Alex Karp's Palantir bet that enterprise value can't be shrink-wrapped: it has to be tailored per enterprise, by engineers deployed inside the customer. Drag the scene, crank enterprise complexity, and run a deployment race between the two models.
Classic software economics: build once, sell identical copies to thousands of customers. Marginal cost near zero, beautiful gross margins — if the product fits out of the box.
In large enterprises it often doesn't. Data lives in decades-old ERPs, spreadsheets and tribal knowledge. The gap between "product installed" and "value created" is exactly the red ring you see in the scene — and it's where most enterprise AI pilots die.
A Forward Deployed Engineer is a full software engineer stationed at the customer — on the factory floor, in the ops center — who builds working software against the customer's real data in days, not quarters.
At Palantir, FDEs ("Deltas") pair with platform engineers ("Devs"). FDEs don't build from scratch: they configure and extend a core platform — Foundry's ontology, a semantic layer mapping raw tables to real objects like pump, shipment, patient — so each deployment is custom fit on shared rails.
The gold arcs flowing back to the core are the model's real engine: every bespoke fix an FDE builds on-site is a product signal. Recurring patterns get generalized into the platform, so deployment N+1 is faster than deployment N.
That's how high-touch service work compounds into product: the service margin is the tuition; the platform is the degree. OpenAI, Anthropic and a wave of AI startups have since adopted FDE-style teams for exactly this reason.
Economics: high touch, high ACV — seven-figure contracts justify embedded engineers; usage expands after trust is built ("acquire, expand, scale").
Criticisms: it looks like a consulting firm wearing a software valuation — headcount scales with customers, margins lag pure SaaS, and revenue concentrates in big government/enterprise deals. The bull case is that the ontology + AI agents keep shrinking marginal deployment cost; the bear case is that "tailored" never stops meaning "labor."