What Is a Forward Deployed Agent (FDA)?

TL;DR: A Forward Deployed Agent is how a Decision Graph gets built for a specific customer in two to four weeks. The work is elicitation then encoding: sitting with the tech lead to learn what the company's words actually mean, then translating that into typed entities and live ETL across their systems. It's a role rather than a tool because the encoding step needs judgment no automated mapper can make.

#What an FDA is, in one paragraph

A Forward Deployed Agent (FDA) is the delivery model Uvi uses to ship a Decision Graph for an enterprise customer in two to four weeks. The FDA is not a project manager or a generalist consultant. They are a specialist in ETL, data science, and the customer's particular toolchain.

They sit with the customer's tech lead for two to four hours of elicitation, then encode the resulting entity definitions and joins across code, tickets, PRs, Slack, and internal services. The output is a queryable graph exposed via GraphQL and MCP, ready for the customer's agents to call.

#The job, week by week

Week 1, elicitation. Sit with the tech lead, walk through ten real questions an engineer asks today, record every implicit reference type: branch-name conventions, flag-key patterns, Slack thread habits.

Weeks 2 to 3, encoding. Translate the elicited language into typed entities and live-ETL plumbing across the source systems.

Week 4, verification. Run a blind benchmark on the ten elicitation questions, iterate the entity definitions until the agent returns the right chain.

End of week 4: the GraphQL surface and MCP server are locked, and the customer's agents can call against production.

#Why this is a role, not a tool

The encoding step requires judgment a tool cannot make. "What is a feature here, in your words?" has a different answer at every company. A naive automated mapper would either over-generalize, and lose precision, or over-fit, and lose generality.

An FDA is a human who has done this twenty times. They know which questions to ask, which patterns transfer, and which to encode as customer-specific. The Decision Graph stops being a generic framework and becomes the customer's actual decision substrate.

#FDA vs professional services vs solutions engineering

Professional services typically delivers a configured product within the vendor's existing object model. Solutions engineering supports pre-sale evaluation.

A Forward Deployed Agent does the work that lives between: shipping production infrastructure that encodes the customer's specific reality. Palantir originated the model, and it's the right shape for any platform whose value depends on capturing a customer's hidden vocabulary.

Frequently asked questions

Is the FDA a Uvi employee or a contractor?
Uvi employee. The role requires deep knowledge of the Uvi platform, the customer's tools, and the standard patterns of decision encoding. Contractor FDAs would have a learning curve that defeats the two to four week timeline.
Does the FDA stay involved after deployment?
An on-call relationship continues. The Decision Graph definitions evolve when the organization's vocabulary changes, typically once a quarter. The FDA owns those updates, which is a few hours of work rather than weeks, and stays the customer's named contact for the platform.
How is the FDA different from a consultant?
A consultant produces a deliverable, a report or a slide deck or a recommendation, and leaves. An FDA produces working infrastructure that the customer's agents call against in production. Their KPI is the success rate of the agent on the customer's real questions, not a project completion checkbox.
What does FDA mean elsewhere?
Palantir uses Forward Deployed Engineer (FDE) for a similar role: engineers embedded in customer workflows to operationalize the platform. Uvi's FDA is the same model adapted to AI agent delivery, with the same proximity to the customer and the same end-to-end accountability, focused on encoding the Decision Graph.

One layer, underneath every agent.

See what changes when your agents stop guessing.

Book a demo