What is a forward deployed engineer?

Google searches for "forward deployed engineer" grew from about 6,600 a month in July 2025 to 40,500 in June 2026. Six times the demand in eleven months, for a job title most people had never heard before last year. You will also see it as forward deployed software engineer (FDSE, the original Palantir title) or AI forward deployed engineer. Same role.

I watched that curve with a strange sense of recognition. I spent 20 years at Verizon, the last stretch as an Associate Vice President leading 450+ people across solutions architects, pre-sales engineering, and customer success in national technical sales. Solutions architects and sales engineers are exactly the roles the FDE conversation keeps defining itself against. So when the AI world announced it had discovered that someone has to make the software actually work inside the customer's building, my first reaction was, welcome. Field teams have been living that problem for decades.

But the FDE role is not just a rebrand, and the companies betting on it are not small. This page covers where the role came from, why it exploded, what it pays as of July 2026, how it differs from the roles it gets confused with, and what the trend means for the thousands of companies that will never appear on Palantir's client list.

01Where the role came from: Palantir

Palantir popularized the role in the early 2010s under the title forward deployed software engineer. Their early customers were intelligence agencies, defense, and heavy enterprise. You could not just install software for those customers. Data sat in silos, networks were air-gapped, and requirements changed weekly. So Palantir put strong engineers at the point of use.

The internal logic mattered more than the title. Palantir's product engineers built one capability for many customers. Their forward deployed engineers built many capabilities for one customer. And when something a field engineer built kept proving useful, it got generalized into the platform. That loop, field invention feeding the product, is what separated the model from consulting. Foundry, the platform Palantir sells today, is the textbook result.

Barry McCardel, a Palantir alum who later co-founded Hex, wrote the essay most practitioners point to on how this culture actually worked. His warning is worth carrying through the rest of this page: most companies that slap the FDE title on a sales engineering team are not running the real model. Real FDE work is expensive, chaotic, and only fits certain markets. The title is easy to copy. The loop is not.

02Why FDE exploded in 2025 and 2026

Enterprise AI hit a wall, and the wall has a name: the last mile. Models look magical in a demo and fragile inside a bank, a hospital, or a factory. Dirty PDFs, mainframes, permission systems, compliance reviews, and processes that live in one employee's head. An MIT report made the rounds in 2025 with a number that stuck: 95% of enterprise generative AI pilots were producing no measurable return.

The companies selling intelligence noticed that the bottleneck was not intelligence. It was deployment. Then the hiring signals stacked up fast:

• OpenAI posts FDE roles publicly, and in May 2026 launched a dedicated Deployment Company, reported at a $4B investment structure with roughly 150 deployment engineers folded in through its Tomoro acquisition.
• Google Cloud leadership went on record recruiting for FDE-style roles, with press coverage putting the push in the hundreds. Anthropic hires for the same work under applied AI titles.
• a16z ran a forward deployed engineer fellowship, reportedly selecting an inaugural class of 65 from thousands of applicants.
• Andrew Ng, in June 2026, called FDE a genuine new career path while predicting there will be far more AI engineer jobs than FDE jobs long term. Both halves of that sentence matter.

Demand for the term followed the money. That 6x search growth is not job seekers alone. A meaningful slice is executives trying to figure out what they just heard on an earnings call.

03What a forward deployed engineer actually does

Strip away the mystique and the work has three phases. I am compressing here from Palantir lore, OpenAI's own job description, and the writeups of practitioners who run this model for a living.

Phase 1: Learn the business reality. The documented process is almost never the real process. "An email arrives" turns out to mean forty senders, no consistent format, half the data in PDFs and screenshots, and the actual routing logic living in one person's head. FDEs sit with the people doing the work, sometimes for full shifts, because the exceptions never make it into the SOP. This phase is where most of the value gets created, and it is the phase pure technologists skip.

