AI industry's size obsession is killing ROI, engineer argues

Large language models (LLMs) face reliability and cost challenges in enterprise settings, despite claims of autonomy. They're prone to errors, especially in multi-step workflows. AI engineer Utkarsh Kanwat notes that even with 95% per-step reliability, a 20-step process would have only a 36% success rate, making LLMs impractical for systems requiring 99.9%+ reliability. Smaller, focused models may offer better controllability and cost-effectiveness. Experts recommend targeted AI strategies over all-encompassing large models for optimal results.