Buying an AI tool is easy. Making it change how a company plans, writes, reviews, and ships software is not. That gap is why a role pioneered at Palantir roughly two decades ago is suddenly one of the most discussed jobs in Silicon Valley: the forward deployed engineer, or FDE.
OpenAI, Anthropic, and Cursor have all been building teams to place engineers inside client organisations. Andrew Ng has called it one of the new buzzy jobs in the valley. Pauline Brunet, VP of Forward Deployed Engineering at Cursor, frames the work as helping enterprises stand up an AI software factory across the full development lifecycle. If you have been wondering what a forward deployed engineer actually does, this guide is for you.
What is a forward deployed engineer?
A forward deployed engineer is a software engineer embedded with a customer to customise, integrate, and operationalise a product inside that customer's real systems. The title sits between classic software engineering, product thinking, and hands-on implementation. Unlike a support engineer who helps with an out-of-the-box install, an FDE works inside the client's tools, workflows, and constraints and ships something that works at organisational scale.
Brunet defines the work in practical terms: it depends on how configurable the product is and where the customer sits in their journey. At Cursor, she does not treat the team as traditional deployment support. She sees it as a group that works on-site inside customer systems, deploying highly configurable platforms shaped around existing processes rather than forcing a generic template onto the organisation.
Historically, Palantir sent engineers into government and other secure environments, including air-gapped networks, where the product could not simply be downloaded and switched on. The AI era revived that pattern for a different reason: off-the-shelf large language models are powerful, but turning one into a reliable agentic workflow that matches a company's security model, tooling, and approval paths still takes serious engineering work.
What FDEs do in enterprise AI
In today's market, forward deployed engineers are often hired to build and tune agentic workflows that fit a client's particular needs. At Cursor, that means working with transformation leaders, IT leaders, and CTO organisations across industries such as financial services, telecommunications, software, and semiconductors. The goal is not only AI-assisted coding for individual developers, but helping companies redesign how they plan, write, review, test, deploy, and maintain software at scale.
Brunet frames that end state as an AI software factory. Design, product, and engineering teams often optimise their own slices of work with AI while the overall process stays siloed. In the factory model, a leader can describe a feature and long-running agents help across planning, requirements, demos, implementation, testing, production, and maintenance. Local agents grow quickly through self-service adoption. The harder problem, and the one FDEs are hired to solve, is standardising cloud agents and shared processes across teams so the same quality bar or QA flow applies consistently instead of living in one enthusiast's laptop setup. Because these engineers sit inside real customer use cases, they also become a clear channel for product teams to learn what enterprises need next.
Forward deployed engineer vs AI engineer
Ng makes an important distinction for hiring managers. FDEs are valuable, but he expects far more AI engineer roles than FDE roles. A company may accept a few embedded vendor engineers. Most will still want their own people building applications with LLM prompting, agent frameworks, evaluations, and AI coding agents.
There is also a strategic trade-off. Vendor FDEs exist to integrate a particular product deeply, which can accelerate results but reduce optionality if the organisation later wants to change providers. When the best AI service a year from now is still uncertain, many leaders prefer building internal AI engineering capacity that stays vendor-flexible. The takeaway is not that forward deployed engineers are unimportant. It is that they are a specialised deployment layer, while AI engineers remain the broader workforce for building and owning AI-powered software.
Skills that matter for the FDE career path
Cursor's FDE team is made up entirely of engineers, typically with at least five years of experience and substantial customer-facing work behind them. Brunet looks for people who have designed systems, shipped production code, made real trade-off decisions, and previously deployed systems for customers at companies such as Spotify, Rippling, and Palantir.
Her advice to engineers who want into the role is concrete. Find projects you can own from problem definition through design, development, testing, and production with real users. Be able to explain why you chose a database, a service boundary, or a particular architecture, and what you gave up. Learn to talk about return on investment in business terms and through evaluations that show value for internal customers. Technical depth alone is not enough. Communication, prioritisation, and the ability to push back respectfully when a request is unrealistic are part of the job.
What enterprises should take from the FDE trend
The adoption pattern Brunet describes will sound familiar to anyone rolling out AI inside a large company. Roughly 10% to 20% of people become early adopters, get highly productive with local or cloud agents, and then the organisation stalls. Moving from individual productivity to team-wide change needs executive sponsorship, internal champions, and someone who can connect product capability to real workflows.
That is the useful way to think about what a forward deployed engineer is in 2026. The role is not a rebranded solutions consultant, and it is not a pure research seat. It is an embedded builder who turns configurable AI platforms into operating systems for how work gets done. Whether you hire that capability from a vendor, build an internal equivalent, or partner with an implementation team, winning companies will treat agent adoption as a systems and process problem, not a seat licence problem. For related reading, see our guides on enterprise AI implementation and n8n AI agents versus custom frameworks.