When you ask an Android team to estimate a video memories feed, the answer is usually around six to eight weeks. It sounds reasonable. You need a horizontal scrolling list, a vertical full-screen pager, ExoPlayer, and some UI overlays. It's just views and media playback.

Four months later, the feature is still in QA. What happened?

The difference between "plays video" and "production video"

The 80% of video playback is easy. You pass a URL to ExoPlayer, and the video plays. The remaining 20% takes 90% of the time. The hidden costs always cluster in the same five areas:

• Video caching and prefetch. Users expect videos to start before they tap. Getting from "video loads eventually" to "video loads instantly" requires a real prefetch and cache layer.
• Seek behaviour under load. Users swipe quickly through a feed. The player needs to handle rapid seek, mid-buffer transitions, and partial loads gracefully.
• Android fragmentation. Video playback issues that only reproduce on specific chipsets, Android versions, or OEM skins are among the most expensive bugs to diagnose.
• Analytics instrumentation. Product wants to know what's being watched, replayed, liked, and shared. That instrumentation needs to be accurate and persistent across sessions.
• Edge cases in the UI. Empty states, error states, network interruptions mid-playback, re-entering the feed at the correct position. Together they take weeks.

The maintenance that never stops

Once the feature ships, the work doesn't end. Video playback bugs surface from real users on device configurations QA never touched. ExoPlayer releases updates that change behaviour. New Android versions break things that worked before.

The opportunity cost

A senior Android engineer costs between $120k and $200k annually depending on the market. Four months of their time on video infrastructure that has already been solved elsewhere is roughly $40–65k in salary alone.

The question isn't whether your team has the capability. Most teams do. The question is whether this is the best use of the time.

Build vs. buy for a memories feature is a question of where you want to spend engineering cycles. The infrastructure problem has been solved. Your competitive advantage isn't in solving it again.