Oracle has spent billions of USD over the past two years helping external clients run AI workloads. However, the company faced internal challenges in adopting generative AI, with co-CEO Clay Magouyrk admitting that until last year, the firm had not identified effective ways to implement these tools across its workforce.
The internal shift began in April and May of this year with the deployment of ChatGPT Enterprise and Codex. By implementing strict security controls and internal policies alongside access, Oracle achieved an 80% adoption rate among employees within three months. To manage costs, the company introduced monitoring mechanisms to track model usage and associated expenses.
CIO Jay Evans noted that OpenAI GPT-6 Astra costs 2.5 times more than other models, leading the company to favor more cost-effective alternatives like Terra for routine tasks. While AI has significantly accelerated code generation—reducing tasks that previously took two to three quarters to roughly one week—this has not yet resulted in faster product delivery.
Oracle is currently redesigning its testing, validation, and deployment processes to keep pace with increased coding speeds. Additionally, the company is testing Anthropic's Mythos model for security vulnerability scanning. While it identified more potential issues in two weeks than in the previous year, approximately 60% to 70% of these findings were false positives, requiring new verification workflows.
