Read summarized version with
A forward deployed engineer, or FDE, is a software engineer who works inside a customer’s team for a defined period, builds in the customer’s own environment, and is judged on what is running when they leave. The title comes from military vocabulary, deployed forward to the point of action, and was in use at Palantir by 2009.
In 2026 it became the most talked-about job in enterprise AI. OpenAI, Anthropic, Microsoft, AWS and Google each announced forward deployed teams in the space of one summer, with more than nine billion dollars behind them. This guide explains what the role is, where it came from, how it differs from a consultant, a solutions engineer and a contractor, and how to tell whether your company should hire one or buy one.
What FDE stands for, and the other FDE
FDE stands for forward deployed engineer, and sometimes forward deployed software engineer, which Palantir shortened to FDSE. Search for the acronym alone and you will also meet flat dark earth, a colour used on firearms and outdoor gear. That FDE has nothing to do with this one, and the two now share a results page, which is why this article spells the role out in full.
Forward deployed engineering, with the ing, is the model rather than the person. It is the practice of putting engineers inside a customer’s organisation to deliver an outcome there. A company runs forward deployed engineering. An individual is a forward deployed engineer.
Where the term comes from
Palantir coined the title in the late 2000s. Its early customers were intelligence and defence agencies that could not explain what they needed from a data platform, and often were not permitted to. Instead of asking, Palantir sent engineers to sit with the analysts, watch the work, and build what the work needed, on site, inside the classified environment.
Inside Palantir the role was called a Delta, against the product engineers called Devs. The company’s own description is the cleanest definition anyone has written. A product engineer’s focus is one capability, many customers. A forward deployed engineer’s focus is one customer, many capabilities. Until 2016 Palantir employed more Deltas than product engineers.
Nabeel Qureshi, who spent years as one, described the job as solving the customer’s problem without worrying about overfitting to it, while the product team’s job was to take whatever the Deltas built and generalise it. That loop, one customer’s messy problem turned into a feature everyone gets, produced most of Palantir’s platform. It is also why the model was controversial for a decade. Sceptics called Palantir a consulting company pretending to be a software company. Palantir’s answer, on its second-quarter 2026 earnings call, was an 86 percent gross margin and 157 percent net revenue retention, and its chief technology officer’s line that only Palantir has FDEs and everyone else has sparkling sales engineers.
What a forward deployed engineer actually does
Strip the vocabulary away and the job has four parts.
- Embeds. The engineer joins the customer’s team for a defined period, attends its standups, works in its channels and commits to its repositories. Palantir’s engineers were on site three or four days a week. AWS’s 2026 model uses pods of five or six engineers. Anthropic’s role description quotes about a quarter of the time travelling.
- Builds in the customer’s environment. Not a demo, not a reference architecture, not a proof of concept on the vendor’s account. The work lands in the customer’s infrastructure, against the customer’s data, under the customer’s controls.
- Owns the outcome. A forward deployed engineer is measured on whether the thing works in production, not on hours billed or a document delivered. OpenAI’s role description has the engineer owning discovery, scoping, design, build and rollout end to end.
- Feeds what they learn back. At a product company that means new features. At a services company it means the next engagement starts further along than the last one. Either way, the role only makes sense if the customer’s problem stops being bespoke over time.
A typical engagement, in the version AWS describes, is a 45-day sprint against a business outcome agreed up front, with the customer’s engineers in the room so that they can run the result afterwards. Anthropic asks its forward deployed engineers to leave behind technical artefacts, MCP servers, sub-agents and agent skills, and to codify repeatable deployment patterns. The common thread is that the engineer’s exit is planned from the first day.
