FDE job listings are up 729% in 12 months.
Salaries hit $785K at frontier labs.
Anthropic and OpenAI raised $11.5B in one week to hire thousands of them.
And almost nobody knows what this job actually is.
Here is the full picture.
The number that created this job
MIT studied 300 enterprise AI deployments.
95% produced no measurable impact on profit and loss.
The models were fine.
The deployments died.
Not because AI didn't work.
Because nobody could make AI talk to a legacy database, pass a compliance review, and survive being handed to the operations team that inherited it.
That gap between what AI can do and what enterprises can actually deploy is worth billions.
Your salary lives in that gap.
What an FDE actually does
Not a consultant. Not a solutions architect. Not a rebrand.
The simplest framing:
An FDE closes the last mile between an AI product and real enterprise value.
Palantir's internal definition nails it:
*"FDE responsibilities look similar to those of a startup CTO: you work in small teams and own end-to-end execution of high-stakes projects."*
You wear three hats simultaneously.
Hat 1: Software engineer
You write real code. On their infrastructure. With their tooling.
Not a prototype on dummy data. Production code in production systems.
Hat 2: Business consultant
You understand their domain, map their processes, translate business pain into technical scope.
Two clarifying questions from you should save three days of engineering.
Hat 3: Product manager
You feed real-world pain back to your company's core product team.
The best FDEs don't just deploy software. They reshape it.
The result: you are a founding CTO embedded inside someone else's company, with the full engineering resources of your own company behind you.
Compensation nobody prints
This is not a research scientist package.
No PhD needed. No published papers. No algorithm puzzles.
Here is the actual ladder:
Entry level: ~$160K
Mid-level: ~$286K (Palantir median: ~$215K)
Senior: $300K base + equity
Staff: $610K total comp
Principal at frontier labs: $1.0M–$1.2M
At Anthropic and OpenAI:
Senior FDE total comp: $785K+
Applied AI Engineer base (senior): $300K+
Why everyone's paying up right now
FDE postings went from 643 in April 2025 to 5,330 in April 2026.
That's 729% in 12 months.
Then one week in June 2026 changed everything:
→ Anthropic announced a $1.5B JV with Blackstone, Goldman, and H&F to deploy AI at enterprise scale → OpenAI announced "The Deployment Company" — a $10B vehicle with TPG — with plans to hire thousands of FDEs
$11.5B raised in one week.
All of it pointed at the same conclusion:
The frontier labs cannot sell AI to enterprises without humans who can close the last mile.
This is that conclusion, denominated in capital.
Meanwhile:
→ 42% of companies now abandon most AI initiatives — up from 17% a year ago (S&P Global) → 40%+ of agentic AI projects projected to be cancelled by end of 2027 (Gartner) → 88% of organizations already use AI regularly (McKinsey)
Near-universal adoption. Near-universal failure to convert it.
Your salary lives in that gap.
3 types of companies hiring FDEs
Not all FDE roles are the same.
Know which bucket you are targeting before you apply.
Tier 1 — Frontier AI labs ($385K mid / $610K staff / $1.2M principal)
Anthropic, OpenAI, Google Cloud.
You work with select high-value customers whose problems push the model's capabilities forward.
Most selective. Best resourced. Highest upside.
Rarely entry level. They hire people who already shipped.
Tier 2 — Applied AI startups (roughly half of Tier 1 comp)
Series A through D, deploying into enterprises.
Same work. Less gatekeeping. Real title on the resume.
Best career growth path for someone who doesn't have frontier lab history yet.
Tier 3 — Fortune 500 AI teams ($150K–$250K)
Most of the postings. Least of the leverage.
Fine as a start. Bad as a destination.
The gap between Tier 1 and Tier 3 is not seniority.
It's the same work priced completely differently.
Skills that actually matter
Here is the counterintuitive part.
The best FDEs are not the deepest engineers at their company.
They are the ones who can hold six domains in their head at once and switch between them without friction.
The technical floor:
→ Python and TypeScript — covers almost everything you touch → One cloud — whichever your target customers actually run → One database you can debug at 2am in a stranger's environment → Enough frontend to ship a usable interface in a day
That is enough. You do not need to be the best engineer in the room.
You need to be the only person in the room who can do all of it.
The AI-native layer (ship it — don't train it):
→ Strong prompt engineering — beyond trial and error → Fluency with major model APIs: streaming, tool use, token budgets → RAG patterns — and knowing when retrieval is the wrong answer → Structured outputs + schema validation → Basic eval discipline — the layer that separates demo from production → One agent framework you've actually built something in
The half nobody practices:
→ Running a discovery conversation with a non-technical stakeholder
→ Saying "we should not build that" to a paying customer
→ Expressing your work in dollars or hours saved
→ Learning an unfamiliar industry well enough to ship inside it in a week
That last category eliminates 60% of technically strong candidates.
