AI Build Log #001 Question: Was the UI actually the bottleneck?
I wanted my videos to look better. That's where this started.
I run two YouTube channels that turn research into explainer videos, built with an editor I made myself. Watching a video from another AI-made channel, mine suddenly looked flat by comparison.
So I opened my editor and thought: I need to make the visuals more dynamic.
If I'd asked AI to just do that, it probably would have built me fancier animations. Instead I asked why first.
Why do I want better visuals? To grow the channels.
But was that actually the thing holding the channel back? I checked the numbers. One channel is new but growing steadily. The other has a strong click-through rate and a high subscribe rate per view — the content clearly works. Its real bottleneck is just raw exposure, which is mostly a matter of time on a new channel. Visuals weren't the problem at all.
So "better visuals" wasn't really about growth. It was something else — I'd seen a slicker channel and felt my own work looked plain. A real feeling, just not the one I started with.
So if visuals weren't the real issue, what would actually make the content more dynamic? I dug into my editor's own code. Turns out it already had most of the pieces — sequencing parameters, a video background option, camera moves. What it didn't have was a way to animate cause-and-effect connections, which is exactly what my videos are about: explaining why something happened. That became a real, concrete todo.
To answer that, I opened my analytics dashboard. That's where I found the real problem.
The real bottleneck was somewhere else entirely
My analytics tool already does a lot: it pulls YouTube performance automatically, tracks which shorts drive longform views, indexes video structure into a database, and monitors trending topics for new video ideas.
Except for one file. A single JSON file mapping shorts to their longform videos, hand-edited. I'd broken its formatting more than once, and each time, I just stopped updating it. Not a technical problem — an annoyance so mundane I'd been living around it instead of fixing it.
So which project should own this fix?
My first instinct was to just patch the JSON file. But the editor already knows which channel and playlist a video belongs to when it's saved. The analytics tool already has a database. The hand-edited file was redundant with data that should exist somewhere else already.
That's where I had to stop and actually think about architecture — which two projects I'd deliberately kept apart were suddenly asking to talk to each other, and why. That part's its own story, coming next.
For now, here's what mattered most about this one: if I'd let AI build my first request literally, it would've solved the wrong problem.
Most of the work isn't building the solution. It's discovering what the problem actually is.
Wrong assumption: The videos needed better visuals. Discovery: The real bottleneck was a workflow problem, not a design one. Next step: Deciding whether two separate projects needed to merge to fix it.