Snipes Retrospective
Interviewed for YC F26.
TL;DR: Snipes was a live map and feed of what's happening around you right now. You'd point your phone at something and go live for a few minutes, and AI cut your stream into clips while you were still live. We launched at UWaterloo for three days: 488 people joined, 1 in 4 went live, and they started 1,632 streams. Then the compute bill came due and we killed it.
Livestreaming platforms are built to answer one question: what is someone I follow doing right now? We wanted to answer a different one: what is happening around me right now?
We bet on three ideas.
Location is an underused social graph
To the vast internet, a random stranger's stream is noise. To a bunch of college kids who recognize that the stream is from their cafeteria, it's the most relevant thing on their phone. Proximity can make content matter before the person filming has any following.
Streaming can be presence, not performance
Twitch asks you to entertain people for hours. Snipes asked you to point your phone at the party, your friends, or whatever funny/shocking/wholesome thing you were doing, for a few minutes. That changes what it takes to supply content, less indexing on the 'talent' required to constantly create clip-worthy moments like Neon or Clavicular.
AI should be the format, not the feature
Live networks die at small scale because nobody is live when you open the app. If a model watches every stream and cuts the best moments into clips before the stream ends, one person's hour becomes a whole feed. The app is never empty, and every ephemeral moment has the chance to travel (like a vine?).
I haven't been a huge fan of the dead internet. However late-stage capitalist livestreaming may be, this was my optimism for something less parasocial.
So the atomic network we targeted was a campus, not a global audience: densely bound, with something always happening, and surface area for serendipity.
How it held up
The supply bet was right. The pattern of 1 in 4 people streaming is much more akin to what Instagram was originally about: seeing your friends.
The format bet had a cost we underpriced. If AI clips are the format, you pay to process every stream, including the ones nobody watches. Clipping compute went over budget within the launch window, and at current compute costs it made no sense to run for free without huge venture backing. That's why we shut it down after three days.
The place bet is the question we never got to answer. On a campus, location is the social graph. Off campus, the risk is that it fades and you end up competing with Instagram on its home turf. Zenly is the cautionary tale.
After office hours with Pete from YC, we decided there were other games we'd much rather play than building a social app. I want to be working on the thing I'd work on for the next ten years.