OpenAI's FDE head Colin Jarvis explains how the team works: start with the outcome, then make itself obsolete

OpenAI published an essay on October 22 by Colin Jarvis, who founded its Forward Deployed Engineering team in early 2025, describing what the job means there. The team's stated mission is to close the "deployment gap" between what models appear able to do and what people can reliably use, and Jarvis says the team is doing its job if it makes itself obsolete every couple of months.

The core value is "start with the outcome, do what it takes to succeed, then follow with scale." In one example, about 15 FDEs embedded with a large semiconductor customer's verification team in mid-2025, some of them new to the industry, built evals and custom harnesses, and got 10+ use cases into production within 3-6 months, reporting around 48% time saved on directed test generation. Later they rebuilt those apps as plugins and skills on Codex, removing roughly 90% of the custom code; extending one use case to a second team took a day after a two-month initial deployment.

The essay also covers Tax AI built with Thrive Holdings (7,000 returns processed in a pilot, about 50% more throughput), OpenAI Presence growing out of customer deployments, and a telecom voice project where FDEs and Research lifted key eval scores from about 60% to 90%+ over six months.

Why it matters for FDEs: it is a first-hand description of what OpenAI expects from the role. Engineers are expected to learn the customer's domain fast, build evals before scaling, feed model gaps back to Research, and keep moving to harder problems as products like Codex absorb earlier work. The team mixes generalists with domain experts, and the post ends with an invitation to join.