An independent survey by Coleman Parkes for Undo asked 300 senior engineers working on mission‑critical C/C++ systems about AI coding agents. The respondents said AI speeds up code writing but pushes the hardest work to debugging and understanding the code. Teams now spend 16.9 hours a week debugging, about 42 % of their time, versus 9.8 hours writing code. About 35 % of AI‑generated code goes to production before the team fully grasps it, and 80 % say agents struggle with complex problems.
Why it matters
For engineering managers, the shift means AI tools may not speed up releases and require more resources for code review and debugging.
