On 2026-10-22, OpenAI published a description of its Forward Deployed Engineering (FDE) team, written by Colin Jarvis, the company’s head of FDE, setting out how the team works with enterprise customers and the results it reports (OpenAI). Jarvis wrote that he founded the team in early 2025 to close what OpenAI calls “the deployment gap”, the distance between what models appear capable of and what people can reliably use to do their work (OpenAI). The company reports that moving use cases onto a common harness with Codex at its core removed roughly 90% of the custom code from each project (OpenAI).
Key Facts
| Item | Detail | Source |
|---|---|---|
| Team founded | Early 2025, to close the deployment gap | OpenAI |
| Semiconductor engagement | Around 15 FDEs on 10+ use cases in mid-2025 | OpenAI |
| Reported time saved | ~48% through AI-based directed test generation | OpenAI |
| Custom code removed | Roughly 90% from each project | OpenAI |
| August 2026 example | Two months for first deployment, one day for second team | OpenAI |
Timeline
- 2026-10-22: OpenAI published “What FDE means at OpenAI”, by Colin Jarvis (OpenAI).
Details (as Reported)
- OpenAI describes the FDE approach as working alongside customers on their hardest problems, with what the team learns shaping its products and giving Research concrete problems to solve (OpenAI).
- In mid-2025, around 15 FDEs worked with a large semiconductor customer on 10+ use cases, aiming to make verification AI-native through directed test generation, root cause analysis and debugging, and an architecture assistant (OpenAI).
- Jarvis wrote that verification can account for ~70% of a project’s effort in that industry (OpenAI).
- OpenAI reports that all of the use cases reached production and early user adoption over the first 3-6 months (OpenAI).
- The company lists these metrics: ~48% time saved through AI-based directed test generation, 22% median performance improvement through kernel optimisation, and weeks of debugging compressed into hours (OpenAI).
- John Deere deployed OpenAI’s APIs to personalise support for its See & Spray weed-control technology, which OpenAI describes as reducing chemical use by up to 70% (OpenAI).
- According to OpenAI, the AI can answer farmers’ questions in natural language, provide tailored setup guidance, and identify when equipment may not be delivering its full potential (OpenAI).
- In life sciences, OpenAI says FDEs’ scientific backgrounds help them work with researchers on pre-clinical drug discovery tasks such as target validation and prioritization, gene-expression analysis, and chemical synthesis-route reconstruction (OpenAI).
- One FDE tested rebuilding an application around Codex, and the team then moved use cases onto a common harness in which custom applications became plugins and skills built on common Codex primitives (OpenAI).
- OpenAI states that Codex also became the interface, allowing the team to retire separate terminal UIs and CLIs written for individual use cases (OpenAI).
- The company reports that extending a solution from one team to the next could be as simple as swapping in new connectors or MCP servers and building fresh evals (OpenAI).
What Is Not Yet Confirmed
The sources do not list open questions.