What Is a Forward Deployed Engineer?

The distinction that makes the role make sense is Palantir's own: a product engineer builds one capability for many customers, and a forward deployed engineer builds many capabilities for one customer. Everything else about the job follows from that inversion.

800-1000%growth in FDE job postings
in the past year, by source

$1BAWS investment in its own
FDE org, June 2026

22 of 311OpenAI open roles that were
FDE-shaped, June 2025

Two of those numbers come from Andreessen Horowitz. Marc Andrusko reports that postings for forward deployed engineers are "up 800–1000% this year, depending on the source", and Joe Schmidt counted 22 of the 311 open roles on OpenAI's careers page falling into the FDE and solutions category. The firm's own headline calls it the hottest job in startups. The third is AWS's, and it is the clearest signal on this page: in June 2026 AWS announced a "$1 billion investment" in a unit called AWS Forward Deployed Engineering, to "embed thousands of experts with customers". A hyperscaler does not name an organization after a job title that is going away.

Where the role came from

The role was built at Palantir to solve a delivery problem. No org-design theory produced it. Andrew Ng's account is the most specific on why: Palantir "sent engineers to government locations to work on secure, air-gapped networks." You cannot support an air-gapped network from your own office. Internally the job was called Delta.

The exact start date is contested, so here is the disagreement. Ng puts it "about two decades ago." Gergely Orosz and a16z's Marc Andrusko both say the early 2010s. a16z's Tom Hollands is the only source to give a year, and his point is that the job began as a retitling: "In 2011, Palantir did something funny. They took their solutions engineers and integration engineers ... and gave them a new title: forward-deployed engineer." Wikipedia declines to date it at all and says only that Palantir "popularized" the role. Early 2010s is the safe reading, with the practice likely predating the title.

How central it became is the part most explainers leave out. As Gergely Orosz documents in The Pragmatic Engineer's breakdown of the role, up until around 2016 Palantir had more forward deployed engineers than it had conventional software engineers. Palantir's own framing of the job, quoted there, is that an FDE's responsibilities "look similar to those of a startup CTO: you'll work in small teams and own end-to-end execution of high-stakes projects."

Why the title is suddenly everywhere

Enterprise AI recreated Palantir's original delivery problem at industry scale. A frontier model is genuinely useful and genuinely useless out of the box: the value only appears once it is wired into one company's data, workflows and edge cases. That work cannot be done from a vendor's office, which is why the vendors started shipping engineers instead of documentation.

EARLY 2010sPalantirDeltas embedded in air-gapped government networks no vendor could support remotely

→

2025-26The AI labsOpenAI, Anthropic and AWS hire the title to get models into production

→

NowRevenue teamsThe same model pointed at go-to-market, where the system is the pipeline

The role is now established enough to have its own Wikipedia article, which records that Amazon Web Services, OpenAI and Anthropic have all hired people under the title, and that "companies listed significantly more job openings for this role between 2024 and 2025." OpenAI's careers page currently lists forward deployed engineer openings across San Francisco, New York, Seattle, Tokyo, Singapore, Seoul and Sydney, including versions specialized by industry for healthcare and for legal work. That specialization is the tell: when a vendor hires per-vertical FDEs, it has accepted that domain context is part of the product.

Andrew Ng gave the role its clearest recent definition in a letter published in The Batch on 29 May 2026. An AI FDE, he wrote, is "an engineer who is embedded within a client organization to help customize solutions, such as building and tuning agentic workflows." He addressed it because people kept asking him about the career path after "OpenAI and Anthropic started building new teams to place FDEs within client organizations."

What a forward deployed engineer actually does

The work is a full delivery cycle owned by one person or a very small team, run inside somebody else's company. Reading OpenAI's own postings alongside Ng's description of the skill set, the shape is consistent.

Discovery in situSit with the real workflow and find where the product actually has to fit

Technical scopingTurn a vague ambition into a system design against live data

BuildingWrite the integration and the custom logic inside the customer's stack

Evals and rolloutInstrument it, prove it works on real cases, push it to production

Saying noPush back respectfully when the ask is unrealistic, per Ng's skill list

Feeding the roadmapSend the gaps found in the field back into the core product

Two of those deserve emphasis, because they are where the job is won or lost. The first is that the skills are not purely technical. Ng's list is explicit that an FDE needs to "speak with clients to understand their needs, formulate a strategy to prioritize projects, explain complex technology, and respectfully push back if a client asks for something unrealistic." An engineer who cannot hold that conversation will build the wrong thing beautifully.

The second is the return path to the product. An embedded engineer who only ships custom work is a consultancy in a trench coat. What makes the model defensible for the vendor is that the FDE brings the pattern back, and the next customer gets it as a feature. That is also the honest answer to why companies accept the margin hit. Schmidt's argument in Trading Margin for Moat is that it is "shortsighted to be optimizing for 80% gross margin" when the alternative is becoming the system your customer runs on.

