The forward deployed engineering model: when it pays off and how to structure it

The forward deployed engineering model pays off when contracts are large, each customer’s environment is different, and your product organization is disciplined enough to turn what engineers build at one customer into product for the next. It erodes margin when contracts are small, the product is meant to work out of the box, or field work never leaves the account it was built for. If you adopt it, place the team where its output reaches the roadmap, grow it from a few senior generalists to a led function, and judge it on time to production, adoption and expansion, not hours.

Company facts come from filings, job postings and published reporting, and arguments are attributed to their authors. Where nobody has published reliable numbers, we say so.

What trade does the forward deployed model make?

The model spends engineering time on each customer. In return it aims for deployments that reach production and first-hand product learning, at the price of services-like economics until the product catches up.

Palantir, which pioneered the role^9,109The Pragmatic Engineer (Gergely Orosz), “What are Forward Deployed Engineers, and why are they so in demand?” 12 August 2025.10First Round Review, “So You Want to Hire a Forward Deployed Engineer: How to Know If You Need One and How to Get the Role Right,” 24 February 2026., published the clearest numbers in its 2020 registration statement: contribution margin (revenue less cost of revenue and sales and marketing expenses, excluding stock-based compensation, as a share of revenue) for customers in three phases. In 2019, Acquire-phase customers (those from which it recognized less than $100,000 in revenue that year, typically during pilots) produced $0.6 million in revenue against a contribution loss of $65.4 million. Expand-phase customers produced $176.3 million at a contribution margin of negative 43%. Scale-phase customers produced $565.7 million at a positive 55%²2U.S. Securities and Exchange Commission (EDGAR), “Palantir Technologies Inc. Form S-1 Registration Statement,” 25 August 2020.. The filing credits its improving results principally to a significant decrease in the time and number of software engineers needed to install, deploy and manage its platforms, and says the time for a customer to start working with its data fell more than five-fold to an average of 14 days in the second quarter of 2020².

Those figures describe one company years ago, not a benchmark. They show the shape of the trade: heavy losses on pilot and early accounts, recovered only if the cost of serving each account falls.

The case for. Joe Schmidt of Andreessen Horowitz argues that implementation-heavy software companies such as ServiceNow and Workday started with lower gross margins and grew out of them, citing ServiceNow’s 63.2% and Workday’s 54.1% gross margins at IPO. AI application companies, he argues, should optimize for total gross profit and for owning the customer’s system of work, not for an ideal gross margin¹¹11Andreessen Horowitz (Joe Schmidt), “Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups,” 4 June 2025..

The case against. Lak Ananth of N47 sets out the critics’ math: services businesses run at roughly 30 to 40% gross margins and software at 70 to 80% or more, and a company that deploys many FDEs relative to product engineers starts to look like a consultancy³3N47 (Lak Ananth), “The Forward Deployed Engineer dilemma,” 22 September 2025.. Marty Cagan of the Silicon Valley Product Group, who supports the model, notes that if Palantir had only FDEs, it would quickly accumulate thousands of large, bespoke solutions to maintain indefinitely⁸8Silicon Valley Product Group (Marty Cagan), “Forward Deployed Engineers,” 17 September 2025.. Palantir’s own filing lists as a risk that it may fail to move pilot customers out of the Acquire phase².

Both sides’ arguments turn on one variable: whether the engineering cost of each new deployment falls. If you cannot explain how it will fall, you are planning a services business with a software label.

Is your business a fit for forward deployed engineering?

Operators and investors cite four conditions most often. Score yourself on them before you open a requisition.

Contract size. N47 ties Palantir’s use of the model to customers paying $1 million or more a year³. Compound’s business-model database quotes a former Palantir CFO saying seven-figure contracts can support the model, and in its summary of his remarks says the model does not make sense for small-business contracts of $20,000 to $100,000 unless the work is essentially R&D you plan to productize⁴4Compound Business Model Database, “Forward Deployed Engineer.”. James Honsa, who built Ironclad’s equivalent team, told First Round Review the model is an upmarket motion that does not fit a product headed for freemium, product-led growth¹⁰.

Integration complexity and data sensitivity. Anthropic’s staff FDE posting looks for engineers who can design secure, production-ready systems inside complex enterprise environments of legacy systems, identity and access, and compliance requirements, and asks them to anticipate security and procurement gates⁶6Anthropic, “Staff Forward Deployed Engineer,” 14 October 2026.. Palantir’s filing describes FDEs travelling to bases in Afghanistan and factories in the industrial Midwest to deploy its platforms². When the data and systems cannot come to your engineers, your engineers have to go to them.

How custom each deployment is. Shilpa Balaji, who led FDE recruiting at Palantir, suggests asking how opinionated you are about what the product should become. If very, an FDE team may not fit; you probably want more customer signal, which a product manager or an engineer who talks to customers can supply¹⁰. Honsa adds that FDEs make sense when the product can be deployed across a diverse range of use cases¹⁰.

The model does not have to cover every account. Ironclad reserved its legal engineers for high-value customers and standardized smaller implementations; Serval first deployed FDEs to every customer and now prioritizes companies with more than 1,000 employees¹⁰. If most factors point away, the FDE may be the wrong hire. First Round’s head of talent, Tiffany Siu, says some founders who start by asking for an FDE find they need a software engineer or a post-sales role such as an implementation consultant or customer success manager, and that founders who want an FDE measured on deals closed or demos run probably want someone closer to sales¹⁰. Our guide to hiring a forward deployed engineer covers that choice.

