Oracle's Clay Magouyrk: AI coding rollout outpaced our processes

Oracle's admission is notable because the company spent billions over the past two years helping other companies run AI, yet its own companywide rollout came much later. According to co-CEO Clay Magouyrk, Oracle had made some progress using AI for customer support but had not found broad use across business units such as developers, finance employees, and sales teams. That gap between selling AI and using it internally is the backdrop for the town hall remarks.

The mechanics of the rollout matter. CIO Jae Evans said the shift began in April and May with ChatGPT Enterprise and OpenAI's Codex, and that Oracle paired the tools with corporate standards, security controls, and internal policies, reaching 80% adoption within three months. The cost side emerged quickly: Evans cited GPT-6 Astra at 2.5 times the price of other models and pointed to Terra as a lower-cost option for routine tasks. Meanwhile, Magouyrk said code that would have taken a team two to three quarters can now be generated in about a week, yet products are not reaching customers faster because testing, validation, deployment, and release management still need redesigning. Oracle's experience with Anthropic's Mythos Preview model illustrates the same tension in security work: Evans said it found more potential vulnerabilities in two weeks than Oracle had found in a year, but roughly 60% to 70% of findings were false positives, requiring new verification steps.

The stakes hinge on whether Oracle can rework its engineering and release processes to match the speed AI brings to individual tasks. If it can, the 80% adoption figure could translate into faster customer delivery; if not, the gains may stay confined to code generation while costs and review burdens grow. The same question likely applies to other large employers making similar rollouts, such as JPMorgan, which the body notes recently set cost limits for employees using Anthropic's Claude.