Researchers Fiona Chen and James Stratton from Harvard University analyzed aggregated engineering analytics from Jellyfish, encompassing roughly 300 million work events—including commits and pull requests—and issue‑tracking data from over 700 k employees across more than 700 software firms spanning 2021 to March 2026. By correlating measured AI tool adoption with GitHub activity, they identified the introduction dates of AI coding assistants (which primarily provide auto‑completion for human‑written code) and AI coding agents (which autonomously generate and submit code). Using a difference‑in‑differences regression framework, they quantified the impact on development metrics. The study found that deploying AI coding agents raised total lines of code produced by about 30 %, increased the average number of commits by 20 %, and boosted pull‑request volume by 23 %. Despite these gains in raw code output, the resolution rate for Jira‑tracked issues and epics showed no statistically significant change, and there was no shift in the size or complexity of those work items. The authors attribute the lack of downstream productivity gains to a bottleneck in code review, noting longer review cycles, higher revision rates, and increased reviewer comments, which collectively offset the initial coding efficiency improvements. Consequently, the research provides little evidence that AI coding tools increase overall software output or reduce employment in the firms examined.