Forward Deployed Engineer: Everyone Copies the Title, Nobody Copies the Loop
- 3 days ago
- 8 min read
On 13 May 2026, in San Mateo, on day two of SaaStr AI Annual, one room had quietly changed its name.
For ten years it was the CCO Summit — the gathering of chief customer officers. Renewal rates, customer health scores, quarterly business reviews. This year the program read: FDE (CCO) Summit. FDE first. Customer success, in parentheses.
First talk, the VP of customer solutions at AssemblyAI, ten years of customer success behind him. His title on the big screen: "Your customers don't need a CSM. They need an engineer." Half the room were CSMs. His confession: they renamed the role "technical account manager" — a hiring disaster, two decent candidates in two and a half months. Renamed it "forward deployed engineer" — the right candidates flooded in. "The actual job barely changed. It was the brand."
That set the tone. The FDE — the engineer deployed at the client — has become the most coveted job in AI. a16z: "the hottest job in startups." ICONIQ, across 205 B2B software firms: FDE openings up 12x in a year. Your HR has probably already put those three letters in a job description. But the FDE your company is about to copy — from title to JD to reporting line — is most likely a coconut-shell airport.
The phrase is worth thirty seconds. During the Pacific war, the US military built airstrips on small islands and air-dropped supplies. After the war, they left. To make the cargo planes come back, islanders rebuilt the runways and control towers out of bamboo and coconut, carved wooden headsets, and queued up waiting. Anthropologists call it a cargo cult. The runway is real, the tower is real, the queue is real — but the plane never comes.
Door one: a job mocked for ten years, suddenly worth 4 billion dollars
In 2011, an enterprise-software veteran mocked Palantir: "In the normal world, those tasks are called pre-sales engineering and consulting." The tone held for a decade: "a consulting company masquerading as software." On the eve of its 2020 IPO, some investors valued it at only about 10.5 billion dollars.
Then the income statement spoke. Gross margin: 67% in 2019, about 80% in 2023, 82% in Q3 2025. For comparison, a large IT-services firm (think Accenture): 32%. Q1 2026: revenue 1.63 billion dollars, up 85%, the fastest since 2020; net margin 53%. The SaaS Rule of 40 (growth plus margin): Palantir scored 127 for fiscal 2025. Market cap: from about 21 billion dollars on day one to about 325 billion by June 2026. Every year, short-sellers; their thesis title unchanged for ten years: it's consulting, not software.
The most telling number is sales efficiency. The CEO, on the same earnings call: "Customers buy despite the fact we have 70 salespeople… only seven actually really sell." A company with 7.6 billion dollars in annual revenue, seven people selling. What replaces the thousands of salespeople you'd normally need is the FDE. The entry ticket: OpenAI put over 4 billion dollars into its deployment arm and picked up about 150 FDEs — roughly 27 million dollars of capital behind each head.
Door two: you copy the title, they hide the loop
Everyone copies the FDE. The problem: most never understood what it is.
The origin. The role was invented at Palantir in 2006, inspired not by the military but by the French restaurant: the waiter isn't a plate-carrier, he's part of the kitchen; he knows every dish well enough to tell you no if you order the wrong wine. The idea: put the person who knows the product best in front of the client, with the right to say no.
The mechanism. The version that circulates — "FDE = engineer who codes at the client" — is wrong. Internally, three roles interlock: an engineer writing production code on site; a deployment strategist who handles the organization, politics, workflows; and, at HQ, a product engineer who generalizes the bespoke solution into product. The former Palantir head of engineering, later OpenAI's chief research officer, gives the best image: "The FDE builds a gravel road on site; the product team then turns it into a paved superhighway — for the next 5, 10 clients."
That is the loop. Each project thickens the product by one layer. Before 2016, Palantir had more on-site engineers than product engineers — ten years as a heavy-delivery company before the 82% margin. The counter-intuitive formula: "doing things that don't scale, at scale."
Compare the market's "FDE" in 2026: the FDE title, the pre-sales work, a POC conversion target, code delivered and left with the client, knowledge in the engineer's head, and the next project starting from zero. The role's inventor is blunt: "Slap a veneer on pre-sales and call it FDE — you've completely cargo-culted it away." And the cleanest test, from an investor: "If the FDE is billable to the project, they work for the project, not the product." Are your new FDEs billed to the project?
Field note — an ex-Palantir FDE who moved on: "Same title, double the salary. Took me three months to see it: at Palantir, every gravel road became a stretch of highway; here, my code is archived with the project sign-off, permissions cut, I can't even open it. Same job title — there a forward post of R&D, here a sales giveaway."
Door three: gaming solved this thirty years ago
If "loop" is still abstract, switch industries. Gaming has already paid, in real money, for every solution and every failure. Hold one mapping: game = client project, engine = product asset.
Those who got it right. In 2016 Capcom developed its RE Engine alongside Resident Evil 7, glued to the game's real needs. The engine wasn't archived after launch: the whole series and several other franchises run on it. Result: 38.5% operating margin, record profits nine years running. Capcom is gaming's Palantir: each game thickens the engine.
