A forward-deployed engineer (FDE) is a software engineer employed by a technology vendor who works inside a customer's organisation, often on its premises and on its infrastructure, to make the vendor's product do a specific job for that customer. The role exists because some software does not work out of the box. Data sits in old systems, workflows are particular to one company, and security rules stop outside tools from connecting. Someone has to write the glue code, and the FDE model puts that person on the vendor's payroll rather than leaving it to the client or a consultancy. The title, which borrows its first word from military usage, came out of Palantir and spread across the artificial intelligence (AI) industry in 2025 and 2026, when model developers found that enterprises were buying AI faster than they could deploy it.

How the role works

Palantir's own internal shorthand still gives the clearest definition. Product engineers, known inside the company as "Devs," work on "one capability, many customers," while forward-deployed engineers, known as "Deltas," work on "one customer, many capabilities," according to definitions published by The Pragmatic Engineer in August 2025. A Dev builds a feature once and ships it to everyone. A Delta takes the whole platform into one account and bends it to fit.

At OpenAI, the work runs in three phases - scoping, validation and delivery - according to Colin Jarvis, who heads the company's FDE team, as described by The Pragmatic Engineer. The distinguishing feature is that FDEs write code directly on customer infrastructure, while OpenAI's solutions architects rarely do. A typical engagement at the OpenAI Deployment Company starts with a focused diagnostic, narrows to a small number of priority workflows, and then moves to systems that connect OpenAI models to the customer's data, tools, controls and business processes, according to OpenAI's announcement of May 11, 2026.

Amazon Web Services (AWS) has published more concrete numbers. Its FDE organisation works in groups of roughly five or six engineers on engagements lasting about 45 days, according to Techstrong.ai. Travel loads vary. Palantir expects about a quarter of an FDE's time to be on site, while healthcare software company Commure estimates up to half, according to The Pragmatic Engineer.

The second half of the job is less visible. Code written for one client is supposed to flow back into the product when it proves general. Palantir's Foundry platform, launched in 2016, absorbed much of the bespoke work its Deltas had built and pulled many of them back into core engineering. Until around that time Palantir employed more FDEs than conventional software engineers, according to The Pragmatic Engineer.

In marketing technology, the role looks similar. A PubMatic posting dated April 10, 2026 described an FDE embedded in go-to-market and operations teams, building Model Context Protocol (MCP) servers, sub-agents and retrieval pipelines that feed into its AgenticOS platform, with pay listed at $200,000 to $300,000. Adverity advertised a London FDE role on October 6, 2026 to support the launch of its Atlas AI product at £65,000 to £90,000. PMG lists an FDE team working on its Alli operating system, covering tagging, audiences and measurement integrations.

Origin and spread

The origin date is disputed. Shyam Sankar, now Palantir's chief technology officer, says he coined the term in 2007 after chief executive Alex Karp asked him why French restaurants were so good and told him to build the equivalent for engineering, according to The Register, which did not independently verify the date. Andreessen Horowitz (a16z) places the title in 2011 in a January 2026 post, and The Pragmatic Engineer dates it to "the early 2010s." Sankar described the purpose on his Substack: "FDEs embed alongside our customers and work to ensure our software solves their problem and not some proxy for their problem."

Palantir's early customers were intelligence and defence agencies, where even a demonstration could require weeks of non-disclosure agreements and security clearances. Sending engineers in, rather than shipping software out, solved an access problem as much as a technical one.

The model's second life began with generative AI. On June 4, 2025, a16z partner Joe Schmidt argued in "Trading Margin for Moat" that AI start-ups should accept lower gross margins early in exchange for deep customer integration. He pointed out that ServiceNow and Workday went public with gross margins of 63.2% and 54.1% respectively, well below the roughly 80% associated with pure software, and later climbed to 79% and 75% by 2024. At the time of writing, 22 of 311 open roles at OpenAI were FDE or solutions positions, by Schmidt's count. "Enterprises buying AI are like your grandma getting an iPhone: they want to use it, but they need you to set it up," he wrote.

Hiring followed. Monthly FDE listings rose more than 800% between January and September 2025, a figure that eWeek attributed to Indeed data and other outlets to the Financial Times. LinkedIn data cited by AWS put the increase in demand for FDE and related roles at 42-fold between 2023 and 2025. In India, Indeed recorded roughly 19-fold growth in postings between June 2025 and June 2026, according to Analytics India Magazine. The base matters: these percentages start from a small number of listings, and none of them measures total headcount. Levels.fyi puts the median total compensation for a US FDE at $216,250, with the 25th percentile at $180,000 and the 75th at $290,000, according to The Register.

Why it matters for marketers

Marketing data is fragmented across ad platforms, customer relationship management systems, clean rooms and measurement vendors, which makes it a natural target for embedded engineering. Zeta Global's second-quarter 2026 materials said forward-deployed engineers are embedded onsite at Gap, which chose Zeta as the system of record for its marketing stack. NielsenIQ (NIQ) built its ConnectAI programme on the same pattern, pairing data access with engineers and data scientists embedded in client businesses and working with the OpenAI Deployment Company. By its third-quarter results, NIQ reported 10 signed ConnectAI charter clients, including Purina and Mars Snacking.

For model developers, FDEs are the route into marketing departments. Sam Altman described OpenAI's FDEs in a February 5, 2026 interview as the people who go into companies that are "not AI native." Anthropic's registration statement says its Applied AI teams, which include forward-deployed engineers, design and deploy custom solutions for enterprises. A Reddit discussion covered by PPC Land in October 2025 found practitioners describing 70-80% of AI agent work as integration, the territory FDEs occupy.

