The number
$216,250. That's the median US total annual comp for a Forward Deployed Engineer on Levels.fyi, with the 25th percentile at $180K and the 75th at $290K. The Register reported it this week alongside two bigger numbers: AWS's $1 billion Forward Deployed Engineering unit (announced earlier this month) and Anthropic's $100 million commitment to training FDEs. Vendors are now spending real money to manufacture the job, and a $110K spread between the 25th and 75th percentile says where you land is a negotiation, not a market rate.
Source: The Register, Oct 22, 2026 (Levels.fyi data).
What changed this week
The Plank FDE census has unfrozen. Fresh head: 1,262 postings observed Oct 18, down from 1,333 on Sept 12, a 5.3% drop (-71) over roughly five weeks. A full company-level diff isn't possible yet, so treat this as a headline, not a trend. A pullback in total postings while vendors announce billion-dollar FDE units is worth watching: either hiring is concentrating in fewer, bigger employers, or the Sept peak was inflated by reposts.
On our own board, we ran an existence check this weekend: 69 stale listings confirmed gone and marked closed (31 on Greenhouse, 38 on Ashby, including 12 at Databricks). Open roles on fde.directory are now 260. Fewer listings, fewer ghost jobs.
data: Plank FDE census (CC BY 4.0); fde.directory board check, Oct 25, 2026.
New roles
Eight seats. Two are brand new this week; the rest are strong open roles with printed bands we haven't featured yet.
Prime Intellect — Applied Research, Forward-Deployed RL SF / New York / London / Remote · $150K–$300K cash + equity The deliverable here is a post-trained model, not an integration: you build the RL environment and run the experiments. New this week. View role →
Peakflo — Forward Deployed Engineer (India/Remote) Remote (India) · ₹10L–₹13L per year (≈$12K–$15K), as listed A rare India FDE seat that prints a number, aimed at 0.5–2 years of experience at a YC W22 agentic-finance company. View role →
Amazon Web Services — Forward Deployed Engineering Manager Mountain View / SF / New York / Seattle / Boston / Arlington VA + · $185K–$288K base + RSUs Pods of 5–15 FDEs embedded at strategic customers. With the $1B unit announced, this is where the org chart is growing. View role →
Amazon Web Services — Senior Security Engineer, Forward Deployed Engineering Mountain View / Boston / New York / Seattle / Arlington VA / Dallas · $175K–$248K base + RSUs Security gets its own FDE track: threat-modeling and securing production AI systems inside customer environments. View role →
Atlassian — Principal Forward Deployed Engineer San Francisco / Seattle / Remote (US) · $206K–$269K Jira and Confluence's owner runs a Principal-tier FDE org for its AI push. Principal-level base, printed. View role →
Zendesk — Forward Deployed Engineer Remote (US: CA/IL/NY/TX) · $200K–$300K OTE (80/20) CX incumbent embeds FDEs with strategic customers to ship AI Agents and RAG workflows. Mind the 80/20 split: base is the lower number. View role →
Databricks — AI Engineer, FDE (U.S. Federal Sector) Maryland / Virginia / Washington, DC · $182K–$250K Public-sector deployment team at a transparent band. The same band shows up across Databricks' Sr. FDE reqs. View role →
Clearly AI — Forward Deployed Security Engineer Remote (US; Seattle/SF preferred) · $125K–$175K + 0.1–0.5% equity "Forward Deployed" as a title prefix, not a team: a YC security startup with a printed band and real equity. View role →
One insight: the Databricks FDE loop, round by round
Databricks runs the largest FDE org outside Palantir, so its interview is the closest thing the market has to a standard. Per Aced's (formerly Exponent) guide, it's fully remote and takes roughly 4–6 weeks across five stages.
Recruiter screen (~30 min): motivation and how your client-facing work maps to the role. Technical screen (45–60 min): short questions across SQL, data manipulation and AI concepts like RAG. You're expected to be strong in two of three domains (applications, data, AI) and credible in the third. Coding (~60 min, notebook): an easy-to-medium data-manipulation function using dictionaries, strings and lists. Graph and concurrency problems are rare; what's scored is clarifying inputs, edge cases and narrating your reasoning.
Then the two that decide it. Decomposition (~60 min): turn a vague business goal into a system design. The guide's advice is to spend the first 5–15 minutes clarifying goal, scope and constraints, and to tie the design to a measurable KPI and an end user. Leadership: a values conversation on customer obsession, ownership and bias for action.
The pattern: the coding bar is moderate; the screen is for whether you can structure ambiguity out loud. Final calls go to a hiring committee, so positive early feedback doesn't guarantee an offer.
Source: Aced (formerly Exponent), "Databricks Forward Deployed Engineer Interview." The loop evolves; some customer-facing loops add a take-home or presentation panel.
One ask
What company should we break down next? Hit reply with a name and we'll put it in the queue.