Five ways to organise FDEs, and which one fits

Ask how FDE teams should be organised and the answer usually starts with headcount. Headcount explains the form a firm already has, not the one that fits it.

Two questions decide the form. Who owns the customer relationship, and who carries the on-site cost. The first decides whether field evidence can change a roadmap; the second, how long anyone can afford to sit in a customer's office before the account pays for itself. Those two predict the failure mode better than revenue or headcount.

Five shapes, and the evidence they leave

The in-house form is what Palantir built: two engineering tracks, Delta and Echo, and now a documented "AI FDE" in Foundry — an agent that operates the platform through conversational commands. It is the only form in which field evidence and the product roadmap share an owner. Its failure is structural: one strategic customer can pull the roadmap into bespoke work.

The partner network buys capacity and channel at once. OpenAI's Deployment Company, announced 2026-05-11 in Boston with OpenAI as majority owner, more than $4bn of initial investment and 19 partners, arrived with roughly 150 experienced forward deployed engineers and deployment specialists through the Tomoro acquisition. The failure mode is variance: two partners deliver two different products, and the brand carries both.

Consultancies and system integrators have the strongest project machinery and the weakest reason to leave an asset behind. Bain, Capgemini and McKinsey invested in DeployCo; Accenture, Deloitte, McKinsey and Morgan Stanley appear among the nominating firms for Anthropic's Frontier Academy — $100m, a target of 10,000 FDEs by the end of 2027, a 12-week residency, nomination only, no badges issued yet. A day-rate organisation is paid to finish a project, not to leave a customer a reusable eval set. Reversion to traditional outsourcing is not a risk here; it is the default.

The boutique studio is several strong generalists and no bench. One operator's self-report puts the workable band at 50,000–200,000 yuan: a 300,000-yuan order holds a small team for one or two months, above 200,000 procurement takes over, and he estimated the odds of the final payment not arriving above 60%. The failure mode is the founder: remove one person and the studio has no delivery.

The enterprise-internal team is the only form where the customer relationship is already owned. A practitioner inside a large company negotiated 50% of his working time in the field and found that two or three people in the whole organisation carried the full chain of judging demand, building and driving adoption. The failure mode is the ticket queue: internal FDEs are the nearest available engineers, so every urgent request lands on them first.

The choice table

Form	Who owns the relationship	Who carries the on-site cost	Failure mode	Choose it when
In-house	The vendor, through named account owners	The vendor, capitalised against a strategic account	One customer bends the roadmap	You have repeatable scenarios and want the evidence to feed a product
Partner network	The vendor owns the brand, partners touch the account	Shared, usually the partner until renewal	Inconsistent quality across partners	Demand outgrows hiring, and you can certify delivery
Consultancy / SI	The consultancy, in a project frame	The consultancy, recovered by day rate	Reverts to outsourcing; no asset stays	The buyer wants a programme, an audit trail, and someone else to carry risk
Boutique studio	The founder personally	The studio, from margin	Founder dependency	Scenarios are narrow, the buyer small, speed beats scale
Enterprise-internal	The business unit	The employer, as salary	Drowned in internal tickets	Systems are sensitive, permissions complex, value is reuse across units

The reporting line decides what the role becomes

The whitepaper's advice is one sentence: an FDE should not report on a single line into sales or into engineering. Placed under sales, the role becomes pre-sales with a keyboard — the demo closes the deal and nobody is paid to stay until the metric holds. Placed under engineering, it becomes a feature team that receives tickets and never meets the person whose week it is changing. Placed under support, it becomes break-fix.

The position that works sits at the intersection of product, delivery, customer success and a business number — awkward to draw, easy to say. A practical design: the business owner holds the metric, IT holds system security, the FDE holds delivery, and management holds adoption. No single line contains all four, which is the point.

A cheap test: read the FDE's performance review. If every line can be observed from inside the office — tickets closed, features shipped, demos delivered — the role is post-sales whatever the org chart says. If at least one line can only be observed at the customer, in their numbers, the reporting line is doing its job. The question of whether an FDE is pre-sales is usually settled this way, by evidence rather than by title.

The title is the last thing to read

Titles are the least informative field in this market. Sierra has no "forward deployed engineer" listing; the function is spread across agent engineering roles and a Tokyo "forward deployed infrastructure engineer". Harvey runs about 180 legal engineers and deploys in pods of one PM, one or two lawyers and an engineer. Scale renamed the job "Frontier Agents Engineer". The same work appears under four names because the work is defined by the account, not the ladder.

So read the form, not the label: who signs the renewal, who pays for the travel, where the field evidence goes. The 2026 wave makes this checkable: deployment organisations now publish their structures instead of hiding them in headcount.

The three names to fill in before choosing

Write three names next to the form you are considering:

1. The person who owns the customer relationship and signs the renewal.
2. The budget line that pays for travel and idle time before the first invoice.
3. The meeting where field evidence last changed the roadmap or the price list.

If any of the three is blank, the form is wrong, and the cheapest correction is usually the reporting line, not the headcount. Which form makes sense also depends on how mature your customers' data and processes already are; the five levels says where they actually sit, and a form built for level four cannot be sold into a level two account. Without the third name, the symptom appears at renewal: the work is priced as hours because nobody can point to what was reused — the degeneration scoping and pricing is written to prevent.