Developer Spends Three Rounds Debugging AI That Couldn't Place a Video Caption Correctly

A developer building a WYSIWYG drag-and-drop caption editor for a video tool called Keryx spent multiple frustrating sessions trying to get an AI coding assistant to correctly position captions on rendered MP4 files. Despite the underlying concept being straightforward — absolute positioning scaled by a ratio — the AI repeatedly fixed one issue only to introduce another across three separate iterations. The root cause turned out to be that both the preview size and render size measurements were being pulled from downstream proxies rather than the actual target dimensions hardcoded in the codebase. The preview element size carried browser sub-pixel and device-pixel-ratio rounding, while the render size came from a scaled-down proxy video whose dimensions were further altered by encoder rounding decisions. The developer noted the irony, having solved an identical problem by hand nearly two decades earlier in a couple of days using JavaScript, with less experience and worse tooling.