Oracle exec says internal AI rollout lagged behind external services

Oracle spent billions over the past two years helping other companies run AI. Its own companywide AI rollout came much later. During an internal town hall this week, co-CEO Clay Magouyrk said that last year, the company still hadn't found a way to make generative AI broadly useful to its own workforce. Now, after rolling out OpenAI's ChatGPT and Codex, employees are writing code dramatically faster, but creating new bottlenecks along the way. "When I think about where we were a year ago," Magouyrk said, "I don't think we figured out how do we make AI really that useful for ourselves." Oracle had made some progress in using AI for customer support last year, but it had not yet found broad use across business units, such as developers, finance employees, and sales teams. That started to change, Oracle Chief Information Officer Jae Evans told employees, in April and May, when the company rolled out ChatGPT Enterprise, and OpenAI's software development tool, Codex. Rather than simply making the tools available, the company established corporate standards, security controls, and internal policies and reached 80% adoption within three months, Evans said. "We made it so easy to use and adopt that we might have gotten a little bit of sticker shock," Evans said, so now the company provides visibility into which models employees are using and how much they cost. OpenAI's GPT-6 Astra costs 2.5 times more than other models, like could be used for many purposes, Evans said, citing Terra as an example of a lower-cost model that could be used for routine tasks. The executives said AI has dramatically accelerated software development, but also created new bottlenecks elsewhere in the engineering process. "We see developers able to generate code using this tool in like a week's time," Evans said. "Normally, that would have taken a team of developers two to three quarters to go develop." For now, AI deployment at Oracle hasn't translated into getting those products to customers more quickly. "When you make the actual act of writing the code quicker, it doesn't mean that suddenly everything is 1000 times faster," Magouyrk said, explaining that Oracle is still figuring out how to redesign its testing, validation, deployment, and release management processes to keep pace. Oracle's experience highlights the next challenge facing companies rushing to bring AI into the workplace. Powerful models can dramatically speed up individual tasks, but they can also rack up big bills and simply move bottlenecks elsewhere. JPMorgan recently set cost limits for employees using Claude from Anthropic. Oracle also received early access to Anthropic's Mythos Preview model, which scans code for security vulnerabilities, and used it to uncover more potential security issues in its first two weeks than it had found in an entire year, Evans said. Announced in April, Anthropic's Mythos is considered one of the most powerful AI systems ever developed, with governments and security agencies warning that its capabilities surpass those of other models, especially for cybersecurity. That model generated a high number of false positives (Evans estimated roughly 60% to 70%), prompting Oracle to build new processes to verify vulnerabilities before engineers began fixing them. Anthropic did not publicly release the model. Have a tip? Contact this reporter via email at astewart@businessinsider.com or Signal at +1-425-344-8242. Use a personal email address and a nonwork device; here's our guide to sharing information securely.