UI, it tested the backend first. It generated a test video with FFmpeg, then used ask vision, the tool I added to GritCode recently, to check that the output was actually correct. It even opened the generated video in its own UI to verify everything worked.
The agent can now see what it built. That still feels wild.
The bugs you only find by using it
The first version was close, but rough. The video played like a slideshow instead of actual playback. The in and out handles were tiny and hard to grab. And there was no way to preview the selected range before export, so I had to scrub around and hope for the best.
But the transcription worked, the captions generated, and the export worked. For a first shot, that’s not bad.
The fix round
So I went back to the AI and listed the bugs. It fixed the playback by swapping in MPV.
This is where software development has changed in the last few months. In 2025, I’d be researching libraries, checking APIs, and figuring out how to wire MPV into a Python app myself. Now the AI just tries things, decides to switch to MPV, plugs it in, and it’s done.
After that, playback worked and the handles were easy to grab. Then a final polish round: bigger subtitles, better styling, keyboard shortcuts. The agent loops on its own now. Improve, run the app, look at it with ask vision, iterate.
Cheap models, real savings
Compare that to paying a freelancer per video, or a monthly subscription to a clipping tool. This app does one job, the job I actually need, and after the build it costs nothing to run. And I didn’t write a single line of code myself.
You might say it’s nowhere near as powerful as a commercial clipper. That’s true. But almost nobody uses all the features of those tools either. That’s the point of bespoke software: build the part you need, and add features as you go. Want auto-reframe? Ask the AI. Want the captions in your brand colors? Ask the AI.
What would you build?
This is the question I keep coming back to. What small, boring, repetitive task are you currently paying a service or a freelancer to do? A clipper, an invoice generator, a lead scraper, a scheduler that reads your email?
A cheap model can probably build it for you, for cents, and you’ll own it.
If you want the full session you can import into GritCode, plus the sources, they’re on my Patreon. Build this one yourself, break it, improve it. And tell me what you’re going to build next.
