Is the FDE a bubble or AI's essential job? From Palantir to OpenAI
Nov 4, 2026Original Y Combinator·KDnuggets·LeadDev·원티드 외 ↗
In 3 lines
- The FDE closes the gap between product and customer on-site; born at Palantir, spread by the AI agent boom
- US postings up 729%, Korean companies posting FDE jobs up 13.8x, but many roles are FDE in name or SI in disguise
- If field learnings reach the product, it's a real FDE; if not, it's consulting with a better title
"There's actually a real risk that it's right." That was Bob McGrew, who helped build the model at Palantir, when asked whether the forward deployed engineer is just consulting with a new name. FDE postings in the US grew more than sevenfold in a year, and OpenAI and Anthropic have poured money into deployment arms. In Korea, the number of companies posting FDE jobs grew more than 13x. Here's what the FDE is, why it took off, whether it's a bubble and where it's headed.
A role born from building software for spies
The FDE (forward deployed engineer) came out of Palantir. Bob McGrew, an early Palantir executive and former OpenAI chief research officer, recalled on Y Combinator's podcast that the company had to build software for spies but "I don't know any spies." Every customer needed something slightly different, so Palantir sent engineers on-site to close the gap between the product and the customer.
Echo and Delta: field analysts and fast engineers
McGrew described Palantir's field teams as two roles: "Echo" and "Delta." Echoes are domain experts, former army officers or healthcare workers, who are also "heretics" convinced the current way doesn't work. Deltas are engineers who ship rough-and-ready code on a deadline. The wrong profile for a Delta, he said, is a craftsman perfecting abstractions to last a dozen years.
Gravel roads, then paved highways
McGrew's metaphor: the FDE lays a "gravel road" to the customer, and the product team turns it into a "paved superhighway" that works for the next five or ten customers. FDE work is product discovery done from inside the customer. a16z, launching an FDE fellowship in May, called this field work "R&D" and named the FDE "the critical enabler for agentic platforms."
Why AI agent companies copy it
McGrew's reason: with AI agents "there is no incumbent product." Unlike, say, billing software, nobody knows exactly what agents should do until you're inside the customer. He summed up the model as "doing things that don't scale at scale." The host noted more than 100 YC startups now list FDE jobs, up from basically zero three years ago.
The boom in numbers, and a small real market
Indeed data showed FDE postings up more than 729% year over year in April, with an average base salary of $171,911 (LeadDev). KDnuggets notes OpenAI launched a deployment company in May with over $4 billion, Anthropic's enterprise services venture was reported at $1.5 billion, and Salesforce committed to 1,000 FDEs. But growth estimates range from 350% to 1,165% depending on who counts, and one tally found just 224 open FDE roles across 39 companies at the end of May.
KDnuggets · KDnuggets ↗"Isn't it just consulting with a new name?"
McGrew didn't dismiss the critique: "there's actually a real risk that it's right." His test is the business model. You may lose money at a new customer, but as the product fits better, the on-site team should shrink and you should earn the right to bigger problems until margins turn positive. KDnuggets managing editor Matthew Mayo put it this way: if what FDEs learn at the customer doesn't change what the company builds next, "it's consulting with a better title."
Signs of froth
A former Palantir FDE told LeadDev the title "might have become a little bit frothy," with many firms labeling anyone customer-facing and somewhat technical an FDE. Palantir's global head of commercial, Ted Mabrey, said every company he saw copying the model offered "a half measure." Some postings require up to 50% travel, raising burnout concerns, and Semafor notes heavy bespoke work could become technical debt as general-purpose models improve. The Pragmatic Engineer newsletter spoke with engineers who were disappointed after taking FDE jobs, finding less software engineering than expected and an IT-services mindset.
LeadDev · LeadDev ↗Korea: up 13x in a year, now with a hiring fair
Korea's numbers are steep too. According to Wanted Lab, the number of companies posting FDE jobs on Wanted from January to September rose 13.8x year over year, and FDE positions rose 14.5x. Wanted is holding an "FDE Recruiting Carnival" in Seoul on Nov 18 for 100 developers with two-plus years of experience, with MakinaRocks, Sionic AI and VESSL AI among the hiring companies. Wanted Lab CEO Lee Bok-gi said "companies and developers still understand this role differently."
Digital Insight · DiTODAY ↗Today's postings split three ways
The 13 Korean companies hiring FDEs in AI News Daily's AI Jobs section (as of Nov 4) fall into three groups: product companies like Toss, Banksalad, Gangnam Unni (Healingpaper), IGAWorks and Madup sending engineers to customers; in-house teams at CJ Olive Young, HL Holdings and Samoo Architects embedding AI in their own business units; and project-style outsourcing or AX work. Eleven of the 13 require three or more years of experience, and only Toss and one internship are open to new grads, which cuts against the idea that Korean FDE work is mostly junior.
AI News Daily · AI Jobs ↗"A mini CTO": a Korean FDE on the difference
Jung Ju-young, an FDE at payments company PortOne, described the role as close to "a mini CTO." SI builds to the requirements in a contract and the experience disappears when the project ends; FDEs define the problem with the customer on-site and feed that know-how back into the core product. He called it a structure that "starts N to 1 but ultimately scales back out 1 to N", the same idea as McGrew's gravel road and highway.
PortOne · PortOne 블로그 ↗"Not a job just anyone can do"
People running FDE companies in Korea offer the opposite diagnosis: the role demands engineering skill plus people skills, quick problem detection and the ability to read org structures, so it's "a hot role, but not one everyone can do." Customers never arrive with a clean problem, so FDEs must carve vague ones into shape, which is why founder or very-early-startup experience is prized. Instead of lone supermen, they argue for 2-3 person teams time-boxed to three to six months, and warn that FDEs who only take problems AI could solve will end up handing their work to AI.
Kkachilhan AI · YouTube ↗The Korean skeptics: "Isn't it just rebranded SI?"
The skeptics are loud too. Some in Korean recruiting say many FDE roles at AX (AI transformation) firms look like SI, building to spec with AI, and call the trend "hype." Tech magazine YozmIT notes that sending engineers to build custom systems for clients isn't new, and that work requiring people on-site is hard to scale. YozmIT plans to publish a survey of working FDEs in Korea this month.
JOB담 (Jobdam)·YozmIT · YouTube ↗What's next: in-housing, and "the title may be a fad, the skill set is not"
Views converge on where FDE work ends: an AI company executive told LeadDev he expects enterprises to bring the role in-house, practitioners in Korea advise leaving maintainers behind inside the client, and in-house FDE roles already show up in Korean postings. McGrew sees capabilities racing ahead while adoption lags, and a big opportunity in closing that gap; KDnuggets concludes "the title may be a fad. The skill set is not." If you're weighing an FDE role, look past the title and ask whether what you learn in the field makes it back into a product or into the company.
