How are businesses approaching AI in cloud architecture to manage operational costs?

Businesses are evolving from unoptimized AI implementations toward more strategic architectures. Initially, many companies are placing 'unmitigated mayhem' into prompts and using expensive models for tasks that don't require them, similar to how organizations once stuffed everything into a single virtual machine rather than using specialized cloud services. The emerging solution involves creating 'generative solutions architecture' that uses the right models for specific tasks at appropriate price points. This means using lightweight, specialized models (like Flash) that cost a fraction of heavyweight ones, implementing RAG systems, and designing architectures that might use only 1-3% AI instead of 90% to achieve the same business outcomes while managing costs effectively.