Forward Deployed Engineer

The engineer who shapes the build inside a customer’s environment — then ships it. Sonitec recruits FDEs for companies deploying AI across Australia and APAC.

The seat, defined

A Forward Deployed Engineer works where the customer works. They sit inside the customer’s environment — its systems, its data, its people — and shape what should be built before a line of it exists: which workflow the AI belongs in, which steps stay deterministic software, where a human stays in the loop. Then they ship it. Not a recommendation, not a demo — working software running against the customer’s real systems.

That two-part motion — shape the build in context, then take it to production — is the whole seat. It is what separates the FDE from a solutions engineer who supports a sale, and from a consultant who leaves a deck behind.

Eval and production separate the seat

Anyone can make a model look good in a demo. The FDE seat is accountable for evidence and for the system after go-live. Two disciplines mark out a genuine FDE:

Eval-driven development. Golden datasets, regression suites, accuracy/latency/cost benchmarks — the build is proven against the customer’s own cases before anyone is asked to trust it.
Operating in production. Guardrails, failure routing, monitoring, audit trails — and ownership of the system once real users depend on it, not a handover at the demo.

If a candidate’s history stops at proofs-of-concept, they are not yet an FDE — whatever the title on their CV says.

How FDEs work

Coding agents are how FDEs work — not the job title. A strong FDE uses agents to move through the build at a pace a customer deployment demands, and stays accountable for everything that ships: the architecture, the evals, the failure modes. The tooling accelerates the seat; it doesn’t define it.

Full 2026 AUD salary bands, city and sector premiums, and the hiring standard we interview against are on our FDE salaries & hiring standards page.

Seat Context

The four skills behind the seat

We frame the FDE seat against Andrew Ng’s AI Engineering Skills Map — four parent skills, held together in one person. It is context for the seat, not a checklist.

Building & deploying AI applications

The technical core of the seat, in its applied form:

LLM foundations — working knowledge, applied not academic
Grounding with data — RAG and retrieval against the customer’s own sources
Agentic systems — tool calling, memory, structured outputs
Eval-driven development — the build is proven, not asserted
Operating in production — monitoring, guardrails, cost and latency
ML foundations — enough to know when an LLM is the wrong tool

Software engineering fundamentals

APIs, cloud, version control, testing — typically Python and/or TypeScript. The AI work sits on ordinary engineering discipline, and it shows immediately when that is missing.

Using coding agents effectively

Directing agents through real codebases at deployment pace — while owning the review, the evals, and the result. A working method, not a role.

Shaping the build

Deciding what to build with the people who will use it — mapping the real workflow, finding where AI earns its place, and saying no to the steps where it doesn’t. This is the skill the customer environment demands most.

How We Qualify

What we interview against

Sonitec is a specialist AI recruitment agency. We staff this seat — we don’t build your AI. Every FDE we put forward is interviewed against the seat as defined on this page.

Shipped in someone else’s environment

Evidence of building inside a customer’s or another team’s systems — integration with what exists, not a greenfield portfolio piece.

Eval discipline

They can describe how they proved a system worked — the datasets, the regressions, the benchmarks — before it was trusted with real work.

Production ownership

They stayed with the system after go-live: the failure modes, the monitoring, the cost curve. Demo-only work is identified as such.