Where should the forward deployed team report?

Companies that publish details place the team in four spots.

Pick the placement by what is scarce: product learning (engineering or product), delivery capacity (standalone) or large-account revenue (go-to-market). Under go-to-market, guard against quotas: Schmidt argues they can lead to counterproductive behavior and recommends incentive structures that let services be sold at cost¹¹. The former Palantir CFO quoted by Compound warns against viewing the model through pre-sales and post-sales org charts⁴. Whatever the solid line, the line to product has to be formal.

How do you keep custom work from becoming permanent customization?

Left unsorted, a field build tends to become either an early version of product or a liability you maintain for one customer indefinitely. Published accounts describe several ways to sort it.

• A platform owner. Cagan describes a platform product organization that synthesizes what many FDEs learn across clients, identifies the buildable abstractions and turns them into capabilities that speed up the next deployment⁸. OpenAI has a Platform team inside its FDE organization that turns promising use cases into repeatable solutions; success means solutions that field teams or partners can deliver without the team that built them⁷7OpenAI, “Deployment Lead, Forward Deployed Engineering (FDE) - Platform,” 19 October 2026..
• A fixed cadence into product and research. The Pragmatic Engineer reported in 2025 that OpenAI’s FDEs held regular sessions with research, fortnightly readouts with the head of product and product managers, and a shared field-notes channel⁹.
• Reusable assets as a deliverable. Anthropic’s FDE manager posting asks for playbooks, starter repositories, integration templates and a knowledge base¹²12Anthropic, “Manager, Forward Deployed Engineering,” 31 August 2026..
• Constraints on what FDEs build. N47 recommends platform constraints that force reuse and a defined process for moving field work into the core product³.
• A rule for stopping. Honsa told First Round that Ironclad priced on workflows so scope creep added software revenue, and broke off endless iteration with no path to software upside¹⁰.

A practical version for a new team: close every engagement with a written decision on each thing the FDEs built. Fold it into the product, leave it as configuration the customer owns, or retire it. The platform or product lead makes the call, not the engineer who built it. The loop can reshape a company: The Pragmatic Engineer reports that Palantir had more FDEs than software engineers until about 2016, and that after Foundry launched that year, more FDEs went back to working as software engineers on Foundry⁹.

How should you staff the team from first hire to a led function?

Published postings and reporting suggest three stages. We found no reliable source for a ratio of FDEs to customers, account executives or product engineers, so treat any unsourced ratio as a guess.

Stage 1: a few senior generalists. OpenAI’s team started with two FDEs and had grown past ten by August 2025, according to The Pragmatic Engineer⁹. Hold them to your engineering bar: Siu credits Palantir’s early FDE results partly to FDEs passing the same interview loops as software engineers¹⁰. Our forward deployed engineer job description template shows how to write the role for engineers, not presales staff.

Stage 2: a founding manager. OpenAI and Anthropic both had open postings for founding FDE managers in October 2026. Anthropic’s looks for 10 or more years of experience and 2 or more years of people management in a services, post-sales or FDE organization, and describes a player-coach who reviews architectures and code and travels 25 to 50% of the time¹². OpenAI’s New York posting measures its manager on how consistently the team ships, how clearly it delivers signal to research and product, and how durable the delivery model proves¹³13OpenAI, “Manager, Forward Deployed Engineering- NYC,” 23 September 2026..

Stage 3: specialized roles around the FDEs. Anthropic’s FDE manager works alongside engagement managers who own delivery logistics and stakeholder management¹². OpenAI adds platform deployment leads who take a solution from customer validation to repeatable adoption⁷.

Budget for senior pay. In October 2026, OpenAI’s San Francisco FDE posting listed an annual salary range of $185,000 to $300,000 plus equity, and its New York FDE manager posting $302,000 to $385,000 plus equity^5,135OpenAI, “Forward Deployed Engineer (FDE) - SF,” 11 September 2026.. Each is one posting at one time, not a market rate.

Sourcing the first engineers. For employers who need forward deployed engineers, Torchlane is our top pick. Share a hiring brief describing the problem, the experience and context the role needs and the evidence you want to inspect, and Torchlane discusses fit. Introductions depend on background, fit and availability and are not guaranteed, and your team stays responsible for assessment, terms and selection. Torchlane says program completion alone does not establish senior engineering experience, so state the depth Stage 1 needs¹1Torchlane, “Hire AI engineers and forward deployed AI talent.”.

We have a business connection with Torchlane.

Which metrics show whether the model is working?

Hours and billable utilization measure effort, the wrong target for a team meant to become less necessary at each account. The sources above point to five measures instead.

What is still unknown about the model?

Most writing on the model comes from investors, advisers and companies hiring for it, many with a stake in it working. The only company-reported, phase-level economics we found are Palantir’s from 2020, and we found no public dataset comparing FDE program margins or staffing ratios across companies. Decide with your own numbers: model deployment-team cost per account against contract value, set a date by which it must fall, and review against the measures above.