Building products that actually work requires engineers who live with customers, not Zoom warriors

The Bold Truth About Product Discovery

Forward Deployed Engineers (FDEs) are techies who embed directly with customers—not in Zoom calls, but in their offices, warehouses, or hospitals—to understand real problems and build solutions that actually work. This model, pioneered by Palantir and now spreading across AI startups, is becoming the secret weapon for creating products that deliver outcomes instead of features. If you’re building anything complex, especially AI agents, this might be the only way to win.

What Makes FDEs Different

Most product teams operate like this: customer interviews → specs → build → ship → hope it works. FDEs flip the script. They live with users, breathe their workflows, and prototype in context. Marty Cagan calls this the “product creator” model—engineers, designers, and PMs who don’t just execute requirements but discover solutions.

The Palantir Playbook

Palantir didn’t invent embedding engineers with customers, but they perfected it at scale. Unlike Accenture (which builds what clients ask for), Palantir signs up for outcomes. Their FDEs work in defense agencies, hospitals, and oil rigs—learning deeply complex domains, then building platform capabilities that scale across clients.

The magic? Each FDE builds prototypes using Palantir’s evolving platform. The platform team synthesizes patterns across deployments and abstracts them into reusable services. Rinse, repeat. This is why Palantir is worth $400B while Accenture sits at $150B.

Why It Works Beyond Palantir

You don’t need a billion-dollar platform strategy to benefit. The core insight is simple: engineers learn better by seeing than by reading specs. Send your best devs to customer offices. Let them watch users struggle. The “aha” moments compound fast.

This is especially critical for AI agents today. Startups building agents for legal, healthcare, or finance are all adopting FDE models because you can’t fake domain expertise. You need boots on the ground.

The Original Use Case

Before Palantir, the FDE model was simpler: send engineers to multiple customers to spot patterns. That’s customer discovery 101. But most teams skip it, building in silos, then wondering why adoption tanks.

Here’s What Most People Miss

I’ve seen this playbook before—just under a different name. When I worked with Japanese auto OEMs like Toyota and Honda, they called it “front loading.” Suppliers built highly spec’d bespoke control units before the project was even awarded. The principle? Prove you can solve the hard problem before anyone asks you to.

Today’s FDE model is front loading for the LLM era. We’re moving into anti-SaaS territory—non-deterministic custom software where code is written on the fly, adapting to the changing, interactive nature of problem-solving. You can’t ship a one-size-fits-all product when every customer’s workflow is unique.

Here’s where most people screw up: they underestimate the cost.

Sales pitches are cheap. Building a demo that proves your tech solves complex, messy problems? That’s expensive. Most companies treat FDEs like consultants—billable hours, scope creep, hand-holding. Wrong. FDEs are an investment, not a service. If they do their job well, the solution becomes sticky. The customer can’t leave because you’ve embedded yourself into their critical workflows.

But here’s the trap: FDEs aren’t just developers waiting for instructions. They’re outcome-focused specialists who own the full stack—GPU infra, orchestration, vectorized databases, pre-training, post-training, evals. They’re not there to take orders; they’re there to deliver results. That’s why they’re niche, scarce, and insanely valuable.

Fun fact: Bill Campbell, Steve Jobs’ legendary coach, used to force Apple engineers to work retail shifts in Apple Stores. Why? Because watching real humans fumble with products teaches you more than a thousand Jira tickets ever could.

Question for you: If you had to embed an engineer with your customers tomorrow, what’s the one workflow you’d want them to observe? Drop it below. 👇