9 Lessons Learned from Deploying GenAI at Scale • Garth Gilmour & Stuart Greenlees • GOTO 2025

Deploying Generative AI at scale within a large enterprise requires a "skin your knees" approach, where organizations embrace iterative failure to refine their strategies. Success hinges on establishing a centralized, governed platform that provides guardrails while maintaining developer autonomy. Because models are not interchangeable, companies must support a diverse range of options, balancing performance against cost and latency. Furthermore, the rapid evolution of AI technology means that custom-built solutions often have a short half-life, necessitating a composite architecture that allows for the quick replacement of obsolete components. Effective scaling also demands rigorous cost observability, as excessive token consumption can quickly spiral. Ultimately, achieving fluency in AI requires moving beyond simple adoption to continuous education, automated evaluation pipelines, and a focus on solving unique business problems rather than chasing the latest, most expensive frontier models.