AI-PULSE: a YouTube channel that publishes itself

Anyone can make one video. The hard part of a daily channel is the thousandth day, and the reason most stop is that the format costs an hour a day of human attention. So the design goal was never 'make a good video' — it was 'remove the human from the loop without lowering the floor'.

The whole pipeline is a scheduled GitHub Actions workflow. On a cron trigger it selects the AI tool trending that day, assembles the long-form cut and the Short from the same source material, and publishes both through the YouTube Data API.

Running it inside CI minutes rather than on a box is the decision that makes it sustainable. There is nothing to keep alive, nothing to patch, nothing to pay for, and the run history is the audit log.

The failure mode of an unattended publisher is not that it stops. It is that it keeps going and publishes something broken. So the pipeline has its own tests, and they run before the publish step. A bad run shows up as a red CI run and no upload, which is the correct outcome — silence is cheaper than a wrong video.

It is a small system, but it is the shape of most of the work I care about now: define the behaviour precisely, let the machine do the repetitive part, and put a test in front of the step that is expensive to get wrong.