Three weeks ago, "Forward Deployed Engineer" was Palantir jargon that nobody outside enterprise sales had heard. Today my LinkedIn feed is full of engineers updating their headlines to it. Dan Shipper made it prediction #8 on Lenny's podcast. Anthropic, OpenAI, and Google have all posted for it. The role is the closest thing tech has to a unicorn job right now.
About half the people reading our blog are PMs. I want to talk to you directly, because there is a version of this story circulating everywhere that I think will send you down the wrong path.
The story you are hearing: FDE is the new dream job, so go learn LangGraph and ship a RAG pipeline this weekend.
What I actually think: the frontier-lab version of the forward deployed engineer role is engineer-shaped and you will lose that fight. But the FDE role inside every other company on earth is PM-shaped, and almost nobody is positioning for it yet.
Let me show you what I mean with the actual job postings.
Key Takeaways
- The frontier-lab FDE roles (Anthropic, OpenAI) are engineer-shaped and pay up to $300K, but there are only ~500 globally, and PMs competing there will lose.
- Across 1,000 FDE job postings, 55% of responsibilities focus on customer work, not engineering architecture, making the role structurally PM-shaped.
- The real FDE wave is inside every company deploying AI: roughly 80% of the work is product thinking (workflow redesign, rollout sequencing, v1 scoping) and only 20% is engineering.
- MCP servers, sub-agents, and agent skills (the three deliverable types listed in Anthropic job postings) require clear workflow thinking, not Python mastery, and PMs can ship all three today.
- Enterprise AI consulting engagements cost $150K to $500K for a single deployment; PMs who own these skills internally become a build-vs-buy decision, not a training line item.
Learn this hands-on
Ready to ship a real production app, not just pick a model? Check out the Master Course: Build and Ship a Production-Ready App with Lovable and Cursor.
What Is a Forward Deployed Engineer?
The role has Palantir roots. At Palantir, Forward Deployed Engineers were the people who lived inside customer deployments, building and customizing the product to fit the client's specific operational reality. Part consultant, part engineer, part product manager. The label fell out of common use for a few years, then AI brought it back at scale.
Forward deployed engineer job postings on Indeed grew from 643 in April 2025 to 5,330 in April 2026, representing a 729% year-over-year increase, according to PYMNTS data compiled in May 2026. Bloomberry's analysis confirms 1,165% year-over-year growth from January to October 2025. The role is structurally new, and companies are scrambling to hire for it before they fully understand what it is.
That ambiguity is your opportunity.
What the Frontier Labs Actually Want
Here is the Anthropic Forward Deployed Engineer posting, paying $200K to $300K. The listed responsibilities are:
"Work within customer systems to build production applications with Claude models" "Deliver technical artifacts for customers like MCP servers, sub-agents, and agent skills" "Provide white glove deployment support for Anthropic products in enterprise environments"
The requirements:
"3+ years in technical, customer-facing roles" "Production LLM experience including prompt engineering, agents, evaluation frameworks, and deployment" "Strong Python proficiency plus additional languages (TypeScript, Java, etc.)"
OpenAI's New York posting at $160K to $280K is the same shape: own end-to-end deployment at strategic customers, build RAG pipelines, ship agents with LangGraph, build eval suites and observability. Travel up to 50%.
I want to be honest with you: if you are a PM with thirty hours of Claude Code experience, you are not landing this role next month. You will be competing against senior backend engineers who left Stripe to do customer-side AI work. They will win.
That is the bad news. Here is the good news. Those are maybe 500 roles globally. The actual FDE wave is downstream, and the bar is completely different.
The Forward Deployed Engineer Salary Picture Is Misleading
When people search "forward deployed engineer salary," they find the Anthropic and OpenAI numbers and assume the whole category looks like that.
It does not.
Bloomberry's analysis of 1,000 FDE postings found a median forward deployed engineer salary of $173,816 across all roles. That is excellent compensation, but it is not the $300K ceiling. And critically, those roles have a very different technical bar than the frontier-lab postings.
The breakdown of forward deployed engineer responsibilities across those 1,000 postings tells a more interesting story:
- 55% of the job is working directly with customers
- 37% is building and deploying AI systems
- 32% is integrating systems and APIs
- Revenue quotas and deal-closure metrics: notably absent. Zero percent of postings mention sales quotas.
Read that again. The most cited responsibility, by a wide margin, is customer work. Not RAG architecture, not LangGraph internals. Customer work.
This is not an engineering role that happens to involve customers. It is a customer-facing role that happens to require enough engineering to ship the integration. That is a crucial distinction.
Why the Downstream FDE Wave Is PM-Shaped
Look at the real production examples. BBVA scaled an AI deployment to 120,000 employees across 25 countries. John Deere built AI-powered planting recommendations that cut chemical usage by 70%. These are the case studies the frontier labs use in their FDE pitch decks.
