The AI job nobody wants (and why it's the right bet)
TL;DR
The fastest-growing role in AI isn't "prompt engineer" or "researcher." It's the Forward Deployed Engineer: the one you send to the customer to make the thing actually run. Postings 10×'d in a year, and yet few devs want it. That reluctance is precisely the signal.
There’s an implicit prestige hierarchy among devs. At the top, research and product at scale. At the bottom, anything customer-facing: support, integration, deployment. We want to build the rocket, not land it on the customer’s muddy strip.
Except landing is exactly what the whole industry is missing right now.
The role the labs are fighting over
The Forward Deployed Engineer (FDE) isn’t new — Palantir invented it in the early 2010s, to the point of having more FDEs than product engineers. The idea: an engineer embedded with the customer who doesn’t hand in a report but builds the system and stays until it runs in production.
What’s new is that AI labs started copying the model at full speed. In 2026, OpenAI stood up a dedicated deployment entity, Anthropic a joint venture of the same kind. The reason is repeated verbatim by their leaders: enterprise demand outstrips any single delivery model. Translation: models don’t deploy themselves. Someone has to wire the agent onto the real tools, the real data, with the real guardrails.
In France, postings for this role have grown roughly tenfold in a year. The AI freelance market tells the same story: most requested work isn’t machine learning — it’s LLM integration, RAG, agents in production. The “make it actually work” part.
Why nobody wants it
Because it’s uncomfortable. Travel, urgency, customer environments you discover as you go, and that prestige deficit I mentioned: in many devs’ heads, being customer-facing is less noble than coding a product seen by millions.
The result: exploding demand against anemic supply. In economics that has a name, and it isn’t “dead end.” It’s a rent for those willing to show up.
The bet
Here’s why I think it’s the right bet, with the appropriate level of certainty: it’s a reasoned conviction, not a prediction.
The agent obliterated the “writing code” part. What stays scarce — and becomes more so as models improve — is everything the agent doesn’t do: framing a badly-posed problem, wiring a system onto a reality that pushes back, deciding what ships, making a team actually use it. These are judgment and contact skills — exactly the core of the FDE, exactly what doesn’t commoditize.
Put differently: the job with the least prestige is the one with the most residual value once AI has eaten the rest. Code becomes free; discernment and deployment don’t.
And there’s an angle the big labs’ job posts forget: this role isn’t reserved for their teams, nor for permanent travel. The underlying skill — wiring AI onto an organization’s reality and making its people autonomous — holds at every scale: an SMB, a mid-market firm, in-house or as a contractor, close to home and on chosen projects. It’s one of the few agentic skills that keeps its full value outside the big shops.
What it changes if you’re a dev today
Stop chasing the most prestigious title. Look where the remaining value is: at the last mile, where the agent stalls and the customer struggles. Learn to frame, to deploy, to drive adoption — enablement in the literal sense — and to talk to people who’ll never read your code. These are the skills nobody wants to build and everybody will pay for.
The job nobody wants is becoming the one everybody fights over. That’s usually the best time to get into it.