The job is not a universally attractive one, and the pages selling it rarely say so. Wikipedia records the objection plainly: "some software engineers consider the role to be undesirable because of the travel requirements and pressure to solve customer problems within relatively short amounts of time." That matters if you are hiring. The people who are good at this actively like being in the customer's building, and the ones who tolerate it as a career step leave in a year.

Forward deployed engineer vs solutions engineer

A solutions engineer is measured on winning the deal, and a forward deployed engineer is measured on what happens after it closes. That is the short answer. The longer answer is that the boundary is genuinely blurry, and anyone who draws it too crisply is selling you something.

Both Wikipedia and a16z are candid that these titles overlap and are sometimes the same job renamed. The Wikipedia article notes the duties overlap with solutions architects, sales engineers, customer engineers and IT consultants. Schmidt is blunter still, describing the work as one often filled by a professional services employee, "sometimes rebranded as a forward deployed engineer or an implementation/solutions specialist". The most useful published distinction comes from inside OpenAI, reported by The Pragmatic Engineer, where FDEs "are more hands-on and typically work with more ambiguity than SAs traditionally do."

So ignore the title on the badge and apply two tests. Is the person allowed to write code that survives in production? Does what they learn change the product? If both are yes, it is a forward deployed engineering role whatever it is called.

Forward deployed engineer vs AI engineer

An AI engineer builds their own company's product, and a forward deployed engineer builds inside somebody else's company. The skills overlap heavily. The org chart does not.

Ng is worth reading on this because he argues against the hype on his own topic. He expects "the number of AI Engineer jobs will be far larger" than the number of FDE jobs, for the straightforward reason that a company might accept a few embedded outsiders while wanting far more of its own employees on its own projects. His organizations hire both and hire many more AI engineers. Treat FDE as a specific delivery role inside a vendor's go-to-market, not as the general shape of AI work.

Forward deployed software engineer and forward deployed AI engineer

These are variants of the same job, and the difference is mostly which company is hiring. A forward deployed software engineer (FDSE) is Palantir's original title, still the most common phrasing where the deployed artifact is conventional software. A forward deployed AI engineer is the 2025 and 2026 version, where the deployed artifact is a model, an agentic workflow or an eval suite. The acronym FDE covers all of them, and job boards use the terms loosely enough that the posting tells you more than the title does.

How much do forward deployed engineers get paid?

The median advertised salary for a forward deployed engineer in the United States is $188,000, and the bands the largest employers publish themselves run from $135,000 to $405,000 before any equity. The median is Lightcast data, reported by Fortune on 3 September 2026, in a piece that puts a traditional software engineering role at about $145,000 on the same data. Doing the work inside somebody else's building carries a premium of roughly 30%.

$188,000median advertised US salary
Lightcast, September 2026

$205,000median self-reported total
compensation, levels.fyi

$135k to $405kfull span of bands posted by
Palantir, OpenAI and Anthropic

Self-reported totals run higher than advertised salary, because they include stock and bonus. The levels.fyi page for the title, read on 7 September 2026, puts median total compensation at $205,000, the 25th percentile at $172,000, the 75th at $277,000 and the 90th at $350,000.

Posted bands are better evidence than either aggregate, because the employer wrote the number and has to honor it. This is what the three companies that define the role were advertising on 7 September 2026.

Bands posted by Palantir, OpenAI and Anthropic on their own job boards, all read 7 September 2026

Read that table with one caution and one conclusion. The caution is that the bands are not measuring the same thing. Palantir publishes a salary estimate and says in the posting that it "excludes the value of any potential sign-on bonus" and the potential future value of long-term incentives. OpenAI attaches "Offers Equity" to every US band without pricing it. Anthropic posts a flat annual salary. At a private lab the unpriced equity is the part that decides the offer, and no posting tells you what it is worth, so treat any total-compensation number above these bands as an estimate rather than a published figure.

The conclusion is for anyone budgeting a hire. Anthropic's floor for an individual contributor, $280,000, sits above Palantir's ceiling for the role Palantir invented, $200,000. The title has repriced since the AI labs started hiring it, and a band anchored to Palantir's scale will not clear the market.

None of this prices the go-to-market version. There is no published salary data for a forward deployed GTM engineer, because almost nobody is hiring under that exact title yet, and any number you are quoted for it today is an estimate rather than a market rate.

How to become a forward deployed engineer

All three of the biggest employers publish their requirements, and the gap between them is the most useful thing in the postings. Palantir asks for "1+ years of relevant, post-college work experience" and a strong coder in Python, Java, C++ or TypeScript. OpenAI asks for five years and Anthropic for four, both of which have to already include customer-facing work. The company that invented the role has the lowest bar for entering it.

