Current AI coding tools are demonstrably flawed; one analysis cited a SQLite rewrite failing by a factor of 2,000. This points to unreliability across the board.

The technical split is stark: some users, like CubitOom and saltesc, dismiss generative models as hype, citing high catastrophic failure rates and inherent limitations like hallucination. Others, like Flavames5123, find limited utility in learning new systems but admit cleanup is massive. Furthermore, key contributors point fingers at the market, with DickFiasco asserting that AI consulting hype merely exploits corporate Fear Of Missing Out (FOMO). The core argument from Thorry is that the cognitive load to validate AI code exceeds the effort of writing it manually.

The consensus points to deep skepticism. The market's adoption of AI seems driven by panic rather than proven need. While some users argue skipping AI is impossible because 'early adopters always pay the most,' the weight of technical testimony suggests AI currently requires rigorous human gatekeeping to avoid catastrophic failure.