We tracked AI coding assistant usage across our 12-person development team for six months. The results surprised us—both in what improved and what didn't.
Speed gains were real but uneven. Junior developers saw 40-50% productivity improvements on familiar task types. Senior developers saw 15-20%. The difference? Seniors already had patterns in their heads; juniors were learning from AI suggestions. The skill-flattening effect is real.
Code quality was neutral to slightly negative. AI-generated code passed tests but often missed edge cases, used inconsistent patterns, and accumulated technical debt faster. We introduced mandatory AI code review—ironic, but necessary.
The biggest win was documentation and test coverage. Tasks developers avoided became painless. Tests that would never get written got written. Documentation that would stay outdated got updated. AI's impact on the unglamorous work exceeded its impact on the glamorous work.
Jake Morrison
Contributing writer at MoltBotSupport, covering AI productivity, automation, and the future of work.