The Complete Guide to the FDE (Forward Deployed Engineer) Model for AI Startups

Bob McGrew, a key AI innovator, draws on his direct experience at Palantir, OpenAI, and elsewhere to cover the origins of the FDE (Forward Deployed Engineer) model, how it's actually applied, and the insights it offers AI startup founders. Starting from Palantir's real-world experience, he vividly explains why this model has become an essential strategy for today's AI agent companies, and what kind of business organizational approach is needed for AI to have real-world impact going forward.

1. Why Does the FDE Model Matter Right Now?

In the AI agent era, there are no established markets or 'ready-made products,' so product discovery requires an enormous amount of effort. Bob explains the context like this:

"There are no pre-existing products stacked up for AI agents. That's why the FDE model is taking off. There's just too much product discovery that still needs to happen."

In fact, as of 2025, countless AI startups in Silicon Valley are putting FDE front and center, and the YC (Y Combinator) startup job board has over 100 open positions with the title "Forward Deployed Engineer." Just three years ago, such positions didn't exist at all.

2. What Is an FDE, and How Does It Work?

What is an FDE? An FDE is an engineer who is embedded at the customer site, directly filling in the technical gaps that the product alone cannot solve for the customer's real problems. Bob explains that this model started at Palantir because they were building software for intelligence agencies (customers whose specific work couldn't be publicly disclosed), which necessitated this approach.

"We were trying to build spy software, but nobody would tell us what spies actually do. So what we did was build a demo, go directly in front of the customer, show it, and ask, 'What don't you like? Want us to change it?' -- and collect their feedback."

This is similar to the classic early-stage startup approach of achieving Product-Market Fit. However, the actual strategy Palantir experienced was slightly different.

Core Structure of the FDE Model

The product doesn't solve just one customer's problem. Each customer had slightly different functional requirements.
As a result, instead of a single fixed product, they expanded into a 'customizable platform' and embedded engineers on-site to tailor the product to each customer's problems.
Palantir's FDE model was close to the philosophy of "ignoring the limits of scale and deliberately repeating the 'inefficient' things you normally wouldn't do."

"The FDE model is a way of doing 'things that don't scale' while actually scaling."

3. How Palantir's FDE Team Actually Operates

(1) Echo Team & Delta Team

Echo Team: Composed of people from the customer's domain (former military, healthcare experts, etc.), responsible for identifying the user's real problems, managing client relationships, and defining demos/use cases
Delta Team: Software engineers responsible for rapidly coding (prototyping) and deploying to the actual customer site

These two teams work organically together:

"The Echo team needs to be heretics. They have domain knowledge but recognize that the existing way of doing things isn't the answer... Delta needs to be engineers who can quickly build things that actually work rather than being perfect."

(2) A Training Ground for Entrepreneurs

The FDE experience is "entrepreneur training itself." This is also the background for why an enormous number of founders have come out of Palantir.

"The experience of working as an FDE is real entrepreneur training. You learn almost every skill a startup founder needs right here."

(3) The Difference Between FDE and 'Consulting'

When questioned whether the FDE model is just customized consulting, Bob honestly admits that "if poorly designed, there is indeed a risk of degenerating into consulting." However, the key point of FDE is that it must simultaneously produce 'results that solve problems' and 'repeatable software evolution.'

"If you're just writing code based on customer requests, that's consulting. But you need to create repeatable value based on the fundamental problems that actually have impact."

4. Product Team, Customization, and the Birth of Palantir's 'Ontology'

The Role of the Product Team

It constantly checks whether customer demands from the field can be 'made repeatable and generalizable for all customers.'
If too many 'specialized features' are added for a single customer's problem, the product just becomes bloated.
At one level of abstraction higher, it evolves into structures (platforms, ontologies, etc.) that can be broadly used across multiple customers.

Example: Palantir Ontology

"Initially it was a database design for specific objects like 'people,' 'money,' and 'ships,' but we couldn't customize like this every time. So the start of the ontology was abstracting everything into modules that could be directly defined and configured on-site."

In this process, tension and conflict between the field (FDE) and headquarters (product team) are a natural phenomenon.

"The product team always wants 'generalization' and 'maintainable code,' but the FDE needs a customer-specific solution right now. The tension between the two was actually the driving force for 'learning' and 'growth.'"

