What is a Forward Deployed Engineer — and why every AI company is hiring for them now
In brief
Anthropic and OpenAI both launched billion-dollar deployment companies in the same week — and both are built around the same type of engineer: someone who moves into a company, builds production AI systems against their actual messy environment, and leaves something that lasts. That engineer has a name now.
Contents
In the same week in early 2026, two of the most powerful AI companies in the world announced something unusual: they weren't just building better models. They were building deployment companies.
Anthropic launched a $1.5B joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman. OpenAI launched "DeployCo" — a $10B operation with McKinsey, Bain, and Capgemini — and acquired Tomoro, a 150-person FDE firm, to staff it immediately.
The question worth asking: if you have the best AI model in the world, why do you need a separate $10B deployment company?
The answer is the last-mile problem. Building Claude or GPT-4o is the hard part of AI research. Getting it to actually change how a 50,000-person pharmaceutical company runs their clinical trial workflows — against their undocumented internal APIs, their 15-year-old ERP, their data scattered across three cloud environments — that's a different problem entirely. It requires an engineer who can operate inside a company, not just consult from outside it.
That engineer is a Forward Deployed Engineer.
The clear definition
A Forward Deployed Engineer embeds directly inside a client company and builds production AI systems against their actual technical environment. Not a pilot. Not a proof of concept. Production systems.
The word "forward deployed" is military in origin — it means stationed in the field, not at headquarters. For software engineers, it means working inside the client's systems, on their infrastructure, against their constraints, with their stakeholders.
Here's what an FDE is not:
| Role | What they do | What they don't do |
|---|---|---|
| Solutions Engineer | Demos software | Writes production code |
| Consultant | Delivers strategy decks | Owns the implementation |
| Sales Engineer | Proves the product works | Stays for the complexity |
| Forward Deployed Engineer | Builds production systems inside the client | All of the above |
The distinction matters because the FDE model is fundamentally different from consulting. A consultant leaves a recommendation. An FDE leaves a working system. A consultant's deliverable is a document. An FDE's deliverable is deployed, running code.
What an FDE actually builds
Pull up the Anthropic FDE job listing and the deliverables are specific:
- MCP servers — Model Context Protocol servers that connect Claude to internal data sources: the CRM, the data warehouse, the internal knowledge base, the legacy ticketing system. FDEs build the plumbing that makes AI actually know what the company knows.
- Sub-agents and agent skills — Not just "use the API." Production agent architectures with tool routing, fallback handling, access control layers. Agents that work in messy real-world conditions.
- Evaluation frameworks — How do you know the AI is actually doing the right thing? FDEs build the eval suites that measure quality, catch regressions, and give stakeholders confidence that the system works.
- Production deployment — Auth, observability, rate limiting, error handling. The full engineering work required to put something in front of real users at scale.
Here's a concrete example. A pharmaceutical company wants AI to help clinical trial managers answer questions about their trial protocols. An FDE's job isn't to write a prompt. It's to:
- Map the existing data sources (trial databases, regulatory documents, internal wikis — probably three different systems that don't talk to each other)
- Build an MCP server that connects Claude to all three with appropriate access controls (clinical trial data has regulatory requirements)
- Design an agent that routes questions to the right data source
- Build an eval framework using real questions from trial managers to measure accuracy
- Deploy it with monitoring so the team knows if output quality degrades
- Hand it off to the client's engineering team with documentation they can actually maintain
That project might take 6–12 weeks. A consultant would have spent weeks 1–4 writing a strategy document. The FDE spent those same weeks building the system.
Where the role comes from
Palantir invented this model. They called their version Forward Deployed Engineers going back over a decade, embedding engineers inside government agencies and defense contractors to build data systems against the most hostile technical environments imaginable — classified networks, legacy databases from the 1980s, bureaucracies that had never shipped software before.
The model worked. Palantir became one of the few enterprise software companies that actually delivered on its promises, because they sent engineers to solve the problem on site rather than shipping software and leaving clients to figure it out.
The rest of the industry is now catching up. EY launched FDE roles in the UK in April 2026. Accenture launched an FDE practice with Microsoft and ServiceNow. Deloitte and Google have deployed similar programs. The model went from "Palantir's unusual strategy" to "the obvious answer to enterprise AI deployment" in roughly 18 months.
The current hiring explosion
The numbers are not subtle. FDE hiring is up 800% since January 2025.
