Forward Deployed Engineer: Why AI Companies Pay Up to $325k to Embed Models Into Real BusinessesArticle · digest for November 21
Editorial scene of a field AI engineer working at the intersection of client workflows and system architecture
Article5 sections≈ 8 min

Forward Deployed Engineer: Why AI Companies Pay Up to $325k to Embed Models Into Real Businesses

According to MIT Project NANDA's report The GenAI Divide (Fortune), roughly 95% of corporate generative AI pilots deliver zero measurable financial return. Companies happily pay for APIs and build slick slide decks, yet their profit-and-loss statements show nothing. Against that backdrop, Forward Deployed Engineer (FDE) job postings jumped more than 10x in a single year, with base salaries at OpenAI and Anthropic reaching $325,000 before equity and bonuses.

Let's unpack why AI labs write those checks, why most programmers walk away from the role, and how people actually land it.

Where the field engineer boom came from

Models themselves are no longer scarce—anyone can buy access in five minutes. The bottleneck hits next, when a company tries to wire AI into messy live operations with a dozen disconnected databases, an aging CRM, and unwritten team rules. The industry calls that gap between buying software and seeing real value the deployment gap.

Palantir battle-tested the fix years ago. They realized quickly that handing a corporation a login and wishing them luck is a sure way to lose the account next quarter. Instead, Palantir sent engineers directly to the customer site to map real workflows and ship production systems.

In 2025–2026, top AI players adopted the playbook. Colin Jarvis launched OpenAI's FDE group in early 2025, delivering deployments for Morgan Stanley, Klarna, and Coca-Cola. Anthropic followed with its Applied AI team, alongside Databricks, Scale AI, Cursor, Cohere, Cognition, and Google Cloud.

Kevin Bai—who trained at Palantir, built the FDE function from scratch at Rippling, and later joined Anthropic Applied AI—summed up the job in one line: "An FDE is just a customer-facing developer. The client isn't buying software or your hours—they're buying a concrete outcome."

Creator Nate Herk's own story (YouTube video) illustrates this well. He worked as a BI analyst at Goldman Sachs, hit a wall of bureaucracy, quit, and started an AI agency for small businesses like real estate brokerages and HVAC companies. He quickly found it was FDE work in miniature: clients show up saying "build me what I saw in that video," while the real job is setting their idea aside and finding where the business actually bleeds money.

Anthropic careers page listing Forward Deployed Engineer openings on the Applied AI team (Source: anthropic.com)

What 200 real job listings actually ask for

Nate Herk scraped roughly 200 live FDE listings across 24 AI companies. Databricks, Palantir, and OpenAI posted the most openings, followed by Scale AI, Cursor, Anthropic, Cohere, and Cognition.

Pay looks strong across the sample: typical base salaries sit between $180,000 and $275,000, with median total compensation around $230,000. Senior roles at Anthropic and OpenAI hit $320,000–$325,000 in base pay alone, and top OpenAI bands pass $400,000 before stock. The surprise sits inside the requirements: people skills outweigh any programming language.

Requirements across 200 Forward Deployed Engineer job listings (Nate Herk analysis)

At the top sits discovery—pulling the real problem out of human conversations, locking down project scope, and managing executives. The technical stack—Python, TypeScript, stitching external systems via APIs, and building multi-step tool-calling agents—follows close behind. Evals (automated AI quality tests) stand out too: companies are tired of flashy demos and demand measurable reliability. Meanwhile, a college degree barely acts as a filter anymore—Anthropic actively recruits former technical founders.

Let's clear up one myth: this is not a "no-code" entry path. Tools like Claude Code and Codex speed up coding, yet top labs still expect strong engineering fundamentals. Experience bars are high as well: most firms ask for five-plus years, and only about one in ten listings considers candidates under three years. That's why most people transfer sideways—developers bored of tickets, consultants or analysts who learned production engineering, or founders of small automation shops.

The flip side of the paycheck: why 9 in 10 engineers say no

If the job pays up to a third of a million dollars, why isn't there a line around the block? Everything comes down to lifestyle.

