While AI technology advances rapidly, what troubles enterprises even more is: how to actually run models in business scenarios and deliver quantifiable value.

In the past, this task was handled by Solutions Architects, who took on the responsibility of "implementation"; but now, more and more AI companies are focusing on the "Forward Deployed Engineer" (FDE).

So how hot is this emerging profession that "puts AI first"?

According to the Financial Times, citing data from the job platform Indeed, demand for Forward Deployed Engineers surged in 2025 — job postings from January to September this year alone grew more than 800% compared with the same period last year.

OpenAI established a Forward Deployed Engineer team at the beginning of the year, expecting to expand it to 50 people in 2025; Anthropic also expanded its Applied AI team, including this role, fivefold.

In the plainest terms, a "Forward Deployed Engineer" is a software engineer role that works close to the customer's front line. They don't just write code — they go to the customer's site to understand business processes, quickly adapting, integrating, and implementing the company's products so customers can truly use them, afford them, and use them well.

How did the Forward Deployed Engineer come about?

According to Business Insider, the Forward Deployed Engineer originated from Palantir's practical needs in its early collaborations with highly classified organizations such as intelligence agencies and the military. Since customers found it difficult to fully describe their requirements remotely, Palantir adopted an approach of having engineers embed directly on-site, work side by side with users, and rapidly iterate to turn prototypes into usable systems; as this methodology was gradually institutionalized, the FDE was born.

Palantir's official blog notes that the FDE is internally called "Delta" and often works in tandem with the Deployment Strategist (internally called Echo), a role more oriented toward product strategy/PM. The core function is to embed long-term with the customer, directly using engineering capabilities to configure and extend the platform (Gotham, Foundry) into workflows and applications that solve specific tasks — not merely offering advice as a consultant; the overall culture is R&D-oriented.

Therefore, Palantir's FDE model essentially treats "customer deployment" as R&D rather than cost of goods sold (COGS) : it allows the front line to "build it first" to solve specific high-value problems, even if short-term efficiency or gross margin suffers, in exchange for long-term commercial returns through platform feedback and wider adoption. This approach requires a high headcount threshold, tolerance for rework and failure, and giving roadmap authority to the front line.

To this day, Palantir officially still calls this role FDSE and emphasizes "we pioneered this role." The job description maintains embedding with customers, delivering impact through engineering, and feeding back into the platform; it also explicitly states requirements such as traveling to customer sites (about 25%).

Why do AI companies rely so heavily on FDEs?

In the great AI era, the reason FDE roles are so hot is that for generative AI/agents to enter every industry, each scenario is highly heterogeneous and lacks mature standards, and enterprises often don't know how to implement and integrate them into their processes.

FDEs can complete "product discovery" and "process redesign" on-site, shortening the distance from technology to value. This is also why AI startups like OpenAI and Anthropic have expanded their FDE/Applied teams in recent years.

Simply put, the core spirit of this model is: distilling frontline insights back into the platform product, reducing the customization cost and time of future projects.

According to Oliver Jay, OpenAI's Managing Director of International, in the process of expanding enterprise customers, the biggest bottleneck over the past year has been moving successful pilot scenarios (PoC/pilots) into production. For example, data permissions, eval standards, risk management, and regulatory requirements — none of these can be skipped when deploying AI. Therefore, OpenAI's response is to send engineers to fight side by side with customers, ensuring that eval frameworks, metrics, and guardrails grow along with the product.

Notably, in Palantir's practical experience, the FDE role is often called an "entrepreneurship bootcamp." Because a mature FDE must be able to talk to investors and customers externally, while internally writing good product code and delivering well. This belief has extended into the AI startup ecosystem: being able to get a project done on-site is the best training for entrepreneurial ability.

In summary, the responsibilities of an FDE can be grouped into the following areas:

1. Requirement implementation: go deep into the customer's site, breaking down vague business pain points into developable technical tasks and deliverables. 2. Product customization and integration: perform secondary development on top of existing products, connecting the customer's data, systems, and permissions, handling compatibility and performance. 3. Rapid delivery: launch usable versions in short iterations, validating value, fixing problems, and scaling up, balancing stability and speed. 4. Troubleshooting: facing the various "pitfalls" of real environments — from networking, data quality, and deployment to security compliance — locate and fix issues immediately. 5. Bridging communication: communicate with both the customer and internal product/R&D teams, turning field feedback into product improvement directions, preventing either side from losing focus.

How good is FDE pay?

Palantir: the founding giant of the FDE

  • Compensation: In the US, Forward Deployed Engineer/FDSE annual total packages fall roughly between $171k–$347k; the Levels.fyi median is about $211k, and Glassdoor estimates a median of about $256k.
  • Stock: RSUs (PLTR is publicly listed), which are tradable equity with market liquidity. So while the salary may look lower than AI companies', Palantir's stock is publicly traded (PLTR), fully liquid — once you get it, it's real money.
  • Job content: frontline embedding with customers, cross-data integration, and production operations; frequent travel and relatively high pressure and working hours consistently appear in both official and employee accounts.

OpenAI: the salary ceiling for veterans

  • Compensation: Levels.fyi shows Software Engineer L4 ≈ $575k, L5 ≈ $998k, L6 ≈ $1.25M per year; the company-wide median compensation is about $538,860 per year.
  • Title differences: On Glassdoor, "Member of Technical Staff" samples estimate total packages of $151k–$220k, median ≈ $181k; this differs from senior engineering levels and should be distinguished when comparing.
  • Stock: OpenAI issues PPUs (Profit Participation Units). Although theoretically "paper wealth," OpenAI regularly holds Tender Offers (buybacks) that let employees cash out (for example, large-scale buybacks occurred in both 2024 and 2025).
  • Job content: more like a "technical special forces unit," responsible for helping major customers (such as JPMorgan Chase and Coca-Cola) deploy GPT-4 models and solving high-difficulty integration problems.

Anthropic: a potential star in a period of explosive growth

  • Compensation: Levels.fyi shows Senior Software Engineer ≈ $550k (base ≈ $318k, equity ≈ $233k), Lead Software Engineer ≈ $700k (base ≈ $328k, equity ≈ $372k); the median for this job family is about $570k per year.
  • Stock: the company's valuation is growing extremely fast (from billions to tens of billions of dollars). If you believe Claude can surpass GPT, the potential multiple returns (upside) on stock here could be higher than OpenAI's.
  • Job content: emphasizes AI safety and enterprise applications; the work culture is somewhat more "academic/rigorous" than OpenAI's, but equally high-pressure.

Conclusion: FDE is a great role — wouldn't you consider switching careers?

In summary, the battle in AI isn't about how fast models run, but how reliably value lands in practice. The role of the FDE is to turn the pretty curves on the demo stage into quantifiable results in real operations: hooking up data pipelines, eliminating process friction, and feeding real user feedback back into the product.

As a next step, rather than chasing the latest parameters, ask first: does your team have someone who can get systems running in frontline scenarios and keep optimizing them? If not, start by recruiting an FDE.

For job seekers, an FDE must be able to write code, handle a client CEO's questions, fight fires over bugs, and turn lessons learned into repeatable solutions. Remember, the market isn't looking for people who can talk about models—it's looking for people who can make models reliably deliver value.

The AI Battle Heats Up! Anthropic Launches Claude Opus 4.5: Does It Crush Gemini 3 Pro in Coding, Agents, and Computer Use?

Sources: Palantir, Business Insider, FT, Big Tech Careers

This article was drafted with AI assistance; compiled and edited by / Lee Hsien-Tai