The Companies Hiring FDEs

The last lesson set out what an FDE does. This one is about who is paying for one, and what their job descriptions give away about the specific work you would be doing.

Reading JDs well is itself a small craft. The paragraph a company spends the most words on is the one it has had the most trouble hiring for. The platform names in the "you will" section are the ones you will be measured against in the loop. And the travel line is not boilerplate, which is a mistake almost every candidate makes the first time.

The three archetypes

The hiring surface has sorted itself into three groups, and it is useful to see them before any single company name, because the group tells you more about the work than the logo does.

An honest aside before the names. The claim that Palantir "coined" the term is contested — the phrase has older roots in defence contracting — but Palantir is where the role as it exists today took its current shape, and the alumni network from Palantir is why you now see the title at Sierra, Anduril, Scale and so on. Saying Palantir popularised FDE is accurate; saying they invented it is only partly so.

Who is hiring, by group

Frontier labs. Anthropic's FDE job family has, as of 2026, a steady stream of openings across SF, New York, London and Dublin, and the public JDs lean hard on MCP server authoring, sub-agent design, and skills as an on-demand loading mechanism. OpenAI's FDE postings are decentralised by team — one written by the API Platform group reads differently from one written by the Enterprise group — but the common thread is eval-driven iteration and an explicit on-call rotation. Both publish rubrics; both say up to 50% travel.

Applied-AI companies. Sierra (customer support agents), Glean (enterprise search), Harvey (legal), Decagon (support), Distyl (vertical rollouts), Cognition (Devin-style code agents), Ramp (fintech) and a few more. The JDs read like product engineering with a customer-facing overlay: the applied layer is the thing, the model behind it is somebody else's. The domain in the title is the signal — a Harvey FDE will be judged on legal-specific instincts, a Glean FDE on retrieval over messy enterprise data.

Palantir and descendants. Palantir's FDE loop is the longest-running of any employer in this list, and its JD style is deliberately short on specifics — the framing is that you will walk into a situation nobody has fully described yet and walk out with a plan. Anduril, Shield AI and other defence-tech companies read similarly. The framing is less about the model stack and more about engineer-as-consultant inside an ambiguous mission.

Infrastructure. Databricks and Scale AI occupy a third shape — their platforms sit under the applied layer. Databricks posted on the order of two dozen GenAI-adjacent FDE openings in late 2026 across its field engineering org, most of them focused on helping a customer get a production RAG or agent system shipped on Mosaic AI. These JDs tend to read as the most technically concrete of the three groups.

What each group presses hardest on

Not every FDE loop is the same loop, and the group is a strong hint about which parts of this course you will be tested on most heavily.

• Anthropic presses on the platform: MCP, sub-agents, skills, the agentic loop, and prompt caching as a cost lever. Expect to be asked to design the extension boundary — what you expose versus what you refuse — and to defend it to a security reviewer.
• OpenAI presses on evals above all else. The eval round is the one candidates most reliably fail, and the one the loop weights heaviest. It also presses on-call readiness and the production debugging instinct.
• Palantir presses on the ambiguous customer problem: given a mess, structure it; given a stakeholder, read them; given a plan, defend it under adversarial questioning. The technical bar is real but the standout round is the ambiguity one.
• Applied-AI startups press on the vertical and on speed. A Harvey loop will press on legal fluency; a Sierra loop will press on customer service as a discipline. "Can you ship something useful in a week" is an explicit evaluation at several of these.

Reading comp and volume figures honestly

Three independent 2026 data pulls on this role — from Perspective, Dexity and Levels.fyi — give three different totals for both the hiring volume and the comp band. The medians agree directionally, the tails do not. The useful posture is: treat any single number as a signal, not a fact. Before quoting a figure in a customer or recruiter conversation, triangulate against the company's own career page and one other independent source. Quoting a number you heard at a dinner party will catch up with you.

The hiring volume, honestly stated: as of 2026 the role is being hired at every scale from a single opening at a seed-stage startup to dozens of simultaneous openings at frontier labs and infrastructure providers, and the rate is still rising. By the time you read this the specific numbers will be stale; the shape of the market will not.

How to use a JD as a study plan

A practical habit worth building now. Pick the three companies you would most like to work for. Open their current FDE job descriptions side by side. Read the "you will" and "what we look for" sections only. Every bullet is a chapter of this course, and the bullets they repeat are the ones they over-index on in interviews.

If "designs eval suites" shows up in two of three, Part 6 of this course is the one to memorise first. If "authors MCP servers" shows up, Part 5. If "scopes POCs with executive stakeholders" shows up, Part 8. The JD tells you what the loop will test. The loop almost always does.

What to take away

The FDE role sits across three archetypes — frontier labs, applied-AI startups, and Palantir-shaped consultancies — and the archetype of the employer tells you more about the work than the logo. Anthropic presses on the platform, OpenAI on evals, Palantir on ambiguity, applied-AI companies on their vertical and on speed. Comp and hiring-volume figures from third-party sources are noisy enough that you should triangulate before quoting them. The job description is a study plan, and the bullets a company repeats across roles are the ones they will test hardest on in the loop.

Next: the three job titles an FDE role is most often confused with, and how to tell them apart in a job posting.