What we know

Harvard researchers have found that the use of AI coding agents has shifted the main bottleneck in software development from writing code to reviewing it. These AI tools speed up code generation, but the resulting increase in code volume and complexity demands more thorough human review. This change could slow down the overall development timeline despite faster coding.

Why it matters

AI coding agents like GitHub Copilot are becoming common tools for developers to automate and accelerate coding tasks. Understanding that the bottleneck has moved to code review highlights new challenges in managing quality and oversight. It suggests that while AI can improve productivity in one area, it may create delays or require more resources in another part of the development process.

What is still unknown

Details about the accuracy of AI-generated code, which programming languages are best supported, and the timeline for wider adoption remain unclear. No independent tests of these AI coding agents’ effectiveness or impact on development speed have been reported.