The forward-deployed engineer is the hottest job in technology. FDE job postings grew more than 700% over the last year. The hyperscalers and frontier labs now employ them. So do a growing number of health tech companies.

This is quite strange. For the past two decades, SaaS economics favored building the same, highly configurable product that many customers could adopt. As Elion’s Colin Durant recalled, “customization was where margins went to die.” So, as AI makes it easier to write code, why are so many companies now employing expensive engineers to sit alongside customers?

It turns out that automation paradoxically requires a lot of expensive manual labor, at least initially. As AI expands what we can build, we still have to figure out what to build and actually make it work.

FDEs are emerging as one answer. But what exactly do they do?

“Enterprises buying AI are like your grandma getting an iPhone: they want to use it, but they need you to set it up.” Joe Schmidt, A16Z

What Is A Forward Deployed Engineer?

Traditionally, product designers and sales teams went to customers, while software developers stayed back at headquarters. Palantir famously pioneered a different approach: discovering and developing products inside customer organizations.

FDEs are technical people who sit alongside front-line users to identify and understand their problems, then build and refine software to solve them.

This proximity has three benefits.

First, it enables better product discovery. FDEs closely observe workflows, surface tacit knowledge, identify edge cases, and uncover problems that they may otherwise miss. This is painstaking work. Palantir Chief Architect Akshay Krishnaswamy explained , “The FDE’s job is to absorb pain and excrete product.”

Second, bundling discovery and building can accelerate execution. After wiring their platform into underlying data and systems, FDEs can iterate quickly with end users, without depending on local tech teams’ bandwidth.

Third, rather than forcing customers into a fixed product, FDEs can adapt solutions to local context. This matters because health systems are a bit like snowflakes. Each has its own electronic health record instance, technology stack, governance rules, team structure, legacy workflows, culture, and more.

If traditional SaaS is like a prefabricated kitchen with configurable options, FDEs are like hiring a craftsman to build a custom kitchen.

These benefits make FDEs particularly well-suited for building AI agents , which, in many ways, act as new employees that must be onboarded, trained, evaluated, and supervised before operating independently.

Nishith Khandwala is the CEO of Bunkerhill Health, whose Carebricks platform functions as a “system of action” for deploying agents. “Forward-deployed engineers are first-class citizens at Bunkerhill,” he told me. “They partner with doctors, nurses, and staff to turn their ideas into agents, then continually iterate to make them work in practice.”

Still, as the model has proliferated, FDE has become “the most overloaded term in the Valley,” according to Sanjay Saraf, Head of FDE at Luminai.

The role varies substantially depending on the FDE’s technical depth, where they sit in the sales cycle, and how mature their company’s underlying platform is. Some FDEs do little technical work and act more like management consultants, while others are deeply technical and build the critical pieces of the platform.

Doug Proctor held several leadership roles at Palantir before co-founding Candid Health, a revenue cycle automation company where he now serves as COO. He believes that FDE is less a narrowly defined role than a way of working. “FDEs are first and foremost a culture biased towards extreme outcome orientation and an ethos of doing whatever it takes to help customers win.”

Working closely with customers is hardly new. Services and consulting organizations have long done it. What distinguishes FDEs is where they start and what happens next.

FDEs start with a specific customer outcome. As they build and iterate on a bespoke solution, they and their product teams identify which parts may be useful for other problems or other customers. They fold those abstractions back into the underlying platform so they can be reused, making similar problems easier to solve in the future.

Bob McGrew, who helped develop forward-deployed engineering at Palantir, says , “The FDE model is doing things that don’t scale at scale.” At its core, it’s a learning process that couples product discovery, development, and delivery, creating value for the customer while the FDE company’s assets compound .

Reusable assets extend beyond code. FDE organizations also accumulate knowledge and best practices across multiple customers, allowing them to see patterns that a single customer wouldn’t see on their own. This is especially valuable now, as healthcare organizations are starting to figure out how best to use AI agents.

Assort Health uses FDEs to deploy AI agents for referrals, scheduling and intake. Its Co-CEO Jeffrey Liu explained, “Forward-deployed engineering has to make the core product smarter over time. If all it does is grow headcount and perpetuate customizations, it isn't working. That discipline is what separates a customer-centric deployment strategy from an expensive services business.”

