Inside OpenAI's Forward Deployed Engineering Team
Coordinate effectively across OpenAI’s Forward Deployed Engineering (FDE), Product, Research, Security, and other relevant teams to deliver reliable customer solutions and contribute to platform development. Establish clear communication channels, define stakeholder responsibilities, and implement structured escalation processes to resolve deployment blockers. Identify customer requirements, manage expectations, and verify integration readiness before proceeding with deployment. Maintain continuous feedback loops to communicate findings, address emerging risks, and support improvements. Develop actionable local plans to sustain progress while broader technical or organizational issues are being resolved. Secure clear commitments from stakeholders, track dependencies, and maintain accountability throughout the engagement. Demonstrate FDE effectiveness through cross-functional coordination, timely issue resolution, and successful customer delivery outcomes.
Larch’s FDE posted this to six teams a week before a customer visit:
“Product, Research, Partnerships, GRC, Security, GTM: Larch is blocked. Please fix before launch. I’ve told the sponsor we’re launching during the visit.”
Days later the supplier export was still missing, and nobody had agreed to fix anything. The message gave everyone urgency and nobody a task. An FDE almost never ships alone, and “How do you move a blocked deployment through people you don’t manage?” is a question to practice. A strong answer turns each blocker into a specific request, gets someone to accept it, and says what evidence will close it. Larch, a procurement customer, runs through the whole lesson.
Which teams does an FDE work with?
OpenAI describes the FDE team as working “at the intersection of customer delivery and core platform development,” which connects delivery for one customer with learning that can improve the platform. The posting names Product, Research, Partnerships, GRC (governance, risk, and compliance), Security, and GTM (go-to-market) as close collaborators. It doesn’t publish reporting lines or say every team joins every engagement, so use the names to frame questions and then confirm who actually decides on your engagement.
Use these as working coordination questions, not fixed team mandates. Product may assess recurring friction in an OpenAI product surface. Reproducible model behavior can be routed for Research investigation through the appropriate internal channel. Partnerships may help with partner or relationship dependencies. GRC can clarify assurance requirements and the relevant risk-acceptance authority. Security can assess connection and access risks. GTM can help align commercial expectations with verified scope and readiness. Confirm the responsible team, decision-maker, and escalation route for each engagement.
The customer’s people keep their own authority. At Larch, customer engineers control the supplier export, and procurement specialists decide what a useful price comparison looks like.
What makes an escalation work?
An escalation someone can act on states five things: the problem you observed, the consequence for delivery, the evidence, the specific decision or help you need, and who will follow it through. Record the recipient’s acceptance separately from the request. A deadline attached to a message is not ...