OpenAI FDE team cuts custom code roughly 90% with Codex

OpenAI frames the FDE team's purpose around what it calls the deployment gap: model capabilities have moved faster than the systems and processes needed to put them to use. Sometimes the missing piece is a product; sometimes a model performs well on a benchmark but cannot handle real-life variables and constraints. The team's stated goal is for each engagement to leave the next customer with a better starting point — if a class of problem no longer needs FDEs the next time, the team counts that as success.

The John Deere example illustrates the pattern. See & Spray, its precision weed-control technology, uses cameras and machine learning to spray weeds rather than treating entire fields, reducing chemical use by up to 70%. But capturing those benefits depends on farmers knowing how to configure the technology and plan their season, so John Deere deployed OpenAI's APIs to make support personalized and scalable. In life sciences, OpenAI says its FDEs' scientific backgrounds help them earn trust with researchers on tasks such as target validation and gene-expression analysis.

The account also describes how the team manages to shrink its own custom work. After Codex grew in capability, one FDE tested rebuilding an application around it and moved use cases onto a common harness, turning custom applications into plugins and skills. OpenAI says this let it retire separate terminal UIs and CLIs, and that the same principle fed into OpenAI Presence, its product for enterprise voice and chat agents. In a telecom engagement, the model scored around 60% on key evals during scoping; over six months, with Research working on capability gaps, performance rose to 90%+.