Palantir

Bespoke Gotham deployments for American intelligence were encoded into platform primitives: ontology, object models, permissions, workflow engines, provenance traceability. Those primitives became Foundry, then Apollo and AIP. Once deployments standardized around Foundry, gross margin rose into the 80 percent range, and Palantir left the forward deployed engineering motion for selling by account.

That sequence is why the company anchors nearly every argument about Forward Deployed Engineering. Jesse Zhang, CEO of Decagon, writing on X on 11 August 2026, treats it as the reference case: Shyam Sankar's line, "FDEs eat pain and excrete product," alongside Joe Lonsdale's admission that the "consulting firm in disguise" reputation rested on an accurate observation. Palantir carried that reputation for close to twenty years before the margin arrived.

Zhang's reading of the precedent is compressed into one claim: "the pain was the input to the product, not a cost of sale." The difficult move is the exit, removing the engineers once the customer journeys are known. Anthropic and OpenAI have built enterprise deployment arms explicitly modeled on Palantir, which copies the entry while leaving the exit untested.

Ethan Mollick's May 2026 provocation approaches the same seam from the other direction: as long as humans are needed to fit AI into client organizations, the adoption bottleneck sits in the organization rather than the model. Whether anyone copying the entry can repeat the exit remains open.