Forward deployed engineer, consultant, solutions engineer or staff augmentation
The role overlaps with four older ones, and the differences are what a buyer is paying for.
| Role | Who they work for | Where they work | What you get | Who owns the outcome |
|---|---|---|---|---|
| Forward deployed engineer | The vendor or provider | Inside your team and environment | Working software in production | The engineer, until handover |
| Consultant | A consultancy | Workshops, documents, your meetings | Analysis and recommendations | You, when you implement them |
| Solutions engineer | The vendor, before the sale | Demos and architecture reviews | A design and a proposal | Nobody yet. The sale has not closed |
| Staff augmentation | A staffing provider | Inside your team, on your backlog | Capacity, working your way | You |
| Managed services | The provider | Your platform, their on-call | A system run for you | The provider, continuously |
Two of these are often confused with the FDE. A solutions engineer advises before the contract is signed. A forward deployed engineer builds after it. Staff augmentation puts engineers inside your team too, but they work your backlog your way and you own the result, whereas a forward deployed engineer brings a method and is accountable for the outcome. Netguru’s description of the difference is precise. Forward deployment inverts the accountability structure.
What forward deployed engineering is as a model
Forward deployed engineering is the operating model built around the role. A provider sells an outcome, staffs it with engineers who embed in the customer’s organisation for a defined period, and structures the commercial terms around the result rather than the time. Three features separate a real one from a contractor with a new title.
- A defined period with a planned exit. Francessca Vasquez, who runs AWS’s forward deployed unit, put it plainly. It does not behove customers to have embedded engineers operating on their behalf for years. The engagement ends with the customer self-sufficient.
- Outcome pricing. AWS structures deployments around shared goals and business results, not billable hours. Palantir’s chief revenue officer said the company does not get paid for clicks, tokens or chats. Databricks runs its forward deployed engagements on shared objectives with milestone-based fixed fees.
- Reuse. Every engagement should make the next one cheaper. Decagon, an AI customer-service company, restructured its forward deployment in 2026 after every customer generated an endless stream of requests that all routed through engineering, and set out to cut the custom engineering per deployment by 80 percent. The investors who like the model most are the ones warning about this. a16z’s Marc Andrusko wrote that if you only copy the embedded-engineer part, you end up with thousands of bespoke deployments that are impossible to maintain or upgrade.
Why 2026 is the year everyone hired one
The role had been a Palantir curiosity for fifteen years. Then enterprise AI made its problem everyone’s problem.
Models got good faster than companies could deploy them. S&P Global found in 2025 that 42 percent of companies had abandoned most of their AI initiatives. MIT’s NANDA group reported that 95 percent of generative AI pilots showed no measurable return. The bottleneck was never the model. It was the integration, legacy systems, data nobody had cleaned, compliance rules, and workflows that were never designed for software that acts on its own. That is the work a forward deployed engineer does.
So the labs built forward deployed organisations, and then spun them into companies. In May 2026 OpenAI launched the OpenAI Deployment Company with more than four billion dollars of outside capital, and Anthropic set up a services venture with Blackstone, Hellman & Friedman and Goldman Sachs valued at around 1.5 billion. In June AWS announced a one-billion-dollar Forward Deployed Engineering unit and, the same day, a partner-led motion to deliver the same method through consulting firms. In July Microsoft followed with the Microsoft Frontier Company and around six thousand embedded experts. Google Cloud put 750 million dollars into a partner fund.
Demand for the people followed. Indeed’s index of forward deployed engineer postings, reported by Business Insider, rose 729 percent in the year to April 2026. Plank’s census counted 1,333 open roles across 565 companies in September. Christian & Timbers, an executive search firm, estimated that only about two thousand of the seventeen thousand people in the United States who hold the title can reliably deliver enterprise AI returns, and that demand would rise a further 2,100 percent by the end of the year. Posted salaries run from a median base around 195 thousand dollars to 350 to 550 thousand at the labs.
Hire one or buy one
If you searched for this term because you are hiring, the market above is the answer to why it is hard. Hiring a forward deployed engineer takes ninety to a hundred and twenty days, competes with the labs on pay, and gets you one person. Most companies that need forward deployed engineering do not need a headcount. They need one platform or one workflow to reach production in a defined time, with their own team able to run it afterwards.
