LLMs are not specifically worse than humans at contributing code — in Schirp's model they're just a mechanism that throws many, many more darts at the wall. The distribution of quality is similar, so the ratio of red-zone (automation-passing, contribution-failing) merges scales up with volume. With limited human review time for the red area, more bugs reach production unless the automation threshold is raised. Solution: invest in narrowing the automation–contribution gap so more bad contributions are auto-rejected.