More on that in a moment.
3 artifacts you actually ship
Anthropic's own FDE postings name the deliverables directly.
MCP Servers — the integration layer
Connects the model to the customer's actual systems.
Their ticketing. Their warehouse. Their internal API with no documentation and one person who understands it.
Here is what a real one looks like:
Notice what makes this real.
Not the code quality. The decisions inside it.
The audit row exists because their compliance team demands it. The docstring says when to call the tool, not just what it does. The messy schema absorbed here — not dumped on the model.
Agent Skills — encode the customer's process
The model follows their workflow, not a generic one.
Subagents — handle long-running tasks
Stop a multi-step job from collapsing under its own context window.
Build one of each against a real system.
That portfolio beats any certificate.
It proves the thing a coding round cannot test.
The deployment you ship before you apply
Every posting screens for the same phrase.
*"Shipped production AI systems."*
Not studied. Not prototyped. Shipped. To someone who noticed when it broke.
You cannot read your way past this one.
So manufacture the experience deliberately.
→ Find one real workflow belonging to someone who is not you → Sit with them. Watch them work. Time the painful parts. → Build the thing that removes the worst hour of their week → Deploy it where they already work — not in a new app → Stay long enough to fix what breaks in week two
That last part is not optional.
The whole job is what happens after the demo.
Then write it up like an FDE post-mortem:
That document is your interview.
Every hiring manager reading it learns more about you than any resume line.
The round that eliminates 60% of strong engineers
Frontier lab interview loops include a customer conversation stage.
Strong engineers who cleared every technical screen lose here.
The failure mode is predictable.
The candidate hears a problem and starts solving it.
They propose an architecture. Some open an editor.
It feels like competence. It reads as the opposite.
The candidates who advance run it like a research interview.
They do not write code.
73% of frontier lab FDEs report that running discovery conversations was the skill they were least prepared for coming from a traditional software background.
You rehearse this in the deployments from the previous step.
Every time you sit with the person whose workflow you're fixing, you are practicing the round that decides the offer.
Where FDE careers go
The most common exit from FDE is founding a company.
This is not a coincidence.
FDEs spend years watching enterprises fail to solve the same problems — with a front-row seat to the pain and a full technical stack to address it.
That's a founder factory.
Other common paths:
→ Product org lead — field exposure converts directly into product instinct
→ Vertical lead — owning finserv, healthcare, or defense for a large vendor
→ Staff IC — deep technical expert embedded in core product engineering
At Palantir specifically, former FDEs now lead the Foundry and Gotham product orgs.
The path from deployed in the field to running the product is well-worn.
Which starting point is yours
You ship production systems.
Your instincts are the scarce half. Most AI-native candidates have never run anything real.
Add prompting, model APIs, structured outputs, evals. Build one MCP server against something genuinely messy.
You are a live candidate faster than you think.
You come from research.
Overqualified on modeling. Underqualified on everything that matters here.
The job is integration, constraints, and stakeholders.
Build something boring that survives a compliance review and an ops handoff. That is your portfolio.
You want a frontier lab but don't have frontier lab history.
They rarely hire entry level.
Aim at Tier 2 — same work, less gatekeeping. Move up in two years with real deployment history behind you.
You are strong with customers but weaker on code.
The loop is explicitly designed to filter you out.
Your entire preparation is one shipped, maintained, publicly documented system. Do that before applying.
You are strong on code but uncomfortable in rooms.
You will clear four rounds and lose the fifth.
Practice discovery on real people before applying — not after.
You may already be doing this job.
Internal platform engineers embedded with business units run this exact motion under a different title.
Rewrite your experience in the role's language:
→ Workflows changed → Hours saved → Constraints navigated
You are a live candidate today.
My honest take
Capability stopped being the bottleneck two years ago.
What is scarce is the engineer who can walk into a company with legacy systems, a compliance department, and an ops team that already survived two dead AI pilots — and come out six weeks later with something that actually runs.
That combination is strange.
Which is exactly why it prices the way it does.
Breadth over depth. Judgment with no PM to escalate to. The patience to sit inside someone else's mess before writing a line.
95% of pilots produce nothing.
Someone gets paid to be the 5%.
Most engineers will keep preparing for jobs that existed in 2020.
If this was useful:
→ Repost to share it with every engineer watching AI from the sidelines → Follow @sairahul1 for more breakdowns like this → Bookmark this — the discovery round script alone is worth saving
Subscribe to theaibuilders.co for more such interesting articles
I write about AI, building products, and systems that work while you sleep.
