The title barely existed two years ago. Search demand for it grew 514 percent year on year to 22,200 US searches a month, peaking at 40,500 in August 2026 (DataForSEO, September 2026), and job listings for the role grew roughly 800 percent between January and September 2025 (Economic Times). Andreessen Horowitz has called it the hottest job in startups. This guide covers what an FDE does day to day, where the role came from, how it differs from the roles people confuse it with, what the work pays, and how to get into it.
What does a forward deployed engineer do, day to day?
The work splits into four things, and an FDE does all of them in the same week.
Learning the customer's business properly
Before any code, the job is understanding how the company actually operates: where the data lives, which system is the real source of truth, which process the AI is supposed to change, and who will be affected when it does. Palantir's benchmark, reported by The Pragmatic Engineer, is that FDEs spend roughly 25 percent of their time embedded on the customer's site, and Illinois Institute of Technology puts the range at 25 to 50 percent. That is the proportion of the job that is not engineering.
Building against reality, not a sandbox
An FDE writes code against live systems with real gaps: duplicate records, legacy schemas, permissions that nobody has documented, an API that returns something different on Fridays. This is the part that separates the role from product engineering. There is no clean environment and no option to declare an input out of scope.
Owning the last mile into production
Getting the thing live is the deliverable, not handing over a working repository. That means security review, monitoring, a path for the errors, and a named person on the customer's side who will own it afterwards. An FDE who leaves behind a system nobody can operate has not finished.
Feeding what they learn back to the product
The other half of the role, and the reason vendors fund it: an FDE sees what the product cannot do before any survey would catch it, and that feedback shapes the roadmap. This is why frontier AI companies run forward-deployed teams rather than outsourcing the work.
What the job is not: a support role, a presales role, or a modeling role. An FDE usually does not train models. They take a model that already exists and make it survive contact with a business. That makes the FDE the build-side role in the wider mix of applied AI talent a company needs to move AI from pilot to production.
The origin of the forward deployed engineer
The role was created at Palantir, where these engineers are known internally as "Deltas" and were sent into customer environments to build the working solution on site rather than hand over software and hope it fit. Palantir's own engineering blog explains that the name is a throwback to the company's early days, when each business development team was named after a letter of the NATO alphabet, and that Deltas sit in business development with a mandate to achieve technical outcomes for customers (Palantir, Dev versus Delta).
For years the model stayed a Palantir specialty, because it is expensive: it only pays when each customer is large enough, and the product complex enough, to justify an engineer on site. The model spread as AI moved from demos to production and the last-mile problem, getting a capable model to actually work inside a specific business, became the real bottleneck. What began inside one company is now a named function, forward deployed engineering, at many of the most advanced AI companies.
Why forward deployed engineering took off in the AI era
The shift has a single cause: models became a commodity and deployment did not.
Any company can call a frontier model through an API today. What almost no company can do is get that model to change a business process that already has twenty years of systems wrapped around it. The result is the statistic that defines the current market: an MIT study widely cited through 2026 found that 95 percent of enterprise AI pilots produced no measurable return. Those pilots were not running on bad models. They were running on the same models that work elsewhere, and failing on integration, data quality and ownership.
That is a people problem, and the market priced it quickly. Job listings for forward deployed engineers grew roughly 800 percent between January and September 2025 (Economic Times). The scale of the commitments is the clearest signal: AWS committed around 1 billion dollars to embedded engineering in June 2026 (CNBC), and Microsoft launched a roughly 2.5 billion dollar business built on about 6,000 embedded experts in July 2026 (CIO Dive, The New Stack). OpenAI set out to build a forward-deployed team of around 50 engineers by the end of 2025. These are not pilot programs. They are the vendors concluding that a model they cannot deploy is a model they cannot sell.
For a buyer, the same logic runs in reverse. If 95 percent of pilots fail on the last mile, the last mile is where the money should go, and the role that owns it is this one.
Forward deployed engineer vs solutions engineer, sales engineer, consultant and AI engineer
Four roles sit near this one and get used interchangeably. The distinction that matters is not seniority or skill, it is what each role is accountable for when the project ends. Here is how the five compare on core focus, depth of engineering, customer exposure and accountability.