Phase 2: Decide where intelligence belongs. This is judgment, not enthusiasm. A 10-step workflow might need an LLM in exactly two steps. The rest is deterministic software: rules, API calls, plain code. The early AI era ran the opposite play, throw the model at everything and hope. MIT's 95% figure is what that era's report card looked like. The FDE's job is to say no to most of the workflow and yes to the two steps where probabilistic judgment actually pays.

Phase 3: Ship it, prove it, and stay accountable. Production code in the customer's environment. Evaluation suites that prove the system behaves before it touches real work. Human approval gates on anything consequential. Audit trails, because a business will not trust an agent it cannot inspect. And when something breaks at 2 a.m., the FDE is the one whose phone rings.

The cleanest one-line test I know: a consultant's deliverable is a recommendation, a sales engineer's deliverable is a closed deal, an FDE's deliverable is a system running in production that the customer's own team actually uses.

04What forward deployed engineers earn (as of July 2026)

Compensation claims for this role are where the internet gets loudest and least reliable. There are viral posts claiming $1M a year. Here is what published sources actually show, ordered from most to least verifiable:

The honest summary: strong senior engineer money everywhere, with frontier-lab equity as the lottery ticket on top. The million-dollar FDE exists the way the million-dollar salesperson exists. Real, rare, and a terrible planning assumption.

05FDE vs solutions engineer, consultant, and AI engineer

I led solutions architects and pre-sales engineers for years, so this is the part of the FDE conversation I can pressure-test from experience. The difference between these roles is not talent. It is where accountability ends.

A great sales engineer's job ends when the deal closes. An FDE's job starts there. That single difference explains the comp, the burnout stories, and why Palantir treated field teams as R&D instead of cost of goods sold.

It also explains the failure mode. When a company renames its SE team to FDEs without giving them authority to build, or without a platform that can absorb what the field learns, it gets consulting margins with software salaries. McCardel's rule: if finance sweats every FDE hour as pure cost and the field cannot invent, you are doing solution engineering. Just say so. Marty Cagan's take at SVPG lands on the same point from the product side: the durable value shows up when field learning gets synthesized into a platform, not when you accumulate a thousand bespoke snowflakes.

06What the FDE trend means if you run a business

Most of the FDE content wave is career advice. Fair enough, the comp is real. But the more useful lesson sits on the other side of the table, because the FDE model is a diagnosis of why AI projects fail, and the diagnosis applies at every company size.

The pattern that works is audit first, evals second, deployment third. Map how the work actually happens, including the exceptions. Decide where AI belongs and, more importantly, where it does not. Build mostly deterministic software with LLM judgment only at the steps that need it. Keep a human approving anything consequential. Then measure against the only three buckets a business cares about: revenue lift, cost savings, risk reduction.

None of that requires hiring from Palantir. It requires someone technical enough to build, senior enough to say no, close enough to your operation to learn how the work really flows. That combination is the actual scarce resource. The title is optional.

07FDE-style deployment for companies that will never hire Palantir

OpenAI's Deployment Company and Palantir's field teams serve governments and the Fortune 500. The economics require seven-figure contracts. Nobody is forward-deploying a $400K engineer into a 60-person manufacturer, a regional lender, or a mid-market healthcare group.

Those companies have the same last-mile problem. The workflows are just as messy, the exceptions just as undocumented, the failed-pilot risk just as real. What they need is FDE-style deployment at local scale: someone who sits with the team, maps the real process, ships on top of the systems already in place, and stays accountable until it works.

That is the work I do now from Spokane, Washington, after 20 years at Verizon, most of them running field organizations, and the AI products I have built since leaving in 2024, which now serve about 30,000 people a month. Embedded AI deployment for companies that will never be on a lab's client list. If that is the problem you are staring at, start with how I approach AI automation for Spokane and Inland Northwest businesses, or see how I audit AI visibility for a specific industry at CredibilityOS.

The FDE wave will crest, and the title may fade. The lesson underneath it is durable, and it is the same one I learned running turnarounds in four Verizon markets: the work happens where the work happens. Software that ignores that fails at any price point.

08Questions people actually ask

Last updated: July 22, 2026