At Sierra, an enterprise conversational AI company with more than 120 engineers, Natalie Meurer argues the Palantir origin customer embedded engineering role should be defined by who owns the customer outcome, not by any particular skill stack.
The term "forward deployed engineer" has become almost nonsensical. The label has been stretched across the AI industry until it can mean anything from a Palantir-style customer-embedded delivery engineer to Sierra's in-house agent builder. The incoherence is the story.
Sierra, an enterprise conversational AI company with more than 120 engineers building agents for customer service, has responded by deleting the title. The role is now called "agent engineer," and the change is deliberate. In a Latent Space Q&A tied to the AI Engineer World's Fair's Forward Deployed Engineering track, the company's head of agent engineering, Natalie Meurer, argues the right way to think about the work is not skill set but accountability: who owns the customer outcome.
Meurer is making a structural argument about how AI engineering teams should be organized, not a branding point.
"Forward deployed engineer" was born at Palantir, where the model was to put highly technical employees inside or next to the customer and hold them responsible for delivery. Meurer spent five years there before joining Sierra. The label spread because the underlying problem, building software for a specific enterprise customer rather than for a mass market, kept recurring in new places.
In the AI era, that problem has gotten worse, not better. The surface where a company meets its customer is increasingly an agent: a piece of software that holds a conversation, takes an action, and fails in customer-visible ways. To build one well, an engineer needs to integrate with customer systems, design the agent's behavior, understand the underlying operations, shape the product, and care about how the end user experiences the result. Sierra's job description for an agent engineer lists all five as required.
The five-piece skill set is not the point, Meurer says. The point is that the same person holds all five accountability lines at once. Product engineering, building the platform, and customer-facing engineering, building for a specific buyer, are no longer separable jobs. They are starting to be the same job, held by the same engineer, on the same team.
Sierra's rebrand is the visible symptom of that convergence, not the cause. The convergence itself is happening wherever AI agents become the primary interface between a company and its customers, whether or not anyone renames the role. What Sierra has done is refuse to pretend the old split still holds.
The remaining question is whether the rest of the AI industry will follow. Right now "forward deployed engineer" is still in use at companies that inherited it from the Palantir playbook, at agent startups that never had a Palantir phase, and at large enterprises trying to staff internal AI teams. Meurer's argument is that the label, in any of those places, only stays useful if someone is clearly on the hook for the customer outcome. If nobody is, the title is doing work it cannot back up.
That is why "almost nonsensical" lands. It is not a complaint about a misused buzzword. It is a critique of an organizational structure that has stopped describing what it actually does.