Hey, it's Cindy 🌱 This is the full starter kit from the reel: what a forward deployed engineer actually is, the honest path in, the practice loop you can run with Claude this week, and the one-page deployment readout that hiring managers actually screen for.
the basics
The job, and why it exploded 🧭
- 🕵 It's not new. Palantir invented the forward deployed engineer more than a decade ago for customers like the CIA who couldn't explain their own requirements, so engineers embedded inside and figured it out. Until around 2016, Palantir employed more FDEs than regular engineers
- 💰 Now everyone wants them. Salesforce committed to building a team of 1,000 FDEs (their own blog calls it "Today's Hottest Role"). AWS just put $1 billion into a new FDE unit, seeded with "thousands" of hires. OpenAI is hiring FDEs right now at $162K to $280K base plus equity. Anthropic's FDE-titled listings filled so fast the role now lives under Applied AI titles
- 📈 The numbers: Indeed listings went from about 643 to over 5,330 in a year, with a median advertised base of $173,816 (Bloomberry's analysis of 1,000 real postings)
- 🧩 Why it exists: MIT found 95% of enterprise AI pilots fail to produce measurable impact, and it's almost never the model, it's the deployment. Deloitte's survey of 3,235 leaders found insufficient worker skills are the top barrier to getting AI into workflows, and 84% of companies haven't redesigned a single job around AI. Companies bought the AI. They can't buy someone who makes it work inside their business
The one-line definition worth memorizing: a consultant leaves you a deck, a solutions engineer leaves when the deal closes, a forward deployed engineer moves in and doesn't leave until the thing works in production.
the honest path
The honest path in 🚪
No sugarcoating: this is an engineering role. 60% of postings want 3 to 5 years experience and only 12% are entry-level, and Python shows up in 66% of them. But here's what the data and the people who built these teams say:
- 🎓 Pedigree is not predictive. First Round's interviews with FDE hiring leaders: Palantir's rosters included recent grads, and 10-year big-tech specialists were sometimes treated as red flags. They hired for fresh perspective, grit, compulsive building, and business curiosity
- 🎭 Half the job is reading the room. An ex-Palantir FDE's memoir: the company handed new FDEs an improv theatre book because the job demands "unusual sensitivity to social context." Their teams included ex-military, ex-intel, even a former SWAT cop as a team lead
- 🤝 47% of postings explicitly require customer-facing ability, the skill most engineers don't have and most marketers, consultants and ops people do
- 📁 Portfolios beat credentials. What screeners scan for is evidence of shipped, documented builds and customer-facing technical work: "It doesn't need to be production scale. It needs to be real, documented, and on GitHub"
the numbers
Real listings, real pay (verified 12 July 2026) 💰
Every number below is printed on a live listing, pulled from the companies' own job boards. Postings expire, so treat this as proof of the market:
- 💵 OpenAI, Forward Deployed Engineer (SF / NYC): $162K to $280K + equity, and their Platform Engineer FDE runs $230K to $385K
- 💵 Google, Forward Deployed Engineer V, GenAI: $262K to $365K + bonus + equity
- 💵 Salesforce, FDE (Mid/Senior), ~27 US locations: up to $316,750 base in CA/NY
- 💵 Sierra, Forward Deployed Infrastructure Engineer: $230K to $390K + equity
- 💵 Scale AI, FDE GenAI: $179K to $224K + equity · Databricks, Sr. FDE: $182K to $250K
- 💵 Handshake, Senior AI FDE: $300K to $375K + equity · Glean, Founding FDE: $160K to $270K · Palantir, Forward Deployed AI Engineer: $135K to $200K + RSUs
The distribution, from the two best datasets: median advertised base is $173,816 with a typical band of $140K to $250K, and at the frontier labs total comp for mid-to-senior FDEs runs roughly $385K to $785K once equity is counted (May 2026 report, 1,200 data points).
🇦🇺 The honest Australia note: OpenAI, Salesforce and Databricks are all hiring FDEs in Sydney (and Databricks in Melbourne) right now, but Australian ads rarely print pay, so there's no reliable AU salary band to quote. The US transparency numbers above are the best available signal.
practice it
The practice loop: do the job before you have it 🔁
This is how you practice being an FDE with Claude this week. The secret is step zero, because "forward deployed" means embedded in someone ELSE'S workflow, not your own:
- 🎤 Discover. Interview the person who owns an annoying process, even 15 minutes. A coworker, a friend with a business, anyone. Map how the work REALLY happens, not the org-chart version. (This is the rare skill: discovery, not coding.)
- 🎯 Pick the one process where the pain is measurable (hours per week, error rate, missed leads)
- 🗺 Plan. Describe the real workflow to Claude and ask for an automation plan with the smallest possible first version
- 🛠 Build scrappy with Claude Code. Working and ugly beats perfect and imaginary
- 📏 Measure. Hours saved per week, before and after. Real numbers only
- 📄 Write the deployment readout (template below) and ship the next one
The discovery prompt to paste after your interview:
🎤 the discovery prompt
I interviewed someone about a workflow they hate. Here are my raw notes: [PASTE NOTES]
First, map the actual workflow as steps, flagging where time is lost and where errors happen. Ask me anything that's unclear before assuming.
Then give me an automation plan: the smallest first version we could ship this week with Claude Code, what it would need access to, and what should stay human. Estimate hours saved per week honestly.
the artifact
The deployment readout (the artifact that gets interviews) 📄
One page, every time. This is what "real, documented, and on GitHub" looks like, and three of these beat any resume line:
📄 the deployment readout template
# [Process name]: deployment readout
THE PROBLEM
Who owns it, what it costs them (hours/week, errors, delays), in their words.
THE WORKFLOW BEFORE
The real steps as they actually happen, including the workarounds nobody admits to.
WHAT I SHIPPED
What the automation does, what it's built with (Claude Code + [tools]), what stayed human and why.
THE RESULT
Before: X hours/week. After: Y minutes. Error rate before/after. Quote from the process owner.
WHAT I'D DO AT SCALE
The next two upgrades, and what would need hardening for production.
learn it free
Free training that actually maps to the role 🎓
All free on Anthropic Academy, with completion certificates:
The links 🔗
📰 Salesforce's own announcement: Today's Hottest Role: Forward Deployed Engineer
📰 AWS's $1B FDE unit: CNBC coverage
💼 OpenAI's live FDE listings: careers search
📊 The best data on the role: Bloomberry's analysis of 1,000 postings
📖 The inside story: Reflections on Palantir
🤖 Claude: claude.com (the practice loop runs on Claude Code, from the Pro plan)
Follow
@cindiezhu for more AI tips every single day 🌱