Two ways to die. One: build the cathedral-engine first, find a game later. Square Enix spent ten years on its Luminous engine for a single title; the studio was dissolved. That's the AI company raising money to build a platform before seeing a client. Two: HQ forces the engine, the front line is held hostage. To save on licenses, EA forced the whole group onto its Frostbite engine; an RPG studio had to use it without even save/load or a third-person camera, stringing together three disasters. A former employee's words: "Frostbite is full of razor blades." In late 2023, EA abandoned the one-engine strategy.
The promised test, then: at the end of every project, ask — did the engine get thicker? Yes: you're Capcom, Palantir. No: you didn't hire an FDE, you hired a billable consultant, whatever the badge says.
The AI-era translation. Engines commoditized: in 2012, 71% of Steam games ran on a homemade engine; by 2024, 13%, with Unity and Unreal taking 79% of new titles. The APIs of GPT and Claude are AI's Unreal and Unity; the commoditization gaming took twelve years to live, AI does in three. But in the same data, the second half: the 10% of games still on a homemade engine capture 41% of revenue. The engine stays the privilege of the leaders — except it's no longer the renderer or the base model. It becomes the layer of private assets that grows from the client's ground: the business's semantic map (ontology), the harness around the model, sector eval sets, workflow templates. The FDE is the mechanism that, in the era of commoditized models, grows that private engine. That's what OpenAI and Anthropic are really buying.
Door four: an on-site engineer builds a game with one player
The global software industry knows this move well: code on site, tweak to the client. Staff augmentation, fixed-fee delivery. Same move, two fates: Palantir gets 82% margin from it; the classic staffing model loses on every contract. Why?
The answer, again through gaming. The deliverable of a staffing project belongs to one client. An on-site engineer builds a game with one player. The match isn't Nintendo (whose game is a copyable asset at near-zero marginal cost); it's the outsourcing studio: making someone else's game, accumulating no engine, keeping no IP. Look at the valuation ladder: Keywords Studios, the world's largest outsourcer, 12,000 people — taken private in 2024 at 2.8 billion dollars. Epic (engine): 22.5 to 31.5 billion. EA (engine plus IP): taken private in 2025 at 55 billion. Outsourcing, engine, IP — each copyable layer adds an order of magnitude. A labor business that accumulates no asset is worth a fraction of an asset business.
One fair caveat: heavy on-site work isn't a sin. Harvey (110 billion valuation, 190 million ARR, 100,000 lawyers in 60 countries) embraces a 2001-looking method: quarterly reviews, heavy presence, big-firm-style change management. Because selling AI to a partner who's practiced for thirty years isn't selling software, it's changing how they work — there, heavy presence is a moat. Its line to engrave: "Logins are not transformation." The difference isn't the move, it's the layer the deliverable lands on: every Harvey engagement feeds the same product; body-shop staffing feeds a one-off. The engine is compiled organizational knowledge — the person leaves, the asset stays. Classic staffing leaves the knowledge in the veteran's head; project over, team scattered, asset at zero. The harshest moment for a ten-year staffing engineer is redoing the CV: no body of work, only seniority.
Three ways out
AI startup founder: before hiring an FDE, build the road home for the code. Concretely: the FDE's review includes a productization contribution, not billable hours; at the end of every cycle, a gravel-road review decides which stretch goes into the product. Trading margin for a moat is valid — but only if the money spent comes back as product asset, not revenue.
Engineer: the FDE may be the highest-leverage CV of the era — in every YC batch, Palantir-bred founders outnumber Google-bred ones. Before you sign, one question: will my code become part of the product, or the funeral offering of a project? The first is a body of work; the second is seniority.
Building enterprise software: for the first time, the amortization formula is shifting — AI drives down the marginal cost of bespoke work; an FDE armed with a model does the work of a whole team. The window: turn staffing into an engine, or keep playing the single-player game.
In 2016, Wall Street mocked Palantir: "a consulting company masquerading as software." In 2026, Sequoia's investment thesis reads: "the next 1T-dollar company will be a software company masquerading as a services firm." Same sentence, order reversed — from insult to thesis. Between the two is not ten years. It's the thickness of the engine.
FAQ
How do you tell a real FDE from a fake? One question: at the end of the project, does the code thicken the product, or stay with the client? If they're billed to the project, they work for the project, not the product.
Should heavy on-site work be avoided? No — if each engagement feeds the same product. It's a moat when the deliverable rises one copyable layer, a trap when it stays a one-off.
What is the "engine" in applied AI? The private-asset layer grown from the client's ground: business ontology, the harness around the model, sector eval sets, workflow templates.
ECTIME AI Lab is the applied-AI research and deployment unit of ECTIME Group. We build, ship and stress-test agentic systems in production. Our focus is verification and runtime governance. We maintain open-source Claude Skills for GEO/SEO and advise European brands on deploying AI that is not just autonomous, but verifiable and authorized.
Comments