The deal dies between the demo and the workflow

Enterprise AI deals don’t die in the demo. They die in the six months after it — in data access requests, security reviews, and the gap between what the product does and how the buyer’s team actually works.

The most-quoted evidence is MIT NANDA’s 2025 finding that ~95% of generative AI pilots produced no measurable P&L return. Treat that headline with care — it rested on 52 interviews and a six-month ROI window, and critics have picked at both. The subfinding is the interesting part anyway: pilots built on purchased tools and vendor partnerships succeeded about 67% of the time; internal builds, roughly a third of that. And unlike the headline, this one has corroboration from revealed preference — vendors are behaving exactly as if deployment is where deals are won. Job postings for forward deployed engineers grew 1,165% year over year through October 2025, per Bloomberry’s analysis of 1,000 postings. The report’s diagnosis, the “learning gap,” matches what any implementation team could have told you: tools that never adapt to the buyer’s workflows stall, whatever the model underneath can do.

So the market has converged on an answer with a job title: put an engineer at the boundary between your company and your client’s, and have them drag the pilot into production. The awkward part is who’s converging. In Bloomberry’s data, 58% of FDE postings came from companies with 11–200 employees — vendors trying to run Palantir’s motion without Palantir’s contract sizes.

The motion works. The economics don’t.

	Number	Status
Median FDE base salary	$173,816	Sourced — Bloomberry; 70% of postings add equity
Base grossed up for benefits (~31% of comp)	~$252K	Sourced — BLS ECEC, all-civilian average
Plus travel, tooling, management	$280K–$400K fully loaded	Our model; top end assumes senior markets and equity
Concurrent engagements per FDE	~2	Our working assumption; complex accounts run below that
Field cost per engagement, per year	$125K–$200K	Model output

If your ACV is $80K and each account consumes six figures of senior engineering attention per year, you’re paying for the privilege of winning. The motion only pencils when contracts comfortably clear the field cost — which is why a16z’s Marc Andrusko observes that companies gladly trade gross margin for momentum on seven-figure deals, and why everyone below that line runs “white glove” as heroics: one senior engineer doing deployment work off the books, at night, unmeasured.

What we’re actually paying $400K for

Audit an FDE’s week and a startling share of it is not engineering and not judgment. It’s coordination: chasing a client-side approval that’s been sitting for four days, retyping status into a weekly email, reconciling the third fork of the integration doc, re-answering a question the last engagement already answered. Nobody has published a rigorous time split for this role — including us, yet; it’s one of the things we’re instrumenting design-partner engagements to measure. But the pattern is consistent everywhere we’ve looked, and it points at the only lever that changes the table above: the coordination is exactly the class of work agents can now carry.

That’s the premise of Gravel. Gravel is a system of record for the forward deployed motion. It ingests the traffic an engagement already produces — client email, shared Slack and Teams channels, meeting transcripts, docs — and agents keep four things current on a surface both sides share: the ticket board, the timeline, the knowledge base, and the goals the deal was sold on. Agents draft; your team approves what the client sees. Blockers on the client’s side get chased daily instead of when a human remembers. Nobody writes the weekly status email again.

The result is a different denominator. Instead of one engineer carrying two accounts, one person supervises six engagements — our design target, not yet an audited benchmark, and we intend to publish the real distribution. Same judgment, most of the coordination labor removed, and the motion that converts enterprise pilots stops being gated behind seven-figure contracts.

The demand side has already moved. The constraint left standing is cost. That’s the one we built for.