Tech leaders at Meta, Google and Microsoft boast that AI now writes a growing share of internal code, claiming it cuts costs and speeds delivery. In April, Google said three‑quarters of new code came from AI, while Microsoft’s CEO said 30 percent and CTO Kevin Scott projects 95 percent by 2030. The narrative is loud, but the reality inside engineering rooms is different.

Developers on Reddit and Hacker News report that the promised productivity gains evaporate once the code is handed over. They describe AI output as buggy, requiring extra debugging, and note that the output often forces them to sift through thousands of lines to spot mistakes. The time spent on verification outweighs any incremental speed for.

Many engineers say the pressure to use AI is built into performance reviews and team structures. One UX designer noted that AI pods became mandatory, with output quality rarely assessed. Another engineer observed that the sheer volume of AI‑generated code creates a “rat’s nest” of technical debt that will become costly as token prices rise.

The backlash shows that AI tools are not a silver bullet. While some developers find value in rapid prototyping or documentation lookup, most report increased cognitive load and a decline in coding skill. As firms push adoption, the industry must balance automation benefits against the risk of eroding core engineering expertise for future innovation strategies.