To deploy a genAI-powered business application, you need more than a general-purpose large language model, even one that’s carefully fine-tuned and prompted. To build and safely operate genAI-powered business applications, you need to assemble an architecture that draws on resources from many layers of intelligence; input gates to shape requests and output gates to protect the firm; application pipes to tie it all together though an API-first workflow; and feedback loops to test, monitor, and ensure quality. You can assemble these applications from parts today as you establish a full architecture to support genAI-powered business applications over the next three years.