Oracle streamlines workflows with ChatGPT and Codex

Introduction

Oracle has been selling software to help other companies run their operations for decades. Now, it is handing a big chunk of its own work to someone else's software.

Oracle's Adoption of ChatGPT and Codex

According to a customer story published by OpenAI, Oracle has rolled ChatGPT and Codex across recruiting, engineering, and operations. The aim is to turn knowledge that lives in a few specialists' heads into workflows anyone on the team can run, quickly and repeatedly.

Adoption Rate

The headline number is adoption. Oracle reached 80% adoption within three months of its internal rollout, which began in April and May 2026.

What Oracle Actually Deployed

The rollout centers on two OpenAI products. ChatGPT Enterprise is the business version of the familiar chatbot. Codex is OpenAI's coding agent, built to write and work through software tasks.

Engineering Speed

The most striking claim involves engineering speed. Developers can now generate code in approximately one week, according to the research findings. The same work previously took two to three quarters.

Broader Implications

Engineering is not the only department involved. Recruiting and operations are also part of the push, with the goal of capturing specialist know-how in repeatable processes.

Cost Visibility

Oracle executives have noted that while engineering output accelerated, new challenges emerged around cost visibility. A traditional software license is simple to budget: you pay per seat and know the number in advance. Agentic tools that run multi-step tasks on their own can consume computing resources in ways that are less predictable, because one request might trigger a short job or a long chain of steps.

Broader Questions

Not every finding was celebratory. The research also flags broader questions about monitoring AI usage as these workflows spread.

Conclusion

For Oracle, the internal rollout doubles as a sales story. The company now offers OpenAI models through its cloud and embeds generative AI in Fusion Cloud HCM. Being able to point to its own 80% adoption figure gives its sales teams a reference customer that happens to be itself.

Conclusion

There are reasons for caution. The productivity figures come from a customer story published by one of the two companies involved, which naturally presents the deployment in a favorable light. Generating code faster is not the same as shipping reliable code faster, and the source material does not detail how output was reviewed or measured.

Conclusion

The cost visibility issue may prove to be the more lasting lesson. Finance teams that are used to predictable software bills may need new ways to track spending on tools that work in steps rather than seats.