Forward deployed engineer

What is a forward deployed engineer?

An engineer who works inside your codebase, on your problem, for a fixed period.

The title comes from Palantir and has since been adopted by the AI labs. The common thread is that the person designing the system also writes it, in the customer’s environment, against the customer’s real data. The alternative is an architecture document handed to a team that then discovers the assumptions were wrong.

Clemence W. Chee sells this alongside two executive engagements. The three are the same scope at different altitudes: what to build, who owns it, and who writes it.

When does this shape fit?

Three situations, all of which look like a stalled build rather than a missing strategy.

You know what you want to build and the team has never shipped a retrieval system before. Your prototype demos well and nobody can say whether it is correct. Or the AI work keeps losing to roadmap items because no engineer is ring-fenced for it.

If instead the problem is that nobody has decided what to build, the three-week teardown is the cheaper answer and it ends in a decision.

What do the first two weeks look like?

Week one is read-only. I take repository access, run the system, and write down what is actually there.

The output of week one is a short note listing what I found and what I propose to build first, sized so that something is in production by week four. If that note says the honest answer is smaller than the engagement, I say so and we scope down.

Week two is the first commit. Shipping something small and real early is what separates this from a consulting engagement, because it forces every hidden constraint into the open while there is still time to respond to it.

What do you actually build?

The layer underneath the model, and the agent on top of it.

In practice: the retrieval and context layer, the permission checks that hold at query time, the evaluation suite, and the agent itself. I work in your language and your framework. I do not introduce a new one so that the system depends on me after I leave.

What I do not build is your platform. Kubernetes, incident response, the CI system itself, identity infrastructure and networking stay with the people who own them, and if those are the hard problem then I am the wrong call.

What does this cost?

Quoted per engagement, based on days a week and duration.

The engagement runs 6 to 12 weeks at 3 to 4 days a week, so the total is between 18 and 48 days of my time. Knowing that range is usually enough to tell whether this is the right instrument before either of us spends an hour on a call.

I publish a minimum rather than a day rate. A day rate invites comparison against agency-supplied contractors, and the comparison is misleading here: the work is scoped to end with a system in production, not to fill a seat for a quarter.

I do not work through intermediaries on a commission basis.