FDEs Are Not Always The Answer

The very things that make the FDE model powerful also create tradeoffs: how much customization do customers really want, and at what cost?

FDEs can quickly create and implement custom solutions. Yet because they are highly compensated— at the extreme, some FDEs at frontier labs earn more than $1 million per year—the model is expensive. And like custom-built kitchens, bespoke software can be harder to maintain and leave customers dependent on vendors.

Brendan Keeler (aka Health API Guy), who leads interoperability at HTD Health, shared a useful analogy. If FDEs build knobs, traditional SaaS configuration involves turning knobs that have already been built. Sometimes turning knobs is all that’s needed.

Meanwhile, health systems with the wherewithal can increasingly build and turn their own knobs using newer self-service approaches, including no-code, low-code, and now even vibe coding.

One such option is Notable, which reports its platform runs over a million automated workflows each day, most of them built by customers. Chief Medical Officer Aaron Neinstein told me that compared to FDEs, self-service may reduce costs and vendor dependence, create greater customer ownership, and be more sustainable.

Similarly, Epic’s Agent Factory enables Epic customers to build the agentic workflows they need. Derek De Young, who leads the program, explained that this lets organizations take what Epic is creating and shape it with their own policies, knowledge bases, and protocols. In effect, this creates “internal FDEs” within health systems.

Still, self-service has limits. Bunkerhill’s Nishith Khandwala argues it may not work for complex workflows that cross multiple teams and data systems, have many exceptions, or change over time.

Luminai CEO Kesava Dinakaran believes self-service is not reliable enough for critical workflows. “If a workflow requires ‘four nines’ of reliability,” he argues, “it may be unrealistic to expect someone within an already-busy IT function to ensure that level of performance at scale.”

Keeler adds that “no-code builders have a long and torrid history of underperforming. But AI could fix them, maybe.”

Regardless of who does the building, someone must still manage the implementation. Amir Dan Rubin, CEO of Healthier Capital, who previously led Stanford Health Care and One Medical, emphasizes that project and change management don’t go away. The work still requires “boundary spanning” across the people and teams involved. Without it, UTMB Health’s Chief AI Officer Peter McCaffrey warns of “chaos” and “zombie workflows.”

And software alone won’t overcome more fundamental challenges. Instead of “simply lobbing software over the wall,” Abhinav Kurada, Percepta’s Head of Hospitals, explained how his team is using cognitive modeling and super-forecasting to define core principles for addressing workforce and inpatient throughput challenges at HATCo’s Summa Health. Percepta brings in FDEs closer to the last mile to build workflows, models, and applications on top of its intelligence platform.

The Changing Shape Of Healthcare Software

As Epic argues, FDEs aren’t entirely new. In some respects, FDEs are a rebranding of traditional co-development partnerships.

At the same time, the lines are blurring. As agentic coding lets developers spend more time with end users and make changes faster, traditional software companies are becoming more forward-deployed. Epic recently announced its new Ranger program, which allows customers to embed Epic-hired and -trained staff for long-term assignments. Self-service-oriented companies like Notable employ FDE-like implementation engineers alongside self-service capabilities.

Still, as today’s custom solutions become tomorrow’s configuration, FDEs could paradoxically eliminate the need for some FDE work. In this way, FDEs could accelerate the path back toward SaaS.

Yet FDEs could persist if the frontier keeps expanding outward. Doug Proctor told me, “It depends on whether you’re playing a finite game or an infinite one.” Nishith Khandwala added, “FDEs will create net new jobs to be done and move up Maslow’s hierarchy to solve harder problems.” Sanjay Saraf tells his team, “The job never gets smaller, the ambition only gets bigger.”

Which is a good thing, since healthcare has no shortage of opportunities.

First Round Capital’s Roy Rosin, who previously led innovation at Intuit and Penn Medicine, told me, “Healthcare’s complexity and the lack of consistency across settings and implementations make FDEs a good fit. And, because the outcomes really matter, we have to get it right. ”

Acknowledgments: I thank the following people for discussing this topic with me: Derek De Young, Kesava Kirupa Dinakaran, Colin Durant, Brendan Keeler, Nishith Khandwala, Abhinav Kurada, Jeffrey Liu, Peter McCaffrey, Aaron Neinstein, Doug Proctor, Roy Rosin, Amir Dan Rubin, Sanjay Saraf, Patrick Wingo.