Why aren't AI productivity gains higher?

Developers across companies explain why real-world AI productivity gains remain modest despite the hype. Key reasons include: coding represents only a small fraction (~16%) of engineering work, AI speeds up code generation but creates downstream bottlenecks in code review and validation, social friction and polarization between pro- and anti-AI engineers slow adoption, skill and tooling gaps limit effective use, and AI tools lack the implicit business and system context that lives in engineers' heads. Additional observations note AI excels on greenfield projects but struggles with large legacy codebases, documentation quality from AI is mixed, and there's growing concern about developers offloading critical thinking and junior engineers bypassing skill development.