Building a forward deployed org when agents do half the job

Here is the strangest number in Bloomberry’s analysis of 1,000 forward deployed engineer job postings: 58% came from companies with 11 to 200 employees. The title Palantir invented for a two-thousand-engineer field army is now being hired for, overwhelmingly, by teams that cannot afford the org it came from. Postings grew 1,165% year over year through October 2025, on top of a quintupling in 2024 per Lightcast data. Most of these companies are about to staff the role as it existed in 2020 — half of which is now automatable — and the miss will cost them their two or three most expensive hires.

What the market is hiring under one title

The title covers three different jobs. The share data below is Bloomberry’s; the reads in the right two columns are ours.

Variant	Share of postings	What it really is	Watch out for
Builder FDE	60%	Production engineering at the client boundary	The real thing; hardest to hire
Sales Engineer+	30%	Pre-sales with a commit bit	Fine, but don’t expect them to carry implementations
Internal tools builder	10%	Ops engineering wearing the trendy title	Title inflation

Two more numbers from the same dataset: 47% of postings explicitly require customer-facing ability, and none carried a quota (listings don’t always disclose comp structure, but the direction is unambiguous). The market seems to have figured out this is not a sales role. It’s an engineering role where the compiler is another company’s org chart.

What actually predicts success

The skills list — Python in 66% of postings, TypeScript in 35%, agent and LLM experience in roughly a third — is table stakes, and tells you little about who survives contact with a client. Palantir’s two decades of selection are more instructive. Qureshi’s account emphasizes generalists in small autonomous teams of four or five, tolerant of travel and ambiguity, able to read a room and navigate a client’s politics. The output of that filter tells you what it was selecting for: ex-Palantir employees show up more often in YC batches than ex-Googlers, despite Google employing roughly 50x more people. You are hiring proto-founders and pointing them at other people’s companies.

Our own screening addenda, for what they’re worth: curiosity about the client’s business rather than the technology, comfort being wrong in public, and the ability to ship something embarrassing in week one. A brilliant engineer who needs a spec will sink without a trace.

The part that changed

Break the FDE week into two piles. Pile one is judgment: discovery, integration architecture, the call on what to customize, earning a skeptical client’s trust. Pile two is coordination: status updates, approval-chasing, meeting-note archaeology, the weekly email, keeping five stakeholders on two sides pointed at the same plan. Pile two consumes far more of the week than anyone budgets for — and pile two is the one agentic tooling now carries. (That’s the half Gravel does: agents work the board, chase the blockers, and write things down, while your engineers stay in Linear and GitHub.)

The org-design consequence: stop staffing for pile two. Hire fewer, more senior people, and change the unit of capacity from accounts carried to engagements supervised. Our staffing model — a design target we’re measuring with design partners, not an audited benchmark — looks like this:

Stage	Old playbook	With agents on coordination (our model)
Founder-led, zero hires	Founder does deployments in the margins of the week	Founder plus an agent crew is a functioning deployment team
First FDE hire	Senior generalist, half their week lost to status work	Same hire supervises up to six engagements
Pod of 3–5	Carries 6–10 accounts	Carries the pipeline you currently decline; one member owns productization full-time

One discipline survives from the pre-agent era, and Andrusko states it well: time-box engagements, hold a ratio of field effort to ARR, and systematically convert bespoke work into templates. Agents lower the cost of the motion; they don’t repeal the rule that field effort per account has to fall over time. They’re how a small team finally gets to enforce it.