#enterprise code

AI coding agents keep failing on real company code

A new benchmark ran eight frontier models against real tasks from private production codebases, and the best one solved under 40%.

Withspecific · 4m