If you've been following talent trends in the international tech world recently, you may have noticed that a job title few people had heard of has suddenly become something that top AI companies like OpenAI, Anthropic, and Google are scrambling to hire: the FDE (Forward Deployed Engineer).
What is an FDE? Simply put, it's an engineer that an AI company sends directly into a customer's company. They first analyze the customer's situation, then carry the AI system all the way to actual deployment, until users are satisfied.
In recent months, US tech companies have been spending big money to recruit FDEs. At the end of June 2026, AWS announced it would invest US$1 billion (about NT$323 billion) to establish a dedicated FDE organization; on July 2, Microsoft announced it would invest US$2.5 billion (about NT$806 billion) to create a new business unit, "Microsoft Frontier Company," with roughly 6,000 engineers and industry experts—some transferred internally, some hired externally. Within two days, the two cloud giants together poured in about US$3.5 billion (about NT$1,129 billion), with the goal of placing their engineers directly inside customers' companies.
Then, on July 8, OpenAI's deployment company (OpenAI Deployment Company) announced it had agreed to acquire Northslope. Most people have probably never heard of Northslope, but it was founded by a former Palantir FDE, and this is the deployment company's second acquisition since its founding in May 2026 (the first was Tomoro). In fact, from its founding in 2024 to this acquisition, Northslope was only about two years old, and its revenue in 2025 grew nearly sevenfold. The fact that such a young company still in explosive growth was snapped up directly by OpenAI shows just how in-demand top FDE teams are today.
According to talent market analysis cited by Business Insider, FDE job postings grew from 643 to 5,330 in one year (April 2025 to April 2026), an annual increase of 729%. Beyond AI and cloud companies like OpenAI, Anthropic, Google Cloud, Palantir, and Stripe actively recruiting, enterprises have started hiring as well. Another study conducted by the headhunting firm Christian & Timbers and obtained by TechCrunch found that in early 2026, only 5% to 10% of surveyed companies planned to hire FDEs, mostly for small-scale pilots, but by the end of Q2, that figure had risen to 70%. The focus of competition in the AI industry is shifting from the race for cutting-edge models to who can get enterprises to actually buy in and use the technology.
Why would a role that, at first glance, just "sends engineers to the customer's site" occupy the most coveted position in the entire AI industry?
What exactly does an "FDE" do?
To understand why FDEs are so hot right now, I think it's related to the trend of enterprises deploying AI over the past two years. MIT's NANDA project, in its report "The GenAI Divide: State of AI in Business 2025," surveyed more than 500 enterprises and found that a full 95% of enterprise generative AI pilot projects ultimately produced no quantifiable impact on the P&L—in other words, companies spent big money adopting AI but the results fell short of decision-makers' expectations.
So is the real cause the models themselves or insufficient compute? I don't think it's that simple. Today's large language models are already quite capable; the answer doesn't lie on the model side, but on the front line: how exactly should AI be connected to existing systems? How do you meet permission and compliance requirements? These are things that change employees' workflows, yet no relevant resources were on-site. This year, people finally realized the importance of the FDE. Customers understand their own systems, AI companies have the models—each side knows half—but only the FDE can stand in the middle and make the system truly work its way into everyone's jobs.
But what makes this role special is what they do all day long.
A traditional consultant sits in a conference room, listens to requirements, produces a report telling you "you should do it this way," and execution is left for the company to worry about. But an FDE sits directly in the customer's office, going to ask the employee who runs the process every day, "Why is this step done this way?" "Where does it get stuck?" They quickly throw out a working version, then keep iterating based on on-site feedback, until the system actually works.
Suppose six months later the system goes live and breaks down in the middle of the night—the person who originally mapped out the problem architecture is the one who gets called in to fix it. While solving the problem, they also bring back what they learned on-site, making the system better.
For many enterprises, the first hurdle in adopting AI is messy, outdated documents—and further down, things get even harder: when you let AI call internal company systems, who has permission to do what, which data can't be touched—all of this has to be wired up piece by piece. These problems never show up at the demo stage; only by actually walking into the company do they surface. That is precisely why the FDE exists.