Roughly two-thirds of the listings in Herk's dataset require travel—25% to 50% of your working hours. OpenAI's Colin Jarvis warns newcomers upfront: expect to spend about half your schedule inside client offices. Instead of quiet sprints and clean code, you deal with flights, messy customer databases, and executives demanding ROI by quarter's end.

Recruiter Patrick Kellenberger of Betts Recruiting put it bluntly for The Pragmatic Engineer: "Everyone wants to hire them, yet barely 10% of engineers actually want the job." For anyone who doesn't mind live negotiations and operational chaos, that filter works in your favor—competition is far lower than in standard backend roles.

How it differs from 'prompt engineer'—and what a week looks like

It's fair to ask: will this title vanish in a year like "prompt engineer" did in 2023? Unlikely. Prompting dissolved into basic literacy as models got smarter. An FDE owns the whole loop—from auditing a workflow to measuring business impact and feeding insights back to product teams. Anthropic's Applied AI team explicitly lists the artifacts you ship: MCP servers, subagents, reusable skills, and golden test sets.

Based on job descriptions and practitioner stories, a typical week looks like this:

DaysWhat happensMain output
MondayTravel to client, discovery interviews with ops teamReal bottleneck identified, not slide-deck guesses
TuesdayOne-page scope documentGoal, single success metric, definition of done, boundaries
Wednesday & ThursdayCRM and database integrations, agent build, evals runsWorking system on live data with measured accuracy
FridayExecutive demo and notes to internal product teamSigned-off milestone, model failure log, reusable template

Friday's wrap-up often matters more than the demo itself: Anthropic expects FDEs to turn one-off fixes into shared templates. Solve a hard problem manually once, package it into a template or tight automation, and save your whole team weeks.

Five steps to break into the role

Nate Herk maps out entry into the role in five practical steps.

First, honestly pinpoint your weaker half. The job stands on two legs: reliable engineering and client trust. Developers usually need practice in negotiation and cutting scope creep; consultants and analysts need production coding chops. Don't train everything equally—attack your biggest gap.

Second, bet heavily on evals in your tech stack. Plenty of people pick up APIs and agents fast, yet almost everyone riding the "vibe coding" wave skips formal tests. On a first pass, an agent usually hits 60–70% accuracy—impressive in a short demo, broken in production. Getting to a reliable 90–95% takes automated checks on real data: demos prove nothing, only evals do.

Third, drill discovery interviews and one-page scope docs. Join client calls and demos at your current job, and practice roleplay drills with an AI at home. After each session, write one page capturing the problem in the client's words, one target metric, the definition of done, and everything out of scope. If you can't name the number you'll move, you don't have a project yet.

Prompt to practice discovery interviews with Claude

"Act as VP of Operations at a 200-person logistics company. Three months ago you bought an AI platform, and it isn't working. I am a Forward Deployed Engineer in our first meeting. Don't give everything away at once—make me ask the right questions to uncover the real problem. After 15 minutes, grade my interview and tell me what I missed."

Fourth, build one end-to-end case study instead of ten toy bots. Take a real bottleneck in your department, a friend's business, or a local company: run the interview, write the one-page scope, build the system, wrap it in tests, demo it, and measure two weeks later whether people actually use it. A short before-and-after breakdown becomes your strongest interview asset.

Fifth, search beyond a single job title and prep for the business case. Check Greenhouse, Ashby, Lever, company boards like Google Cloud, and VC talent networks from a16z, Accel, and General Catalyst. The same job often hides under Deployed Engineer, Forward Deployed Software Engineer, Deployment Strategist, or AI Consultant. Founding FDE roles at startups are another great entry point—Kevin Bai joined Rippling as its first field engineer, and the team grew to 25 within a year. In interviews, alongside system design and AI-assisted debugging (a round Google uses), expect a tough business case: Anthropic specifically tests whether you can walk into a room with only a business problem and walk out with a clear, defensible technical plan.

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 3