That is what a provider sells. The buying question is therefore not whether you can afford the title but what a provider’s engagement actually contains. Who embeds, for how long, against what outcome, who owns the code, and what is left behind when they leave. The next section is the checklist.
What to demand from a provider
- A defined scope and length. One platform or one workflow, one objective, a fixed number of weeks. If the provider cannot say what will be different at the end, it is staff augmentation.
- Engineers who make changes, not slides. Ask who commits code, into which repository, and from which day. The honest answer starts in the first week.
- Code in your repositories from the first commit. Plank’s buyer research found no two production deployments used the same tool stack, and that clients who own the code and the documented patterns carry the rest of the lifecycle themselves. Ownership on handover is the whole point.
- A measured baseline. The engagement should take a measurement in week one and read it again at the end. Cost, latency, recovery time, alert volume, deployment frequency, whichever the objective names.
- A planned exit. Your engineers in the room for every decision, and a backlog you own at the end. A provider that needs to stay is not forward deployed engineering. It is a managed service, which can be the right answer, but it is a different one.
- Outcome pricing. A fixed price against the objective, or milestones. Hourly billing puts the incentive in the wrong place.
How Dedicatted does it
Dedicatted runs forward deployed engineering as a four-week Mission on cloud platforms. Two senior platform engineers, a Lead Platform Engineer and a Senior Platform Engineer, embed with a team of four to six of the customer’s engineers and one platform owner. The Mission names one bounded platform, one objective and a prioritised backlog of three to five priority items. Engineering begins within days, not after an assessment. Week four validates, measures against the week-one baseline and hands over. The price is fixed and agreed before the Mission starts.
There are eight Missions. Five on the platform itself, from performance and cost to security hardening and observability. Two on AI, bringing AIOps into operations or taking an AI pilot to production on the model platform the customer already uses. One the customer defines. Missions run on the platform the customer’s team owns, on AWS, Microsoft Azure or Google Cloud, on Kubernetes, and on data, AI and on-premises platforms. Dedicatted is an AWS Premier Tier Services Partner, and the method does not change with the platform. The full offer, the comparison against consulting and managed services, and the questions customers ask are on the forward deployed engineering page.
FAQ
Is a forward deployed engineer the same as a consultant?
No. A consultant produces analysis and recommendations that your team implements. A forward deployed engineer implements, inside your environment, and is accountable for what is running at the end.
Is FDE a real engineering job or a rebrand?
Both views exist. Wikipedia’s entry records engineers who see it as title arbitrage. What separates a real one from a rebrand is a defined period, outcome pricing, code in the customer’s repositories and a planned exit. Without those it is contracting with a new name.
What does a forward deployed engineer earn?
Posted base salaries in 2026 cluster around 195 thousand dollars in the United States, with a quarter of postings above 300 thousand. OpenAI and Anthropic benchmark the role at 350 to 550 thousand in total compensation.
Which companies have forward deployed engineers?
Palantir invented the role. In 2026 OpenAI, Anthropic, AWS, Microsoft and Google Cloud all announced forward deployed teams, along with Databricks, Snowflake, Scale AI, Cohere and Mistral. Consulting firms including Accenture, Deloitte and EPAM run partner-badged versions.
How long does a forward deployed engagement last?
It varies with the model. AWS runs 45-day sprints. Accenture’s pods work in 90-day cycles. Dedicatted’s Mission is four weeks on one platform. Longer than a quarter usually means the exit was never planned.
Do forward deployed engineers work on site?
Often, and increasingly remotely inside the customer’s working day. Palantir’s were on site three or four days a week. Most 2026 roles quote 20 to 50 percent travel. Dedicatted embeds on site when a Mission needs it and remotely inside the customer’s working day.
Want two of them inside your team for four weeks? See the Mission.
Figures and quotes are as published in 2025 and 2026 by Palantir, AWS, OpenAI, Anthropic, TechCrunch, Business Insider, The Pragmatic Engineer, Wikipedia and Plank’s State of FDE report.
