The 80% Kill Rate on Prototypes
The Story
Boris’s team iterated on a single terminal spinner feature with 50 to 100 versions. 80% of them never shipped.
A condensed file view went through approximately 30 prototypes, followed by a full month of internal testing before it was released.
When building becomes nearly free, the bottleneck shifts from “can we make this?” to “should we ship this?” Boris holds the exact same quality bar regardless of whether code was written by AI or by a human. “If the code sucks, we’re not going to merge it.”
Boris personally ships 10-30 PRs per day while introducing only 2 bugs per month (versus an estimated 20 bugs if he’d written everything by hand). The secret isn’t speed. It’s judgment applied at speed.
Lesson for Creators
When AI lets you produce 10x more drafts, your editing taste becomes your competitive advantage. The creator who publishes everything AI produces will drown in mediocrity. The creator who produces 10 drafts and ships 2 will win. Volume of production is no longer the bottleneck. Quality of selection is.
Related
- Unshipping Features at Instagram — pruning as quality strategy
- He Deletes Underperforming Tweets — curating public output
- Three Modes of Working With AI — same quality bar regardless of source
- One Great Article Is Worth 50x Average — week-plus per article, killing anything that doesn’t meet the bar