What's this about?
Ali Ghodsi answered the question he says he was asked repeatedly — how do you grow 80% at $7 billion in revenue? His answer: AI agents have finally started working in enterprises. The models have been very capable for a while, but lacked the basic context that lives in people's heads or inside a system-of-record. Organizations are now deploying forward-deployed engineers and building ontologies to capture that context and feed it to the AI. It's labour-intensive work, and it's what changed.
What exactly is a "forward-deployed engineer"?
An FDE is an engineer who works inside the customer's environment rather than on the vendor's product — learning how that specific company actually operates and building the integration that makes the software work there. Palantir made the model famous. It's the opposite of the usual software economics: instead of one product serving everyone identically, each deployment carries substantial human labour. The reason it's reappearing for AI is that the missing ingredient isn't model capability, it's context that nobody ever wrote down.
Why is this worth noticing?
- It's a concrete answer to why enterprise AI results have lagged demos: the model was never the bottleneck, the undocumented context was.
- "Labour-intensive" is a significant admission from a vendor. If capturing context requires embedded engineers per customer, enterprise AI has services-business economics, not software-business economics.
- The ontology framing connects to today's Spatial Memory Agent and MindMemOS papers from a different direction — structured, persistent context is emerging as the thing that separates working agents from impressive ones.