The forward-deployed engineer (FDE) is a role invented by Palantir. An FDE is an engineer embedded inside the customer who writes production code and feeds the core product roadmap. Palantir invented the FDE role to solve a problem its first customers in the US government (CIA, NSA, US Army intelligence units) could not solve with traditional consultants.
The origin of the role is inseparable from the nature of that early work. Intelligence agencies deal with sensitive, classified, and constantly-changing problems that cannot be specified in advance. These problems can hardly be solved by standard SaaS, which requires generic enough use cases to justify building a solution that can be sold at scale to any enterprise. It’s hard to reuse requirement analysis docs and PRDs of a system built to fight ISIS.
Traditional consulting companies offer two inadequate responses: either consultants à la McKinsey who cannot write production code, or system integrators à la CapGemini that have no proprietary product to build upon. So Palantir embedded its own engineers inside customer facilities for weeks or months, learning by observing, experimenting, and building in real time.
Palantir’s FDEs write custom code with two destinations. One is the customer’s deployment. The other is Palantir’s own core repo. A bespoke solution for one customer is studied, the repeated pattern gets extracted, and it becomes a platform feature every future customer gets by default. The cost of each next deployment gets cheaper over time, thanks to the compounding effect of each closed-won customer.
In the last 12 months, FDE became the fastest-growing role in enterprise AI, with multi-billion-dollar bets from many companies like OpenAI, Anthropic, and AWS. According to the Financial Times, job postings for the position increased more than 800% from the start of 2025 through September 2025.
OpenAI launched the OpenAI Deployment Company in May 2026, a majority-owned subsidiary with more than $4 billion of initial investment, and agreed to acquire Tomoro, bringing ~150 experienced FDEs and Deployment Specialists from day one.
Anthropic calls its FDE function “Applied AI Engineer”, an engineer who “builds production applications with Claude models, delivers technical artifacts like MCP servers, sub-agents, and agent skills”, and “identifies and codifies repeatable deployment patterns and contributes insights back to our Product and Engineering teams”.
Google Cloud had 59 distinct FDE roles open across its careers site spanning the US, London, Paris, and Hong Kong, split across financial services, healthcare, retail, and public sector.
Databricks runs an “AI Forward Deployed Engineering (AI FDE)” team that “delivers professional services engagements to help our customers build and productionize first-of-its-kind AI applications”. Notably, Databricks explicitly labels this “professional services” and places it in the PS org.
AWS entered the race in June 2026, announcing a $1 billion internal investment in a new AWS Forward Deployed Engineering organization to be seeded with “thousands” of FDEs. AWS intends to send small teams of five to six engineers into customer sites for ~45-day engagements, and stated that “FDE engagements are structured around shared business outcomes, rather than billable hours”.
Salesforce publicly committed to building a team of 1,000 FDEs, while Microsoft announced in July 2026 the Microsoft Frontier Company, with a $2.5B investment and 6,000 industry and engineering experts to co-design, co-innovate, deploy and continuously improve AI systems at scale based on measurable business outcomes.
Are we witnessing a legit increase in Palantir-style FDEs, or is it a big rebranding of existing roles?
There is definitely a risk that FDE label is being applied to two fundamentally different jobs that are not forward-deployed engineering:
Fake #1: sales/solutions engineering with a new title. A solutions engineer is a pre-sales role: demos, RFPs, POCs, and technical close support. The SE helps Sales close the deal and moves on. A real FDE writes and deploys production code in both the customer environment and the core product repo.
Fake #2: systems-integrator (Accenture or Capgemini). Systems integrators are in the labor arbitrage business, they bill for time. Revenue is headcount x duration. An SI optimizes billable hours and utilization of their workforce. Nothing the embedded engineer builds is expected to become the vendor’s product.
If you are considering making a move and join this FDE surge, this is a useful framework to assess where a FDE role lands in the “Capgemini to Palantir” spectrum.
Compensation structure
A Palantir-like FDE role is paid like a staff-level product engineer (salary plus meaningful equity, benchmarked against research/product engineering, not against pre-sales). An SI-in-disguise role has a commission-based sales-engineer comp, with bonuses tied to utilization or bookings. Frontier labs deliberately set FDE comp within striking distance of research-engineering bands. A commission plan or utilization bonus is a red flag that the role is really sales or services.
Length of engagement
Real FDEs engage in multiple repeatable deployments over 60–180 day cycles. Capgemini-style engagements are open-ended, multi-year staff-augmentation embeds where the engineer effectively becomes long-term contract labor and success is measured by continued billing.
Deploy vs. POC
Real FDEs ship to production and own it (i.e. hold the pager). On the other side of the spectrum fake FDEs produce endless proofs-of-concept, demos, and decks. OpenAI and Google job postings both emphasize taking prototypes to production-grade agentic workflows.
Pre-sales vs post-sales
Real FDEs are primarily a post-sale motion to build and deploy, with field signal flowing to the roadmap. Fake FDEs are primarily pre-sale closing. AI labs are ambiguous to this respect, with Anthropic saying that FDEs work “closely with our Post-Sales, Product, and Engineering teams”, while OpenAI FDEs “collaborate closely with Sales, Solutions Engineering, Solutions Architects”. It’s fair to expect some level of collaboration with sales, but a role that is only about closing, with no production ownership, is sales engineering.
Who pays for the engineer
FDEs can be free for the customer. This is the model of hyperscalers, where the FDE’s cost is absorbed by cloud/platform-consumption economics to drive product usage and larger renewals. FDEs are basically COGS to sell more usage-based products, when this subsidizing makes economic sense because the downstream value is large enough to justify it. The SI model is billable-hour professional services, where the engineer’s cost is recovered directly through fees. Middle of the road are fixed-fee engagements based on outcomes. If the customer is billed by the hour for the engineer, the vendor’s incentive is to maximize hours. If the vendor absorbs the cost to grow usage-based consumption, the incentive is to productize work that maximizes usage. Fixed-fee engagements are better aligned with customer’s and vendor’s incentives.
If you are evaluating a role in the FDE space, evaluate these five axes on the actual JD, not the title, and ask for concrete evidence on each if you have the chance to talk to them. The Capgemini-in-disguise FDE sits at one extreme (billable hours, indefinite embeds, POCs, pre-sales) and the legit Palantir-style FDE sits at the other (staff-level salary + equity, 60–180 day cycles, production ship, feedback-to-roadmap, outcome-based).
Do you like this post? Of course you do. Share it on Twitter/X, LinkedIn and HackerNews