Video Production
Stopping Product Videos From Going Stale as You Scale
A system that catches when a product video no longer matches reality, then regenerates only what changed — instead of rebuilding it from scratch.
Drift map
Every existing video mapped to the product flow it depicts, so it's immediately visible which ones no longer match what's actually shipped.
Staleness detection
An automated check comparing a video against the current state of its flow, flagging a mismatch before a customer or exec does.
Targeted regeneration
Every product flow broken into modular, reusable segments — the atomic unit a video is built from — so a single change triggers an update to just the affected segment instead of a full re-edit.
Maintenance loop
The ongoing cycle of catching drift, regenerating what changed through Claude Code, and confirming the library is current — running in the background instead of as an occasional cleanup project.
Challenge
Every demo video was a one-off. Each took days to make, and once it was published nobody owned the question of whether it was still true. One small UI change meant re-editing from scratch, so people either stopped asking for new videos or kept sharing old ones that didn't match the product anymore. We usually found out when a customer pointed at something on screen that wasn't there.
Process
I started by going through every video we had and writing down which product flow it showed. Then I rebuilt each flow as a set of segments, short reusable pieces, with instructions for Claude Code describing what "current" looks like for each one. The first version treated a whole video as one unit, so any change still meant a full rebuild. Once the flows were split into segments, two things became possible: checking automatically when a video had drifted, and regenerating only the segment that changed.
Research
The expensive part was never the production quality. It took me a while to see that. What cost us was treating every update as a full rebuild, and nobody owning "is this still true?", so the stale videos piled up quietly until one surfaced at a bad moment.
Design
Claude Code does both halves. It compares each video against the current state of its flow to catch drift, and when something has changed it regenerates just that segment. The whole library stays consistent without depending on one editor's memory, including mine.
Outcome
We don't find out about an outdated video from a customer anymore. Regenerating one segment takes minutes instead of a re-edit, so keeping the library current turned into background work. The check still misses copy changes, a renamed button or a new label, so I go through the list by hand every week or so.