What is a Forward Deployed
Engineer, actually?

The word doing the work is forward — a military borrowing. A forward-deployed unit is stationed where the action is rather than at home base. Applied to engineering: you are not in the product org waiting for a spec to come back from the field, you are the field. You sit inside the customer’s constraints — their network, their data mess, their security review, their sceptical middle manager — and you build there.

That relocation changes the physics of the job. A product engineer optimises for many users, over quarters, behind a roadmap. An FDE optimises for one customer, over weeks, against a problem nobody has written down yet. Speed beats polish; fit beats generality; a demo that lands beats an abstraction that is correct.

One more distinction that matters for search results: “forward deployed engineer” is now a family of titles, not one job. FDE, Forward Deployed Software Engineer (FDSE), Deployment Strategist, Applied AI Engineer, Agent Engineer, Solutions Engineer (AI) and Forward Deployed AI Engineer all describe overlapping slices of the same work. Section 03 untangles them.

Read those four numbers together and the story writes itself. Enterprises bought AI. Most of the pilots did not turn into money.¹⁷ The diagnosis the industry converged on is that the model was never the bottleneck — the last mile was: the customer’s data is messy, their systems are old, their security team says no, and nobody on either side had defined what “working” means precisely enough to test. Fixing that is not a product feature. It is a person, on site, with commit access.

So the labs started buying the person. In May 2026 OpenAI stood up a dedicated deployment company with a $4B raise at a $10B valuation.⁵ In July 2026 Anthropic and Blackstone announced “Ode”, a $1.5B AI-deployment venture.⁶ Microsoft committed $2.5B to a 6,000-expert Frontier organisation embedding engineers with customers.⁷ AWS built a Forward Deployed Engineering practice. a16z gave the pattern a name — services-led growth — and argued that trading gross margin for deployment depth is a defensible moat, not a spreadsheet mistake.⁸ TechCrunch called FDEs “the AI industry’s latest talent obsession” on 30 July 2026.²

A snapshot of open roles from one 2026 tracker gives the shape of the market: Palantir 51 openings, OpenAI 31, Databricks 12, Mistral 11, Cohere 10.⁴ Frontier labs and data platforms lead; the long tail is vertical AI startups who discovered that their product only works after somebody wires it into the customer.

An analysis of a thousand FDE job postings found the title splits about 60 / 30 / 10 between builder roles, pre-sales roles, and internal ones.¹ They are genuinely different jobs with different day-to-day work, different interviews, and different exits.

Anthropic’s Applied AI / forward-deployed job description describes the split roughly as 40% building, 30% customer-facing work, 30% internal feedback — carrying what you learned at the customer back into the product.¹⁰ That third bucket is the one outsiders never predict, and it is where a lot of the leverage lives: an FDE is the highest-bandwidth sensor a product team has.

Note the shape: you are never blocked by one kind of work. When the security review stalls, you build. When the build is waiting on data access, you go sit with a user. Engineers who need long uninterrupted blocks to feel productive tend to bounce off this; engineers who are energised by unblocking a stuck thing tend to thrive.

Across the whole market, one 2026 census put the median posted base salary at $185K for strict-definition FDE roles.³ That number and the eye-catching “$1M FDE” screenshots are both true — they describe different tiers of a very wide distribution, and reconciling them is its own page.

Four adjacent jobs get confused with this one. The short version, with the full breakdowns one click away:

There is one more that people rarely name: an FDE is not a support engineer. Support is reactive and ticket-scoped by design. An FDE arrives before there is a system to support, and leaves once someone else can run it.

First Round’s guide for companies hiring FDEs makes the point bluntly: the profile you are looking for is staff-engineer material who also wants to talk to customers, not a junior who can be spared from the product team.¹¹ The Pragmatic Engineer’s deep dive on the role reaches a similar conclusion about the seniority the work demands.¹²

The reason is structural. In a product org, hard decisions get distributed — an architect, a staff engineer, a PM, a security team, an SRE rotation. At a customer site there is one of you. You will decide the data model, negotiate the security exception, judge whether the eval is good enough to bet a renewal on, and be the person paged when it breaks — sometimes in the same week.

Interviews reflect that. Palantir’s loop leads with a decomposition round — an open, non-coding problem you have to scope out loud. OpenAI’s process centres on a multi-hour take-home you then have to defend on video. Anthropic’s has a values round that fails more candidates than the coding does. Roughly half the loop, at most companies, is not algorithms at all — which is exactly why generic interview prep does not transfer. The interview prep hub covers each round with practice you can actually run.

Three genuine risks, and what each one is actually worth worrying about.

The strongest argument for durability is the one the money makes. Roughly $8B of institutional capital was committed to deployment-services organisations in about ten weeks of 2026.⁵⁶⁷ Institutions do not build 6,000-person practices for a title that evaporates in a year.