A METR study found developers using AI tools took 19% longer on real tasks but perceived a 20% speedup. This illusion of productivity masks real costs.
Code Rabbit analysis of 470 PRs showed AI-co-authored code had 75% more logic errors and 2.74 times more security issues. Harness survey of 500 leaders revealed 67% spend more time debugging AI code. Stanford data shows junior dev pay 20% below 2022 peak.
Stack Overflow's 2026 survey shows most devs use AI but only 6.6% trust it for critical decisions. The author advocates breaking work into small, planned steps rather than relying on large AI-generated PRs.
Source: Towards Data Science · Summarized by HeadlinesBriefing