AI coding agents generate more code, but not more software

A new study shows that while AI coding assistants produce more lines of code, human review becomes the bottleneck, meaning productivity gains don't translate to faster software development. This challenges the hype around AI replacing programmer work and matters to tech workers and companies.