Oracle: 80% AI Adoption, Code Gains but Bottlenecks and Costs

Key Takeaways

1Oracle deployed ChatGPT Enterprise and Codex in the spring, achieving 80% adoption in three months

2AI tools accelerate code generation but shift bottlenecks and do not yet speed up delivery to customers

3Anthropic's Mythos model has enabled the detection of more security vulnerabilities but generates 60 to 70% false positives

Why it matters : Oracle's experience illustrates the operational and financial challenges of large-scale AI deployment in enterprises, balancing quick gains with new obstacles to overcome.

Oracle has rolled out ChatGPT Enterprise and Codex widely in the spring, achieving 80% adoption within three months. Management notes significant accelerations on the development side, but also acknowledges shifted bottlenecks, cost issues, and product process adjustments that still hinder faster delivery of innovations to customers.

Products Not Faster and Processes to Redesign, According to Co-CEO

Oracle's management asserts that the widespread adoption of AI tools has not yet accelerated product availability for customers. Clay Magouyrk clarifies that gains in code writing do not make the entire cycle "1000 times" faster. He indicates that the stages of testing, validation, deployment, and version management still need to be rethought, while executives recognize shifting bottlenecks within engineering.

Costs Under Scrutiny and Enhanced Governance of Models

CIO Jae Evans explains that the ease of use has caused a "small shock," and that Oracle now provides visibility into the models used by teams and their costs. He mentions a discrepancy: GPT-6 Astra costs 2.5 times more than models like Terra, which are more suited for routine tasks. To regulate usage, the company states it has established standards, security measures, and internal rules, achieving an adoption rate of 80% in three months.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Internal Framework, Deployment Timeline, and Catching Up on Usage

Jae Evans identifies the turning point in April and May, with the deployment of ChatGPT Enterprise and OpenAI Codex. Clay Magouyrk reports an adoption rate reaching 80% in the spring, whereas a year earlier, the company was still figuring out how to make AI useful internally. Last year, usage was primarily advancing in customer support, rather than among developers, finance, or sales.

Code Acceleration and Security: Quick Gains, High False Positives

According to Jae Evans, developers can generate in one week code that would have taken a team two to three quarters to produce, illustrating the gains observed after the arrival of ChatGPT and Codex. In security, Oracle had early access to Anthropic's Mythos Preview model for code analysis and claims to have detected more potential vulnerabilities in two weeks than in a year. However, Evans notes that 60% to 70% of these were false positives, leading to the implementation of checks before corrections. Mythos, announced in April and not yet publicly released, is touted by authorities as surpassing other models, particularly in cybersecurity.

Previous Investments and External Signals on Costs

Oracle claims to have invested billions over the past two years to help other companies adopt AI, after years spent building its infrastructure. The large-scale internal deployment came later. At the same time, other players are establishing financial safeguards: JPMorgan recently set cost ceilings for the use of Claude. Internally, Clay Magouyrk, co-CEO, acknowledged to teams that the widespread adoption of AI has not been without its challenges.