It was Palantir that first turned this approach into a scalable business. In 2011, Palantir merged its previously separate Solutions Engineers and integration engineers into a single role, called the Forward Deployed Engineer. At the time, many people thought this model was too expensive and couldn't scale like pure software—that it was a costly consulting service. It wasn't until the past two years, when OpenAI, Anthropic, Google, and others wanted to make money on the enterprise side, that they discovered simply handing tools to customers to use on their own wouldn't work, and so they too began hiring for this role at scale.
The concept of the "forward deployed engineer" actually existed in the hardware industry long ago
Seeing this wave of FDE enthusiasm, I've long wanted to weigh in on this topic, because for anyone who has worked in hardware, this is nothing new.
In the semiconductor and hardware industries, there has always been a role called the FAE (Field Applications Engineer). FAEs work closely with customers' products and engineering teams, providing technical support at different stages, from component selection and design-in to testing, debugging, and mass production ramp-up.
To build a phone, you need memory, display panels, and all kinds of components. A supplier's FAE isn't just responsible for selling components to the customer; they also go deep into the customer's R&D and testing sites, helping the product all the way to mass production. Foundries have similar customer support mechanisms. Take TSMC, for example: its teams collaborate with customers on design support, masks, manufacturing, and new product introduction, ensuring chip designs transition smoothly into the process and into mass production.
The most classic example is MediaTek. In its early years, it offered turnkey (one-stop) solutions for white-label phones, bundling chips, reference designs, software, development tools, and testing support into a package. Buyers could almost just slap on their own brand and ship. The key to making this model work was precisely that group of FAE engineers embedded at the customer's site. You could say it was the FAE playbook of "staying with the customer until success" that propped up MediaTek's business.
So when I saw Silicon Valley AI companies expanding FDE hiring in recent years, my first reaction was: "Isn't this just the hardware industry's model of helping customers achieve rapid deployment, replayed on a new battlefield?"
AI is a general-purpose technology, but the more general-purpose it is, the more enterprises need someone to go deep into existing systems, data, and workflows to complete that final stretch of deployment. The responsibilities of FAEs and FDEs aren't identical, but the core logic is similar: it's not about delivering a product and being done, but about working with the customer to truly put the technology into their workflows.
If FDEs and FAEs are so effective, why doesn't every software company do it?
If this model is so effective, why isn't every company doing it?
Because it conflicts with the business models of many software companies. A software company's ideal business is to build one standardized product and sell it to as many customers as possible—ideally, customers can use it on their own after buying—so that marginal costs approach zero. Asking them to station engineers on-site for long periods, customizing for one customer after another, simply looks unprofitable in a financial model.
So some software companies choose another path: building a product tailored to a specific industry and serving that industry exclusively, which can also be very effective in the right scenarios. But whenever a new technology emerges, you inevitably have to change along with the industry's needs.
SaaS's success is built on standardization, while the value of the FDE comes from diving deep into different needs. Yet the complexity of AI adoption is pushing back the boundaries of standardization, forcing many software companies to rethink this balance.
Coming back from the wave in Silicon Valley to Taiwan, what impact will it bring?
It's worth noting that when AWS announced its investment in FDEs, it specifically emphasized how they differ from traditional consultants: traditional consultants do assessments and give recommendations, treating each case as a one-off project, whereas what FDEs leave behind for users is not just a system that runs, but also a team and the capability to operate independently and keep innovating.
In fact, Taiwan has a huge manufacturing and hardware sector, and the FAE spirit of "walking into the customer's site and staying with them until success" is something we've always been most familiar with and best at. When waves from Silicon Valley reach Asia, they often take on a different shape, but to me, this FDE craze is just another confirmation of something the hardware industry understood long ago: real value is always found at the customer's site. I'm also looking forward to seeing what kind of change this will bring to Taiwan's software industry.