Limitations and criticisms

The oldest objection is that the FDE is a consultant with a better title. "Forward deployed engineering is being framed as a panacea right now," James Honsa, a former Ironclad executive who now runs Genera, told First Round Review, calling it "a pretty blunt instrument" for building a whole business. Semafor's Reed Albergotti described the trend in July 2025 as a tacit acknowledgement of the limits of AI, since products that worked unaided would not need engineers attached.

Economics are the second problem. Every FDE is a salary attached to one account. Frank Bien, former chief executive of Looker, which ran an early version of the model, said the company "needed to be in the neighborhood of $25,000 a year per customer" to justify it, according to First Round Review. Consultancy Alvarez & Marsal lists "perpetual bespoke work" and teams that "solve the sponsor's easy problems" among the failure modes, and warns of drift into pure services.

Buyers face lock-in. Mukul Saha, a senior director at Gartner, advises clients to understand their use cases before FDEs arrive and to "figure out the exit strategy," according to The Register. Code written by a vendor's engineers on a client's systems tends to deepen dependence on that vendor's platform and pricing.

What it is not

Solutions architect or sales engineer. Both advise and demonstrate, usually before a sale. FDEs build production code after one, and stay longer.

Consultant or systems integrator. Accenture or Deloitte work across vendors and bill for time. An FDE works for one vendor, is usually not billed separately, and is expected to feed code back into the product.

Frontier deployed engineer. Anthropic's Claude Frontier Academy, announced October 2, 2026, uses the same initials for a certification. It trains engineers at partner firms and customers, such as Deloitte and McKinsey, rather than Anthropic's own staff.

Echo or deployment strategist. Palantir pairs Deltas with "Echo" staff - embedded analysts and account leads with domain knowledge - according to Alvarez & Marsal. They own the customer relationship, not the code.

Recent developments

Large vendors moved in 2026 from hiring individuals to building units. OpenAI launched the Deployment Company on May 11, 2026 with more than $4 billion of initial investment, majority owned by OpenAI and backed by 19 firms led by TPG, and agreed to buy consultancy Tomoro, bringing roughly 150 FDEs and deployment specialists. AWS committed $1 billion to an FDE organisation, reported on July 1, 2026, which Techstrong.ai said would be staffed by thousands of engineers. Microsoft announced a $2.5 billion "Frontier Company" in July 2026, according to The Register. Anthropic committed $100 million on October 2, 2026 to train 10,000 engineers by the end of 2027.

Software makers outside AI followed. Salesforce chief executive Marc Benioff told investors in December 2025 that customers using its FDEs deployed its AI 33% faster, according to The Register. On November 11, 2026, Xero introduced Xforce, a team of FDEs and partner consultants for accounting firms, in beta in Australia, New Zealand and the UK.

Timeline

  • 2007 - Shyam Sankar says he coined "forward deployed engineering" at Palantir (his own account).
  • 2011 - Date a16z gives for Palantir creating the FDE title, known internally as "Delta."
  • 2016 - Palantir launches Foundry, moving much bespoke FDE work into the core product.
  • Start of 2024 or early 2025 - OpenAI forms its FDE team; eWeek and The Pragmatic Engineer give different dates.
  • June 4, 2025 - a16z publishes "Trading Margin for Moat," calling the FDE the hottest job in start-ups.
  • January-September 2025 - Monthly FDE job listings rise more than 800%.
  • August 2025 - OpenAI's FDE team reaches more than 10 people in eight cities.
  • December 2025 - Salesforce says FDE-supported customers deploy AI 33% faster.
  • January 27, 2026 - a16z publishes "Forward-deployed Job Titles."
  • February 5, 2026 - Sam Altman describes OpenAI's FDE work in a live interview.
  • April 10, 2026 - PubMatic posts an FDE role tied to AgenticOS.
  • May 11, 2026 - OpenAI launches the Deployment Company with more than $4 billion and agrees to buy Tomoro.
  • July 1, 2026 - AWS's $1 billion FDE organisation is reported.
  • July 2026 - Microsoft announces a $2.5 billion "Frontier Company."
  • August 2026 - Zeta discloses onsite FDEs at Gap; NIQ links ConnectAI to the OpenAI Deployment Company.
  • October 2, 2026 - Anthropic launches Claude Frontier Academy with $100 million for 10,000 engineers.
  • October 14, 2026 - Indeed data shows FDE postings in India up about 19-fold in a year.
  • October 26, 2026 - Anthropic's IPO filing names forward-deployed engineers in its Applied AI teams.
  • November 11, 2026 - Xero announces Xforce.

Summary

Who: Software engineers employed by vendors such as Palantir, OpenAI, Anthropic, AWS, Salesforce, PubMatic, Zeta and NIQ, working inside customer organisations, often alongside account and domain specialists.

What: A role that combines product engineering with on-site delivery, writing production code on the customer's systems so a vendor's software or AI models work in that customer's workflows, with reusable code fed back into the product.

When: Created at Palantir between 2007 and the early 2010s, depending on the source, and revived from 2025 as AI companies scaled enterprise sales, with dedicated units launched by OpenAI, AWS and Microsoft in 2026.

Where: At customer sites and inside customer infrastructure, with demand concentrated in the US and growing quickly in India and Europe, and spreading into marketing technology, measurement and ad tech.

Why: Enterprise AI and data products rarely work without integration, and vendors use FDEs to shorten deployments and lock in accounts, at the cost of lower margins, a risk of becoming a services business, and dependence for the buyer.