- Forward deployed engineer. Core focus: getting a system live inside one customer's business. Depth of engineering: high. Customer exposure: high. Accountable for: the outcome in production. The deepest, longest engagement of the five, and the only one that also feeds the product roadmap.
- Solutions engineer. Core focus: making the product fit a customer's requirements, mostly around the sale. Depth of engineering: medium. Customer exposure: medium to high. Accountable for: the fit, usually for a shorter engagement than an FDE, and anchored to one product rather than building ad hoc.
- Sales engineer. Core focus: the technical side of winning the deal. Depth of engineering: medium. Customer exposure: high, but pre-sale. Accountable for: the deal, not post-sale delivery.
- Consultant. Core focus: the recommendation. Depth of engineering: varies, often advisory only. Customer exposure: high. Accountable for: the advice, not the implementation, unless implementation is bought separately.
- AI engineer. Core focus: building AI capability into a product. Depth of engineering: high. Customer exposure: low. Accountable for: the system, usually the vendor's own rather than a customer's.
The shorthand: a sales engineer gets you to buy it, a solutions engineer gets it configured, a consultant tells you what to do, an AI engineer builds the capability, and a forward deployed engineer is the one still there when it has to work on a Tuesday morning in your environment.
The skills that define a forward deployed engineer
The profile is unusual because it is a combination rather than a specialty, and the combination is what makes it scarce.
Technical skills
Strong engineering fundamentals, usually full-stack breadth rather than depth in one layer, because the work crosses the whole path from data to interface. Real data-integration skill: reading an unfamiliar schema, handling the records that break the assumption, building a pipeline that survives the source changing. Enough familiarity with models and evaluation to know what the model can and cannot be trusted with, without needing to train one. And the operational side: deployment, permissions, monitoring, and the discipline to leave something another team can run.
Non-technical skills
Customer communication, which here means sitting with a non-technical team and leaving with a correct understanding of their process, not a list of feature requests. Comfort with ambiguity, because the brief is usually a direction rather than a specification. Business judgment, which is the ability to tell which 20 percent of the request is where the value is. And the willingness to own an outcome in someone else's organization, which is as much a temperament as a skill.
Why that combination is rare
Each of these is common on its own. Engineers with full-stack breadth are not rare. Engineers who are comfortable in front of a customer are less common but not rare. Engineers with both, plus the judgment to scope, plus the appetite to own a result they do not fully control, are genuinely scarce, and that scarcity is what the salaries reflect.
How much do forward deployed engineers earn?
Reported forward deployed engineer salary ranges for US base pay cluster between 110,000 and 180,000 dollars. Glassdoor data cited by Index.dev in December 2025 gives 120,000 to 180,000 dollars base; FDE Academy gives 110,000 to 180,000 dollars. Total compensation at frontier AI companies runs considerably higher once equity is included, and reported packages there reach well into six figures.
For context on where that sits, the US Bureau of Labor Statistics 2023 figures cited by Illinois Institute of Technology put research scientists at around 145,000 dollars and data scientists at around 108,000 dollars. An FDE at the top of the reported range is being paid above both, which tells you what the market thinks the deployment constraint is worth.
Three things move a package inside that range, and they are not the usual ones:
- Whether the employer is the vendor or the customer. FDEs employed by frontier AI companies are compensated against engineering bands at those companies, which is a different market from an in-house integration role at the same title.
- Equity, not base. At the high end the base is the smaller half of the story. Reported packages at frontier companies are equity-weighted, which is why base-salary comparisons across employers mislead.
- Travel and on-site expectation. The role carries real embedded time, with Palantir's benchmark at roughly 25 percent of time on site (The Pragmatic Engineer) and Illinois Tech reporting 25 to 50 percent. Roles at the higher end of that are usually compensated for it.
Why the figures move so fast: demand for the title grew 514 percent year on year, and searches specifically for forward deployed engineer salary grew 806 percent (DataForSEO, September 2026). Any salary figure for this role, including the ones above, should be read as a snapshot of a market that is still repricing.
How to become a forward deployed engineer
There is no standard path into the role yet, which is good news if you are trying to enter it. The people hired as FDEs today came from product engineering, from solutions engineering, from consulting with real build skills, and from founding something small that failed. What they have in common is evidence, not credentials.
