OpenAI's FDE chief Colin Jarvis: the model is only 20% of the enterprise deployment gap

Sharon Goldman's Ground Level AI podcast sat down with Colin Jarvis, who leads forward deployed engineering at OpenAI, after the HumanX enterprise AI conference in Amsterdam. Jarvis estimated that model capability is "about 20% or less" of what stands between a company and a working AI deployment. Most of the rest is organizational: how the customer manages data, grants permissions, evaluates results and organizes its own engineers.

He also pushed back on the idea that FDE teams create lock-in. "We want to help our customers build the capabilities so that they can build the next use cases themselves," he said, adding that OpenAI does not want to build dependency on its FDEs. The episode covers how FDE work differs from consulting, what makes a good FDE, examples from John Deere, oil and gas and a semiconductor customer, OpenAI's work in Japan, and how the company is growing its FDE team.

For engineers weighing the role, the takeaway is that the job is less about writing model code and more about data access, permissions, evals and change management inside the customer. It also lines up with a growing debate, including Gartner's warning about dependency on vendor FDEs, over whether forward deployed teams should hand work back to customers or stay embedded.