Who's hiring:
The AI labs directly:
- Anthropic (via their $1.5B enterprise AI services joint venture)
- OpenAI (via DeployCo + Tomoro acquisition)
- Scale AI (FDE team supporting enterprise customers)
The Big 4 and consulting firms:
- EY (April 2026, UK launch, expanding globally)
- Accenture (Microsoft + ServiceNow FDE practices)
- Deloitte (AI Center of Excellence, FDE roles)
- Capgemini (part of the OpenAI DeployCo structure)
Enterprise companies hiring internally:
Banks, pharma companies, logistics firms, and government contractors are all building internal FDE functions rather than outsourcing indefinitely. If you're at JPMorgan's AI group, or Pfizer's digital transformation team, you're doing FDE work with a different title.
Specialized firms and job boards:
fwddeploy.com has become the dedicated job board for this category. FDE Academy launched as a training pathway. This is a role category large enough to have its own job board. That's a meaningful signal.
What companies look for
The Anthropic FDE job listing is the clearest public specification for what this role requires. It's worth reading in full if you're considering the path.
The explicit requirements:
- 3+ years in a technical, customer-facing role — not just building, but building in contact with users and stakeholders
- Production experience with LLMs: advanced prompt engineering, agent development, evaluation frameworks, deployment at scale
- Strong Python — plus TypeScript and/or Java
- Experience shipping production applications — not just side projects
The deliverables specified in the listing: MCP servers, sub-agents, agent skills in production workflows.
Two phrases appear in essentially every FDE job description, across every company:
"High agency" — The ability to identify what needs to be done and do it without being told. In a client environment, you often don't have a manager who knows the right answer. You have to figure it out.
"Navigate ambiguity" — Client environments are messy. Requirements are incomplete. Systems are undocumented. Stakeholders disagree. FDEs have to build in conditions where the specs change while you're building.
Compensation
The range is wide because the seniority range is wide:
| Level | Typical Total Comp |
|---|---|
| Entry-level FDE (3–5 years exp) | $180K–$250K |
| Mid-level FDE (5–8 years exp) | $250K–$450K |
| Senior FDE / Staff FDE | $450K–$700K+ |
Why does it pay this well? Because it requires a combination of skills that's genuinely rare:
- Deep technical ability — You have to be able to build production systems quickly in unfamiliar codebases
- Business acumen — You have to understand what the client actually needs, not just what they said
- Communication under pressure — You're in the room with executives. You explain technical decisions in non-technical terms, in real time
- High tolerance for ambiguity — You can't wait for full specs. You have to build while discovering what you're building
Most engineers are strong on #1. Some have #4. Very few have all four at a level that makes the FDE model work. That scarcity drives the compensation.
Is this role right for you?
The honest filter:
Who thrives as an FDE:
- Engineers who find standard product roles too slow and too internally focused
- People who genuinely enjoy figuring out undocumented systems
- Engineers who are as comfortable in a stakeholder meeting as in a codebase
- People who get energy from delivering something real for a real user, not from shipping a PR
Who struggles:
- Engineers who need clear specs before starting (you will not have them)
- People who find client work draining rather than energizing
- Engineers who want to go deep on one codebase for years (FDE assignments rotate)
- People who prefer building greenfield projects to fixing what exists
The FDE model is not for everyone. It's a specific combination of engineering and embedded operator. If the idea of walking into a Fortune 500's data infrastructure and building something production-worthy in 10 weeks sounds exciting, this is the path. If it sounds exhausting, that's useful information too.
How to get there
If you're a developer or recent CS graduate considering this path, the career guide covers the four-part skill stack and the three realistic transition paths in detail.
→ See How to become a Forward Deployed Engineer
If you want to build a portfolio that signals FDE readiness, the portfolio guide covers the five specific projects that hiring managers look for.
→ See The 5 portfolio projects that signal FDE readiness
The bigger picture
The FDE hiring explosion is a consequence of a structural reality: the gap between what AI can do in a research environment and what it actually does inside a real enterprise is enormous. Models are capable. Deployment is hard. The engineer who can close that gap — who can walk into a company, understand their systems, build production AI against their actual environment, and leave something that works — is worth an extraordinary amount.
Anthropic and OpenAI didn't launch billion-dollar deployment companies because they ran out of model improvements to make. They launched them because they understand that the model is not the product. The deployed, working system inside the client is the product. And building that system requires a specific kind of engineer.
That engineer has a name now. The role is here. The demand is real.