AI is making software developers faster — just not at actually shipping software
SMRTR summary
Seven hundred and forty-one percent more lines of code. Twenty percent more software actually shipped. That striking gap sits at the heart of a new MIT and Wharton study tracking over 100,000 developers, and it's raising serious questions about the billions companies are pouring into AI coding tools.
Researchers traced developer activity all the way from raw code through to finished releases, finding that while AI dramatically accelerates the writing of code, the later stages, reviewing, testing, and managing releases, still demand human judgment and coordination. No amount of extra code changes that.
Perhaps most telling, a separate trial found developers using AI believed they finished work 20% faster when they actually took 19% longer.
The researchers call this the "weak-link hypothesis." AI removes one bottleneck but exposes the others. And as one decade of industry data from Google already suggests, lines of code were never the right thing to count in the first place.
SMRTR provides this summary for quick context. The original article belongs to Quartz.
Read the original article