Now look at what those projects actually required. About 20% was engineering: the agents, the prompts, the integrations. About 80% was everything else: which department to roll out to first, which workflows to redesign, which signals to trust, which trade-offs the customer would accept, which features to ship in v1 versus v2.
That 80% is product work. It is exactly what a PM does every day.
The only missing piece is "and now you ship the code yourself" instead of "and now you write a spec for the engineers."
This same tension is reshaping how companies think about the product engineer role: the hybrid who owns the full workflow from customer insight to deployed code. FDE is the consulting-flavored version of the same underlying shift.
Claude Code and Cursor compress that missing piece from a two-year computer science detour into a few focused weekends. That is the unlock.
As Dan Shipper put it on Lenny's Podcast in May 2026, "Any PM who gets really AI-native will be incredibly dangerous because the building is done for you — what matters is figuring out what to build and if it's great."
Any PM who gets really AI-native will be incredibly dangerous because the building is done for you — what matters is figuring out what to build and if it's great.
Forward Deployed Engineer Responsibilities Any PM Can Own Today
The Anthropic posting lists three deliverable types worth pulling apart: MCP servers, sub-agents, and agent skills.
These sound intimidating until you see what they actually are in practice.
An MCP server is a connector that gives an AI agent access to a tool or data source. PMs who have set up a Notion MCP or a GitHub MCP for Claude Code have already built one. It is a configuration file with some glue code, not a systems engineering project. If you want to go deeper, our guide on building a custom MCP to encode your expertise walks through the whole process from scratch.
A sub-agent is a markdown file that describes a specialized task and hands it off to a focused AI process. The signal-capture agent, the release-package agent, the competitor digest runner, all of these are sub-agents. PMs who have followed our AI tools for product managers agent stack guide have shipped several.
Agent skills are reusable prompt blocks or Claude Code slash commands. If you have written a custom /prd-draft command for Claude Code that automatically pulls in your design system context and formats output to your team's spec, you have shipped an agent skill.
None of these require Python mastery. They require clear thinking about workflows, customer needs, and what a v1 that delivers real value looks like. That is PM territory.
The fastest way to get hands-on with all three is the How to Master Claude Code series, which covers MCP setup, custom commands, and sub-agent configuration in a single structured track.
Three Moves to Become Your Company's In-House FDE
Stop trying to land the Anthropic version of this job. Become the FDE your current employer does not yet know they need.
Move 1: Ship one internal AI workflow this quarter.
Not a feature in the product. An internal tool. A sub-agent that triages support tickets. A custom Claude Code command that turns meeting transcripts into PRD drafts. A workflow that runs nightly and posts a competitor digest in Slack. Something your team will use on Monday morning. This is the BBVA-shaped work, miniaturized.
Move 2: Document the rollout.
Who used it, what changed, what broke, what you would do differently. This is exactly what FDEs do at frontier labs, and almost nobody at your company is doing it yet. The artifact becomes the proof of your capability.
Move 3: Repeat, and ride the models.
Every time a new model drops, test it against your workflow. Claude Sonnet 4 versus the previous version on your support-triage agent. The advantage compounds fast, and you accumulate a track record that no engineer-turned-FDE can replicate because they were not thinking in PM terms while building.
Do this for two quarters and your title will catch up to your work, whether your employer calls it FDE, AI Product Lead, or Head of Internal AI. The label matters less than the position you will be in.
The Frame Shift That Changes Everything
I have been rewriting how I think about team training in light of the FDE wave. The old framing was "your PMs learn to code." The new framing is: your PMs become the in-house forward deployed engineers your company will otherwise be paying outside consultants for in 18 months.
That lands differently in a budget conversation. It is not a training line item. It is a build-vs-buy decision.
The consultants are already showing up. OpenAI launched a $4 billion deployment venture in early 2026, backed by TPG, Goldman Sachs, and McKinsey. Anthropic announced a joint venture with Blackstone and Goldman Sachs for exactly the same mission. These are not philanthropic efforts. They are expensive, structured services that will invoice your company to do the work your PMs could own if they had the right skills.
The cost difference is not subtle. A six-week engagement with an enterprise AI consulting firm will run $150K to $500K for one deployment. The Claude Code for PMs cohort at vibecodingacademy.ai covers the same skills across 3 live sessions over two weeks, and your PMs keep those skills permanently.
What This Means for Your Career Right Now
If you are a PM and you have been watching the FDE conversation from the sidelines, this is the moment to step in.
The path is not to compete with engineers on their terms. It is to position yourself as the person inside your company who can do the 80% that engineers are not trained for, while also being capable enough to ship the 20% that used to block you from owning the whole workflow.
That combination does not exist in large numbers yet. Which means it is still an early-mover position.
Open Claude Code tonight. If you have never used it before, the Claude Code tutorial for product builders is the fastest way to go from zero to shipping a real workflow. Pick the smallest internal workflow on your plate. Ship the v1 by Sunday.
The BBVA-scale deployment started with someone shipping a v1 for one team.
