In the past two years, Oracle has earmarked substantial funds to assist other companies in harnessing the power of AI. However, when it comes to fully integrating AI within its own organization, Oracle was somewhat late to the game. Last year, the company was still grappling with finding an effective way to implement generative AI across the board. In April and May of this year, Oracle took a significant leap forward by deploying ChatGPT Enterprise and Codex. Simultaneously, it rolled out a set of standards, security control measures, and internal policies to ensure smooth and secure operations. Remarkably, within a mere three months, the utilization rate of these AI tools soared to 80%, leading to a notable boost in employees' code-writing speed. Yet, this rapid acceleration also unveiled bottlenecks in other areas. The faster development pace didn't necessarily translate into quicker product delivery. Presently, Oracle is delving into the restructuring of relevant processes to address this issue. When experimenting with OpenAI's GPT-6 Astra, Oracle discovered that its cost was 2.5 times higher than that of other models. Consequently, the company made the strategic decision to switch to the cost-effective Terra for daily tasks and set up an internal monitoring mechanism to keep tabs on performance. Moreover, Oracle proactively employed Anthropic's Mythos Preview model to scan for code security vulnerabilities. This move proved fruitful, as it uncovered more potential issues in just two weeks than had been identified in the entire previous year. However, approximately 60% to 70% of these findings turned out to be false positives. This prompted Oracle to establish a new verification process to ensure the accuracy and reliability of its security scans.