Forward deployed engineer: the career recruiters called job-hopping

Why everyone suddenly talks about it

The role comes from Palantir, which has had engineers building on-site at clients for years. OpenAI and Anthropic have hundreds of openings for it. Aaron Levie, CEO of Box, announced on X on April 30 that Box will hire and retrain internal forward deployed engineers to get agents working on critical business processes.

Uber built a structure around it: agentic pods, where its most AI-capable engineers sprint for two weeks with a domain expert from finance, legal or marketing. Sixteen pods in two months. One pod cut budget decisions across 150 cities from 15 hours to 30 minutes (X thread by Uber’s CTO, July 2026).

The Netherlands is joining in, and not just at startups. Deloitte, Accenture, Capgemini and McKinsey told Dutch financial daily FD that their clients increasingly ask for exactly this profile: engineers who design, build and embed AI solutions in day-to-day operations. The reason, in Capgemini’s words: companies are past the experimentation phase and want to know how to run AI at scale. The big firms are recruiting for it and retraining existing staff (FD, 24 August 2026).

Last spring a Dutch IT director posted a LinkedIn vacancy for an “AI-native colleague”: no traditional role, no fixed department, someone who talks as easily as they build. That is the FDE profile, only the name is missing. My inbox carries the same question from directors: where do I find this person?

I did this job before it had a name

My CV: the app team at KPN, marketing intelligence at insurer Reaal, the acquisition team at Eneco, Head of Data at ride-hailing platform Free Now in Hamburg. Then I co-founded an AI consultancy, which I handed over this summer. I never stayed anywhere longer than two years; recruiters saw a risk.

The work was the same everywhere, only the domain changed: telecom, insurance, energy, mobility, AI. Land in the middle of a business team, learn the systems, build what is needed, bring people along, make myself redundant and move on. Free Now was the same kind of platform as Uber, the company now held up as the textbook example of this way of working.

A career path that fit no org chart in 2015 is the profile companies ask for in 2026.

The same rhythm, four times, in 2026

This is what the work looks like when you express it in dates instead of a job title. Four systems I built myself this year, each at the table with the team that had to work with it. The sector and the dates are here; who my clients are is on the clients page.

Training provider, childcare

Migration of a learning environment. Scoping on 20 July, the migration dossier that same evening, content sign-off on 21 July. 192 non-H5P content pages moved, automated final check with zero errors. H5P content and export acceptance continued into August; there was no handover yet.

Tiling contractor

Web app that extracts bathroom and toilet dimensions from a floor plan. Engagement on 12 May, first client login on 5 June. I continued building until 16 July; the engagement is still active.

Recruitment

Dossier feature in an existing product. Promised on 21 July, live on 29 July. Testing and maintenance continue.

Timber manufacturer, family business

Order-planning tool that reads orders with OCR and a language model. Review set on 4 June, accepted on 5 June and live on 11 June. The evaluation sets for product counts, components and paint allocation scored 100%. Further development is agreed through September, so no exit here yet.

What comes after the application: an agent and an owner

The next step of this trade is already visible. Dan Shipper of Every, which runs its own consulting practice through an agent in Slack, puts it this way: “every agent needs a human” and “the further away an agent is from a human who’s doing it, the worse it does” (TBPN, 22 May 2026). Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, over cost, unclear value or inadequate oversight (25 June 2025).

For a forward deployed engineer that means: what you leave behind is not just a system, but an agent with an owner. Since August 2026 my handover is therefore a fixed list of five. That is the handover I settle on, and the reason I leave after the build.

One owner

One name inside your organisation that owns the system. Not IT, not me.

An evaluation set that keeps running

The test set I build against runs on your machine, not mine. If the score drops, you see it first.

A one-page runbook

What it does, what it must not do, where to look when it stalls.

Who steps in

An agreement on who corrects a mistake, and a button to do it.

A path for the next model version

The model that works today can be replaced within six months. Who retests then, and when.

Honest about where this stands

This list is younger than most of the projects I draw on. None of the four cases above has a completed handover yet where all five points are ticked at the client. That is the missing line of proof; it is here as a standard, not as a result. Ask any agency, me included: what do you leave behind, and who owns it then?

What does a forward deployed engineer earn?

In the US the median base salary is $210,000. The 25th percentile earns $165,000, senior and staff profiles reach $243,000 (Recruiting from Scratch, based on 300+ technical placements, 2026). The Netherlands has no reliable benchmark yet; the supply is too thin.

That scarcity is the real problem. Recruitment firm Christian & Timbers mapped over 17,000 FDEs in the US; only about 2,000 of them have implemented AI at large organisations more than once with demonstrable results (cited in FD, 24 August 2026). You compete for this profile with OpenAI, Anthropic, Palantir and, since this year, the big consulting firms. For most Dutch companies the question is not what this person costs, but whether you can get them, and give them enough work to stay.

When you should hire one in-house

This may cost me an engagement, but the honest answer: sometimes you should simply employ this person. The line is not your company size but the nature of your constraint. Do you have an established IT landscape, your own data team and a sizeable marketing and sales operation, with the delay sitting mostly in communication between departments and the question of who owns what? Then this role belongs inside your organisation: someone who stays, learns the dynamics and can be held to account.

The IT director with that vacancy is doing the right thing. For that type of organisation an external builder is a patch on an organisational question.

And when not: hire the work, not the role

Below that line, say 50 to 500 employees without a data team, this hire is the wrong solution. You will not win the talent fight with the labs. And if you do, that person runs out of things to build after the first bottlenecks are solved.

The big consulting firms see it too. Nicole van Det, CEO of Accenture Netherlands, told FD that AI and tech companies increasingly deploy engineers directly at clients to get from prototype to real use. She calls that a new competitive field (FD, 24 August 2026). Companies with 50 to 500 employees therefore compete with the consulting firms for the same scarce profile.

What you need is the work, not the job: someone who builds in your systems, shows that it runs and then makes themselves redundant. That is what I do. Fixed price, end date, and an honest verdict up front; even when that verdict is that you should hire someone instead.