5. Why AI Startups Are Flocking to the FDE Strategy

The uniqueness of Palantir's FDE model was born from the process of repeatedly engaging with diverse government agencies and enterprise customers, each with different requirements. And the situation for today's AI agent startups is completely analogous.

"In the AI agent market, there are virtually no incumbent products. The real work to be done differs for each customer/sector. That's why the field engineer approach has become necessary."

While traditional SaaS generates revenue by 'repeatedly solving the same problem (scaling),' AI startups are fundamentally different in that they need 'new answers (product discovery) for every customer's problem.'

"In five years, AI agents won't be a single concept but will be split into countless sub-markets and workflows. This is exactly why the FDE model is taking off."

6. Practical Tips and Success Criteria for the FDE Strategy

Bob repeatedly emphasizes that not using an FDE strategy is the easier path.

"At first, I tell everyone, 'Don't do it.' If you can avoid it, a standard product strategy is easier. Use FDE only when truly necessary, when you can't avoid it."

(1) Key Principles of the FDE Model

Remember that you're selling "actual problem-solving outcomes" for the customer, not an "installation" or simple software delivery.
Contracts and pricing should also be set not by 'number of seats or usage' but by the problem solved / unit of outcome.
Success can be measured by contract size with customers or the declining unit cost of delivering 'more valuable outcomes.'

(2) Common Failures

Misunderstanding the FDE model and getting stuck in a simple service role or actual consulting
Blindly following on-site customer requests and building software disconnected from the actual organizational impact or core problems

"If FDEs just help customers with their immediate difficulties on-site, what you may actually get is a meaningless product. You need to focus on solving fundamental problems that can have organization-wide impact."

7. Product/Organization Growth, Demo-Driven Development, and the Meaning of FDE

(1) The Unexpected Power of Demo-Driven Development

In both Palantir and AI startup culture, 'polished demo' development was the most realistic way to verify actual customer needs and viscerally feel the product's direction of evolution.

"Every time a feature was added, the key was demoing it to see how it helps existing scenarios and whether the customer actually wants it. This is what creates a product that 'sparks thirst.'"

(2) 'Learning Organizations' and FDE

Bob emphasizes that the FDE organizational environment, unlike established corporations or successful companies, is fundamentally about 'endless learning and trial-and-error.'

"If you go to a company like Google (or Meta) that's already running smoothly, there's not much to learn. Going to a {new, young organization} and experiencing change together is the real stage for growth."

Both Palantir and OpenAI have preserved this learning-organization culture,
And it's precisely the reason why people with FDE experience are best suited for AI startup founding later on.

8. Bob's New Challenge: Applying FDE in the US Army Reserve

Bob has recently been participating as a member of a technology innovation unit in the US Army Reserve. Remarkably, a strategy similar to FDE is being applied there as well.

"We're not just advisors -- we're actually commissioned as officers, discovering problems on the ground, and when necessary, bringing solutions all the way up to senior leadership. ... I'm re-experiencing how the FDE model is a perfect fit when driving organizational change across decades of entrenched habits, bridging frontline missions and leadership above."

9. The Real Opportunity Now Open to AI Startups

(1) Record-Speed AI Advancement, Sluggish Real-World Adoption

Bob uses the recent OpenAI example to argue that while AI capabilities (models/technology) are growing incredibly fast, the speed at which they spread into the real world is much slower and more tedious.

"Over the next five years, AI capabilities will keep skyrocketing, but the 'gaps that need to be filled' before it becomes truly useful and reaches real-world applications will be far greater than people imagine."

(2) What FDE Organizations/Startups Must Do

Bridging the gap between what AI can do and what actual users/organizations want is the biggest opportunity for FDEs and AI-era startup founders.
He likens the entire AI organization to a 'research lab (headquarters),' while FDE startups are the 'field execution units' applying this technology to the real world.

"If OpenAI is the headquarters product team, startups play the role of FDEs rushing out into the field to drive real adoption. This will be the real backdrop for AI innovation going forward."

Conclusion

The FDE (Forward Deployed Engineer) model doesn't just mean specialized customer-tailored development -- it's the only way to apply innovative technology to the real world and create impact in the AI era, and it's an organizational culture that instills 'learning and growth' into both product and business. As AI technology advances explosively, entrepreneurs and engineers who can bridge this gap, along with organizations equipped with the ability to coordinate between field and product, will become increasingly important. The growth experience of learning by directly solving problems on the ground, and the practical capability of truly building customers and markets -- this is the brightest opportunity for today's AI startups.