Years asked for1+ at Palantir, 4+ at Anthropic, 5+ at OpenAI

Code, in writingProduction-grade and full-stack. Python named by all three

LLMs in productionPrompt engineering, agents, evals and deployment at scale, at the two labs

Customer-facing alreadyRequired in the prior experience at OpenAI and Anthropic, not at Palantir

TravelUp to 25% at Palantir, estimated 25% at Anthropic, up to 50% at OpenAI

In the officeThree days a week at OpenAI, hybrid by location at Anthropic

The requirement lists are worth reading rather than summarizing, because of what they weight. OpenAI's San Francisco FDE posting gives eight bullets under "you might thrive in this role if you". Two are about writing software. Four are about temperament, including "model calm and judgment when the stakes are high" and "spot risks early and adjust without slowing down". Anthropic's forward deployed engineer posting runs ten bullets, two of them technical, and asks for "strong communication skills to conduct discovery with customers and to convey technical concepts to diverse stakeholders while maintaining a low ego". The technical bar is a floor you have to clear. The rest of the list is the job.

Anthropic adds one line that tells you how thin the supply is: "Former technical founders are also encouraged to apply." A company that would rather hire an ex-founder than wait for a matching CV is a company that cannot find the exact background, which is the same signal the salary bands are sending.

If you are starting out, the route into the title runs through the original. Palantir's board carries a new-grad forward deployed software engineer role at $135,000 to $145,000, an internship version of the same job, and a "Year at Palantir" internship track, and its main FDSE posting asks for one year of experience rather than five. Counted on 7 September 2026, 77 of the 310 open roles on Palantir's job board carry "forward deployed" in the title, against 17 of 781 at OpenAI and 6 at Anthropic. Palantir is where the role is taught at volume, and it is the only one of the three posting an entry-level band for it.

If the deployment target is the revenue organization rather than the product, the pay data does not exist yet but the shape of the portfolio does. What transfers is having built one company's system end to end: a signal engine, an enrichment waterfall, routing that encodes a real process, then a handover somebody else can run. I build these systems from inside revenue teams, and what settles it is a system running in production somewhere, rather than a stack of tool certifications. The role that names those skills is the GTM engineer.

Forward deployed engineering applied to go-to-market

Almost nobody has written that down, and the reason the model carries over has little to do with AI. It carries over because domain context cannot be shipped remotely. The same condition holds anywhere the software only pays off once somebody understands the business, which describes go-to-market systems exactly.

Think about why GTM tooling underdelivers so reliably. The tools are good. Clay, Unify, Common Room, Cargo and Octave all work as advertised. What fails is the assembly: the signal logic that only makes sense against one company's ICP, the enrichment waterfall tuned to one data reality, the routing rules that encode one team's actual process, the CRM debt nobody outside the building can see. Buying another tool does not close that, because it is a deployment gap rather than a product gap. It is the same gap Palantir closed by putting engineers in the room.

The signals that this is where the model goes next are already in the record, just uncommented. PostHog's own explainer notes in passing that FDE roles now "range from product development to sales to RevOps" and then does not follow the thread. Andrusko's description of the commercial mechanic, putting engineers on planes to land seven-figure deals, is a revenue motion described in engineering language. And Clay, which coined the title GTM engineer in 2023, defines the role as one that builds automated revenue systems with AI, data enrichment and workflow automation, at "the intersection of commercial thinking and technical building." Point that person at one company for a fixed period with a mandate to ship, and you have described a forward deployed GTM engineer without using the phrase. What a GTM engineer is covers that underlying role on its own terms.

My own reason for caring is that I build these systems from inside the team. A growth engineer hired full-time eventually learns your business. An embedded engineer starts by learning it, on a clock, with the systems as the deliverable. The org chart that receives that person is the part most companies get wrong first.

If you are deciding whether you need one

You need a forward deployed GTM engineer when your tools already work, nothing is wired together, and the wiring has to ship this quarter. The demand data then answers the supply question before you ask it. Postings are up 800 to 1000% in a year on Andrusko's count, AWS committed a billion dollars to an FDE org, and OpenAI is hiring the title seventeen roles at a time across eight cities. Budget is rarely what blocks these hires. The scarce input is people who have actually done the job, and most of them are already deployed somewhere.

So the practical decision is usually a timing question: can you get embedded capacity onto the problem this quarter, or do you start a search you will still be running in six months? The questions worth answering first:

• Is the blocker a tool or an assembly? If your stack is fine and nothing is wired together, buying another tool will not help.
• Can the person write code that stays? An advisor who leaves a deck has not deployed anything.
• Who owns it after the engagement? The handover is the deliverable, not an afterthought.
• Is the metric a system or a slide? Production adoption for a product FDE, pipeline for a GTM one. Pick before you start.

If you have already worked through those and what you need now is the capacity, the embedded GTM engineer profiles show who you would actually get and how the matchmaking works.

Common questions