The New Agentic FDE

The Palantir product was a platform for integrating messy, sensitive, real-world data (counterterrorism records, fraud signals, supply chain telemetry, etc), and no two clients had the same data, the same permissions structure, or the same operational reality. Shipping a generic product into that environment didn't work. So Palantir embedded engineers directly inside client sites, gave them read/write access to live systems, and made them responsible for turning the platform into a working solution against the client's actual problem. The FDE was part software engineer, part solutions architect, part product manager, part customer-facing operator. The role was a response to a specific kind of complexity that resisted productization, and for fifteen years, it was treated as a Palantir quirk rather than a general pattern. It could be argued Palantir was a first scaled example of the convergence between services and software (https://alten.capital/blog/when-services-revenue-starts-to-look-like-software).

Two things changed. First, foundation models commoditized the core capability that AI startups were selling. When every vendor has access to roughly the same frontier models, the moat is no longer the model; it's the implementation work that wires the model into a client's data, permissions, regulatory architecture, and workflows. Second, the kinds of products AI companies are trying to sell: agents that automate sales, support, legal, claims, and underwriting, require exactly the kind of embedded, context-rich deployment work that Palantir had been doing all along. The result is that the FDE has gone from a Palantir-specific job title to one of the most-hired roles in enterprise AI. HFS Research, writing in March, argued that 93% of enterprises are stuck in AI pilot purgatory, not because the models aren't good enough, but because no one is doing the FDE work that turns pilots into governed production systems.

The version of the role that's emerging now is recognizably the Palantir FDE, but the technical substrate is different. The new FDE works in agents. They design the agent loop, build the eval infrastructure that catches the failure modes humans miss (a confused human escalates; a confused agent produces a confident, plausible, wrong answer at scale), engineer the context that goes into each model call, wire the agent into governed data with the right permissions, and build the human-in-the-loop override paths that regulated industries require. FDEs do this on-site, talking mostly to operations managers rather than other engineers, translating "we want AI that figures everything out automatically" into "let's start with these specific inputs and outputs." Some are hired under the title FDE, others as AI Agent Engineers, Solutions Engineers, or Agent Product Managers. The titles are converging on the same role.

For technology services firms, this is the most consequential talent pattern of the cycle, and it's not really about hiring. The firms that figure out how to industrialize the agentic FDE by training them at scale, deploying them across portfolios of clients, building the reusable ontologies and eval libraries, and workflow templates that compound across engagements, will capture the layer of the market that HFS calls "build and integrate" and "run and govern," where switching costs accumulate and margin density improves over time. The firms that treat agent work as a tooling decision, staffing existing developers against it without restructuring the delivery model, end up selling hours into a market that is rapidly repricing toward outcomes, which we wrote about in https://alten.capital/blog/from-time-to-outcomes.

Alten Capital invests in technology services businesses. Reach out to explore potential partnerships.