Human reviews absorb coding-agent gains in a large firm study
Harvard researchers Fiona Chen and James Stratton analyzed engineering records from more than 700 software firms; Ars Technica reports more generated code and longer reviews, without a statistically significant rise in completed issues.