A forward deployed engineer (FDE) is a software engineer who works inside a customer's organization, close to the people and systems the software serves, to build, integrate and run it in production. Where most engineers build one product for many customers, a forward deployed engineer builds for one customer at a time, writing real code against real constraints until the thing works there.

What does a forward deployed engineer actually do?

A forward deployed engineer owns delivery at a single customer, end to end: understanding the problem, scoping it, designing the system, writing the code and getting it into production. Palantir, which made the role famous, puts the split neatly in its Dev versus Delta post on engineering roles at Palantir: a product engineer is "one capability, many customers", a forward deployed engineer is "one customer, many capabilities".

The job descriptions now carry the same shape. OpenAI's forward deployed engineer role asks the engineer to "own discovery, technical scoping, system design, build, and production rollout", and also to turn what worked into tools and playbooks other teams can reuse. That second part is easy to miss. A good forward deployed engineer doesn't only ship for one customer. They bring back what the field taught them, so the product and the next deployment get better.

That's also the line between a forward deployed engineer and a software engineer: the software engineer is measured on the product, the forward deployed engineer on one customer's system reaching production. A solutions engineer, by contrast, mostly helps sell and configure it.

Where did forward deployed engineering come from?

Forward deployed engineering was popularized by Palantir, whose platforms needed engineers on site to connect them to each customer's data before they were useful. The term spread when AI companies hit the same wall: models are general, but every enterprise has its own systems, permissions and messy workflows.

The numbers moved fast. In June 2025, Andreessen Horowitz called it the hottest job in startups and counted 22 of 311 open roles at OpenAI as forward deployed or solutions roles. Bloomberry's analysis of 1,000 postings found forward deployed engineer job postings up 1,165% year over year for January to October 2025. And in May 2026, OpenAI launched a Deployment Company that agreed to acquire Tomoro, bringing around 150 forward deployed engineers and deployment specialists in from day one.

I wrote about the AI agent development companies that embed forward deployed engineers. The pattern holds across labs, platforms and large consultancies: the people who wire the system into your company now sell alongside the system.

Does the forward deployed model apply to your company?

The forward deployed model applies to most CTOs in one of three ways, and each one asks a different question of you. Which one you're facing decides whether you're buying the model, building it or borrowing it.

1. A platform partner sends forward deployed engineers to you. You've bought an AI platform or model access, and the partner's engineers come in to make it work on your data. The question is ownership: what's built on their platform, by their people, and who maintains it after the engagement ends.

2. You hire forward deployed engineers to serve your own customers. You sell a complex product to enterprises, and onboarding needs engineering, not a help article. The question is margin. The same a16z piece is candid that this customization "initially results in lower gross margins and higher burn rates", and argues it pays off only if the workflows you learn turn into product.

3. A partner embeds a forward deployed team to take AI into production. You have a pilot, a model or an agent, but not the people to get it running in your own systems. The question is results: what each use case should move, and who owns it once it's live.

The third is what we do. Our forward deployed teams work inside your systems, on your data, one use case at a time. That’s different from our dedicated development teams, which run long-term product builds.

Which roles does a forward deployed team need?

A forward deployed team needs engineers to build and integrate the system, a product manager from day one to choose the use case and the result it should move, and designers when people have to use what's built. One forward deployed engineer can get a system working. A model on its own still isn't something your users can use.

• Forward deployed engineers connect the system to your data and tools, build against real edge cases and get it through your reviews.

• A forward deployed product manager picks the first use case with you, agrees the result it should move and keeps the work pointed at it.

• Forward deployed designers shape the flows and decide where people step in, for the decisions AI shouldn't make alone.

That's the step from engineer to team. It's also how we set ours up: engineers and a product manager from the start, with designers added when the work needs them, one use case at a time.

When does the forward deployed model work, and when doesn't it?

The forward deployed model works when the hard part is getting AI from a pilot into your real systems, data and reviews, and someone will own it once it's live. It fails when the problem is already well understood, or when nobody is accountable for what the system should change.

It tends to work when:

• An AI pilot impressed everyone, then stalled on real data, security reviews and legacy systems.

• Your agents exist but can't reach the systems and data they need to act on.

• You have the model but not the product: nobody has designed where people step in.

• Your engineers own the roadmap, and AI work keeps losing to it.

It tends to fail when:

• The problem is complex but known, and your own team could solve it with the right tools.

• Access to systems and people is blocked, so the engineers end up guessing.

• No one is named as the owner of each system once it's live.

• The work is really a long-running product build, which a dedicated development team fits better than a deployment.

The friction in every version is the same. Getting access, permissions and fallbacks right is most of the work, and it's the part nobody can do from outside your company.

How much does a forward deployed engineer cost?

A forward deployed engineer in the US earns a median total compensation of $220,000 a year on Levels.fyi, with the 25th percentile at $171,000 and the 75th at $290,000. Advertised pay runs lower: postings that disclosed a range had a median of $173,816. In the UK, the median forward deployed engineer earns £140,302 in total compensation, based on a small sample of self-reported salaries.

Those figures are for one person. Hiring makes sense when forward deployed engineering is a permanent function, the second path above: you sell a product that keeps needing engineers at customer sites. When the need is getting specific AI work into production, a full-time hire tends to outlast the job, and one engineer rarely covers product, design and engineering at once. That's where a forward deployed team usually fits better than a hire.

Why the result matters more than the title

The forward deployed engineer became one of the hottest jobs in startups because models alone don't reach production. Someone has to sit inside the company and make them fit. That part is real, and it isn't going away.

But the title is the least interesting part. The better test for a CTO is the result. Agree what each use case should move, whether that's hours saved, cost per request or revenue, measure it against a baseline, and make sure every system in production has a named owner. Pick the model, and the partner, that's willing to be measured that way.

If you have an AI pilot that stalled, or an agent that can't reach the systems it needs, that's where we can help. You can hire forward deployed engineers for production AI from us: tell us what you want running in production, and we'll suggest the right setup.