- Do the job before you have the title. In your current role, find the project where something technical has to land inside a team that is not yours, and own it end to end, including the unglamorous parts: the data cleanup, the permissions, the handover. That is the experience being screened for, and it is available inside most engineering jobs.
- Build breadth, deliberately. If you are deep in one layer, the gap is usually the data path or the deployment path. Pick the one you avoid and take a project in it.
- Learn to run a conversation with a non-technical owner. Not presentation skills. The specific skill of sitting with someone, understanding their process well enough to describe it back correctly, and identifying which part of it actually matters. Most engineers who do not get hired for this role fail here, not on the technical screen.
- Get something into production and be able to narrate it. One system, live, with a business number attached, that you can explain from the first conversation to the handover. One of those beats a long list of projects that shipped to a demo.
On courses and credentials: several training programs and university tracks now market directly to this role. They can help with the technical gaps, and nothing about them substitutes for a production system you can talk about. Hiring managers for this role screen on evidence of ownership.
Which companies use forward deployed engineers?
The model started in enterprise software and has spread in three waves, with a fourth now forming.
The frontier AI labs. OpenAI and Anthropic both run forward-deployed teams, because a model that an enterprise customer cannot deploy is a model that customer will not renew. OpenAI set out to build a team of about 50 forward-deployed engineers by the end of 2025.
The hyperscalers and incumbents, at a scale that changes the market. AWS committed around 1 billion dollars to embedded engineering in June 2026. Microsoft launched Microsoft Frontier, a roughly 2.5 billion dollar business built on about 6,000 embedded experts, in July 2026.
Everyone selling complex software to complex businesses. Palantir, where the model originated, plus Splunk, Salesforce, Ramp and a long tail of AI-native vendors. The pattern is consistent: the product is powerful, the customer's environment is messy, and the gap between those two facts is where the revenue is lost.
And, increasingly, the buyers themselves. The newer development is companies hiring the profile for their own side of the table: not to deploy a product to customers, but to get bought AI into their own production systems. That is the version of the role most mid-market companies actually need, and it is the one the market has the least supply of.
Sources: TechCrunch, CNBC, CIO Dive, Illinois Institute of Technology, TSIA (2026).
When should you hire a forward deployed engineer?
Hire this profile when a pilot works in a demo and keeps failing to ship, when the blocker is your systems and your data rather than the model, or when you need someone who will own an outcome inside your team rather than hand over a repository.
Do not hire one when the use case has not been chosen yet: at that point senior judgment is worth more than build capacity, and a fractional AI leader is the better first hire. Do not hire one when the work is a fixed-scope build you would rather not manage: that is an agency engagement. And do not hire one as a substitute for a data owner, because the data problem will outlast them.
If you want the hiring side in detail, see how to hire a forward deployed engineer.
Is the forward deployed engineer role here to stay?
The honest answer is that the work is permanent and the title might not be.
The case for permanence is the gap itself. As long as general-purpose models have to be adapted to specific businesses, someone has to do the adapting, and that someone has to be an engineer who is willing to sit on the customer's side. Nothing on the horizon removes that. Better models make the adaptation easier; they do not remove the need to know which process to change, which system is the source of truth, and who will own the result.
The case against the title is that it is a 2025 and 2026 label for something software companies have done under other names for twenty years: field engineer, implementation engineer, solutions architect. Some commentary, including The New Stack, expects the naming to settle into something more conventional once the hiring surge cools.
What is unlikely to change soon is the supply problem. Executive search firm Christian & Timbers estimates that only about 2,000 engineers in the US have the full mix the role demands at its most valuable: sector knowledge, the presence to work with senior client teams, and hands-on applied AI experience. It is one firm's reading of its own market rather than a census, and it uses a narrow definition. But it points the same way as everything else: demand for the title grew 514 percent year on year, while the skill combination behind it, engineering breadth plus customer judgment plus appetite for ownership, is the slowest kind of supply to build.
Put a forward deployed engineer in your team
If the pattern in this guide describes where your AI work is stuck, the fix is usually one embedded engineer rather than a bigger pilot. Pangea.ai matches vetted forward deployed engineers to a single brief, from a network vetted to the top 7 percent across more than 500 data points, with a shortlist back in 72 hours. 4.9 rating across 1.5K ratings, 1,500+ teams matched.

.webp)
.avif)
