Status: running. Backlog cleared, new videos handled automatically each week.
The problem
My wife Fabiola and I run the Nomads Amor YouTube channel, documenting full-time RV life. By early this year the channel had 55 videos and the website had almost nothing on it.
Every one of those videos could have been an article. Each had a transcript, a story, and search terms people actually type. Turning a video into a written piece is an hour or two of work, and I was never going to sit down and do that 55 times.
That is the shape of task worth handing to an agent: well defined, repetitive, and stalled for years because no human was going to grind through it.
How it works
An agent pulls the transcript for a video, drafts an article in my voice from a style guide I wrote, converts the prose into page components, and posts it to the site as a draft through the Swiftrics API. I read it, fix what needs fixing, and publish.
The part I like most is the images. A wall of text about an RV repair is dull without pictures, so the workflow matches sections of the draft against transcript timestamps, seeks to those moments in the video file, and pulls frames to illustrate the article. Getting it to choose a usable frame instead of a motion-blurred one took some iteration.
All 55 videos went through that loop. New ones do not pile up either: a weekly job checks the channel feed, and when a new video appears the same pipeline drafts and posts the article. The backlog problem became a maintenance problem.
The side effect
I own the whole stack here, from the Swiftrics codebase down to the API the agents call. Pointing an agent at your own API and telling it to create real pages over and over surfaces things a human tester never would. Odd component nesting cases. Whitespace handling that quietly broke inline links. Operations I assumed were covered and were not.
Because the same person runs the content workflow and maintains the platform, the loop is short. The agent hits a gap in the morning, I get a report of exactly what it tried and what came back, and the fix ships the same day.
An agent does not work around a rough edge the way a person does. It hits the edge, fails, and tells you. If you build a product with an API, making an agent a real user of it is the most honest testing I have found.
Where the line is
Everything the pipeline creates lands as a draft. A human decision pushes it live. That is deliberate, not a technical limitation. I have caught drafts that got a detail wrong or flattened a story into something generic, and the only reason those never went public is that review is mandatory.
The full write-up
More on where agents fit in my work and where they stop: How I actually use Claude Code and Cowork.