Palantir's Forward-Deployed Model

Many people in the software industry hold a quiet assumption: forward-deployed engineering is a specialty for unusual companies, not a model that serious software businesses should copy. Under this assumption, a company that sends engineers to sit with customers must be doing something exceptional, perhaps because its customers are government agencies with difficult problems and slow purchasing processes. This article starts from that belief and shows why it no longer fits the current period. Seeing why it is wrong explains why so many companies are now adopting the model.

The word 'forward' signals where the engineer is positioned. The engineer sits at the customer's location rather than only at the vendor's headquarters. The work itself consists of getting software running against the customer's real systems, not a idealized test environment. That placement is the defining feature of the model: the engineer's day is spent inside the customer's world, close enough to see the systems, data, and workflows that the software has to fit.

The niche reading was not overturned by an argument. It was overturned by buying behavior. In 2024, Palantir's commercial business changed direction. Enterprise customers, watching the AI hype cycle and uncertain how to respond to it, began buying Palantir's AI Platform.

The source's explanation for this buying pattern is specific. Enterprises bought because Palantir was already on-site and the FDE capability was already in place. The model that had been treated as a niche turned out to fit the way enterprises were now buying AI. The shift is therefore about fit between a delivery model and a buying pattern, not about a change in how engineers write software.

The adoption did not stay with Palantir. The source lists OpenAI, Anthropic, Google Cloud, and Salesforce among firms that have adopted FDE models, and it notes that the movement spread through many organizations. OpenAI formalized a Forward Deployed Engineer team in early 2025 that grew from two engineers to more than ten within less than a year. Anthropic built an Applied AI engineering function on a similar pattern. By the end of 2025, more than a hundred Y Combinator companies were hiring for the title, starting from a base of zero three years earlier.

The stronger claim in the source is that FDE is a gating factor for enterprise AI adoption. The source describes the bottleneck as neither model capability nor capital. It is the supply of people who can sit at a customer's site and ship code against the customer's actual systems.

On this reading, a company may have a capable AI product and the money to deploy it, yet still be stuck until someone can connect that product to the messy, specific systems the customer already runs. The product is not the constraint. The connection work is.

The diagram above is a teaching model built from the source's description, not a figure from the source. The key point it shows is that the product and the customer's systems are both present, and the FDE is the link that makes them work together. Without that link, a capable product can remain unused.

The source treats the labs' investments as a single proposition seen from two sides. Both labs are committing large sums to producing this kind of engineer. The source also cites the December 2025 Anthropic-Accenture training partnership, which put more than thirty thousand Accenture staff through Claude training in a single quarter, as the same bet arriving from a different direction.

The source describes two new vehicles that structure FDE work. They differ in market positioning as well as in structure, so the comparison below is the best way to see how they relate.

On structure, the OpenAI vehicle is the larger of the two and is financially structured around its backers. The Anthropic vehicle is smaller. The source does not give a dollar figure for either venture, so the capital comparison has to stay qualitative. The source's own table reports figures where it gives them and says so where it does not, and this article follows that rule.

On target market, the OpenAI vehicle is described as a machine for distribution through its backers' portfolio companies. It is also described as a way to gain delivery capacity quickly through an acquisition. The Anthropic vehicle is aimed at companies one tier below the largest customers. Goldman Sachs' asset and wealth management head described its purpose as to 'democratize access to forward-deployed engineers', and the source reads the implicit argument as bringing embedded expertise once reserved for the largest customers to companies one tier down.