Forward Deployed Engineers: what they are and why everyone is hiring them

A Forward Deployed Engineer (FDE) is a senior engineer embedded in the client's operation who writes production code and owns the business outcome, not a deliverable. The role was invented at Palantir more than a decade ago, and in 2026 it became the standard delivery model for enterprise AI: FDE job postings grew more than 1,000% year over year (GetPerspective, 2026).

Where the role comes from

Palantir organized its delivery around two roles: engineers who sat with the client and shipped code against real data, and strategists who defined the use case, the metric and the business case. Wonderful and HappyRobot copied the titles word for word. Distyl was founded by former Palantir executives. Primero's co-CEO came from 8VC, the fund of a Palantir co-founder. The template is the same everywhere because it works.

What an FDE actually does in a week

• Monday: reads the client's ticket queue or call logs, not a slide about them. Finds the twelve cases that break the agent.
• Tuesday: writes the integration with the ERP or the CRM in the client's repo, with the client's CI, under the client's permissions.
• Wednesday: runs the eval suite against last week's real conversations, fixes the two regressions, ships.
• Thursday: sits in the operations stand-up. Hears that the refund policy changed on Friday. Updates the context layer the same day.
• Friday: demo with the KPI on screen. Not a status report. A number.

The point is proximity. An FDE is inside the ceremonies, the data and the politics of the process being automated. Most of the failures behind the 95% of pilots with no P&L impact (MIT NANDA, Aug 2025) happen in the gap between the vendor and the operation. The FDE closes that gap by standing in it.

FDE, consultant, staffing

Three models look alike on a rate card and behave differently in production.

• A consultant owns a recommendation. The deck is the deliverable. Production is someone else's problem.
• Staffing owns hours. You get a good engineer and you manage the outcome yourself.
• An FDE owns the outcome. The pod signs a KPI and is measured on it. If the number does not move, the engagement does not scale.

Alvarez & Marsal (Apr 2026) describes the cost of that third model: lower gross margins and headcount that scales with revenue. Wonderful went from 120 to 650 people in ten months. The model is expensive for the vendor, which is exactly why it is valuable to the client. Someone is paid to be accountable.

Why the labs institutionalized it

In May 2026 OpenAI launched The Deployment Company with more than $4B in initial investment and Anthropic launched Ode with $1.5B. Both are services companies. Their thesis is that the last mile of enterprise AI is engineering inside the client, not another API. a16z opened an FDE Fellows program in July 2026. Nobody does that for a fad.

The pod, not the hero

An FDE alone is a talented contractor. A pod is a delivery unit. Ours has three roles: a Forward Deployed Engineer who ships the integration and the agent, a Deployment Strategist who defines the case, the baseline and the KPI with the client, and an AI Engineer who builds evals, guardrails and the context layer. The pod works in your repo, your cloud and your ceremonies, and takes one process from discovery to production in 8 to 12 weeks.

What to look for when you contract one

• Production record: ask for the last three things they shipped that are still running.
• Data comfort: an FDE who cannot read your SAP tables is a prompt engineer.
• Eval literacy: if they cannot explain how they know the agent got worse, they will not notice.
• Exit discipline: the best FDEs leave documentation, runbooks and a team that does not need them.

The last point matters most. The job of an FDE is to become unnecessary. If a vendor's model depends on staying inside your operation forever, it is not forward deployment. It is staffing with a better title.

What it looks like from your side

You give the pod a seat in the stand-up, access to the repo and the sandbox, and one person who can answer policy questions the same day. You get weekly demos with the KPI on screen, code in your GitHub from week one, and at the end of the pilot a two-day workshop, the runbooks and an exit plan we can execute in five working days. Your team keeps everything. We would rather you stay because it works.