A forward deployed engineer is a software engineer who works directly with a client’s teams: they observe the work, build the solution in the client’s systems and put it into production. The model comes from Palantir; AI labs such as OpenAI have adopted it to put their models to work for their customers.
Where the model comes from
Palantir distinguishes two engineering jobs: those who develop its platforms, and those who work with clients, the “Forward Deployed Software Engineers”. The latter start from an often vague question asked by the client and build the answer with them, from design to data integration.
The model spread with generative AI. OpenAI, for example, hires “Forward Deployed Engineers” who lead a customer project from discovery through to production rollout.
What changes compared with consulting or classic development
- No translation between need and code. The person who listens to the teams is the person who builds. The nuances of the business do not get lost in a specification.
- The real work, not the theoretical process. The engineer sees how cases are actually handled, exceptions included, and builds for that work.
- Short, measured deliveries. Every week, something runs on real data; what does not work is fixed while the context is fresh.
- Production as the goal. The deliverable is not a recommendation or a demo, but a system running in your tools, with its access controls and guardrails.
Why it matters even more with AI
An AI agent cannot be fully specified in advance. Its quality depends on details you only discover by confronting it with real cases: an unexpected invoice format, an exception everyone knows but nobody wrote down, a rule that changes from one customer to the next. The forward deployed engineer is there when those details appear.
They also build the evaluation sets with the teams: real cases whose right answer is known, replayed before every change. That is what lets an agent evolve without breaking what worked.
When to choose this model
| Situation | Forward deployed engineer | Other approach |
|---|---|---|
| A process specific to your company, data in several systems | A good fit | A standard tool will handle your exceptions poorly |
| A vague need, to be clarified with the teams | A good fit | A specification will freeze the wrong assumptions |
| A standard task, a market tool already does it well | Of little use | Buy the tool |
| A stable, complete specification, an internal team available | Possible | Classic development |
At Extra Enterprise
Every build is led by a forward deployed engineer alongside your teams, from diagnostic to production. Then the run takes over: monitoring, evaluations and changes, with the same team.