Forward Deployed Engineers: AI's Hottest Role

Every few months, a new job title tears through developer communities like it's the answer to everything. The communities I follow — the LangChain Discord, the LlamaIndex GitHub discussions, the Hugging Face forums where people are actually debugging RAG pipelines at midnight — they've been talking about forward deployed engineers for a while now. But lately the LinkedIn layer has caught up, which means the hype-to-signal ratio has gone sideways.

So let's actually look at this thing.

Marina Wyss, a senior applied scientist at Twitch/Amazon who coaches people into AI and ML roles, put out a breakdown of the FDE role this week that's worth engaging with seriously. Not because it's perfect — she's making a career content video, not an investigative piece — but because it's honest in ways that a lot of FDE discourse isn't. She leads with the question most creators bury: is this actually good for you specifically, or is it just good content?

What an FDE Actually Does

The clearest framing of the role comes from Palantir, which according to their own published breakdown of engineering roles is credited with popularizing the title. Their definition cuts through a lot of noise: a normal product engineer works on one capability for many customers. A forward deployed engineer works on many capabilities for one customer.

That inversion matters more than it sounds. The FDE isn't a consultant who hands over a report and flies home. They write production code on the customer's actual systems, stay with the deployment long-term, and often push fixes back into the core product. At OpenAI, Wyss explains, a typical customer engagement runs through three phases — scoping (a few days on-site, mapping workflows, prototyping with synthetic data), validation (building out evals, the quality checks that determine whether the AI is actually doing what it's supposed to), and delivery (the real build, the demo, the handoff).

The time split Wyss describes is revealing: roughly 25% writing code, 50% integration and plumbing work, 25% in meetings. That middle category — the plumbing — is the whole job, really. Getting a model to perform coherently in a lab environment is a solved problem. Getting it to perform coherently against a 15-year-old enterprise database with inconsistent schemas, three different authentication systems, and a compliance team that wants to see every prompt before it touches production — that's what FDEs actually do all day.

This is also, incidentally, why the role exists at all right now. As Wyss frames it, the AI models themselves have become interchangeable. Every major enterprise buyer can access GPT-4 class capabilities. What they can't do is implement them. The companies winning in this moment are the ones absorbing that implementation burden so customers don't have to.

Databricks launched a formal FDE organization in 2026. OpenAI stood up something called the OpenAI Deployment Company. Anthropic announced a parallel enterprise services arm. The hiring surge is real. But Wyss flags something important that most coverage skips: these new deployment orgs at OpenAI and Anthropic are run by external companies. The engineers working inside them aren't core employees of the labs. The equity they receive isn't equity in OpenAI or Anthropic.

That distinction quietly deflates a significant portion of the compensation conversation.

The Money, Without the Fantasy

On levels.fyi, Palantir's forward deployed software engineer role runs between $170,000 and $300,000 annually, with a median around $230,000. That's a real number for a real job.

Perspective AI's 2026 Forward Deployed Engineering Compensation Report, which aggregated roughly 1,200 data points, puts median total comp higher — around $385K at mid-level, $610K at staff, over $1 million at principal for frontier lab positions. Wyss is right to attach a large asterisk to those figures. Most of that compensation is private equity, and private equity is a promise, not a paycheck. For the engineers being routed through external deployment companies rather than the labs themselves, the equity story gets murkier still.

The honest read: FDE roles pay well. Probably very well if you're senior and at the right company at the right moment. The million-dollar number is technically not a lie, but it describes a narrow slice of outcomes weighted heavily toward illiquid startup paper.

The Open-Source Blindspot

Here's where I want to add something that Wyss's video doesn't address, because it's the thing that actually interests me about this role.

The tooling FDEs rely on — LangChain, LlamaIndex, LiteLLM, the whole agentic scaffolding stack — is almost entirely open source. That's not an accident. That infrastructure was built by communities, maintained largely by volunteers and underfunded startups, documented through thousands of GitHub issues and Stack Overflow threads and Discord threads where someone at 1am figured out why their retrieval pipeline was hallucinating and wrote it up for everyone.

The FDE role industrializes that knowledge. An experienced FDE accumulates deep, hard-won expertise in deploying these systems at enterprise scale — the edge cases, the failure modes, the evaluation tricks that actually work when a customer's data is messy in ways the docs don't anticipate. And that expertise lives inside enterprise contracts. It doesn't flow back to the communities that built the tools.

I'm not saying this is malicious. It's just how consulting economies work. But there's a governance question worth asking: does the FDE boom accelerate a privatization of implementation knowledge that used to circulate publicly? The GitHub issues that used to get filed — "here's what broke and here's the fix" — are now, increasingly, the proprietary playbooks of deployment firms. That's a sustainability question for the open-source projects the whole industry depends on.

Who This Job Is Actually For

Wyss's take on entry-level fit is probably the most useful part of her video, and it's the part that's most at odds with how FDE gets marketed on career content channels. Palantir does hire new grad FDEs — that's documented and real. But the role generally expects prior engineering experience, and Wyss is direct about why: the job is sink-or-swim by design. Limited mentorship, undefined specs, customers who will ask for things that make no sense, and production systems that page you at 2am when something breaks.

That's not a bad job. For the right person — self-directed, comfortable with ambiguity, energized by watching something they built get used by actual humans — it could be a great one. But it's a genuinely different psychological profile than most early-career engineers are selecting for when they picture themselves doing AI work.

The technical requirements are substantial but learnable: Python (showing up in roughly two-thirds of FDE postings), TypeScript for the frontend work that inevitably materializes, solid cloud platform knowledge (AWS or GCP), containerization, and the full AI engineering stack — agents, RAG, model API integration, and especially evaluation. Wyss notes that newer postings are starting to name MCP, sub-agents, and agentic coding tools explicitly, which tracks with how fast that space has moved.

But the harder skill set is the interpersonal one. Translating between engineers and non-technical stakeholders. Pushing back on bad requirements diplomatically. Finding the business logic inside a customer's vague ask. Wyss's framing here is accurate: it's easier to teach a strong engineer the customer side than to teach someone with great customer instincts how to write production code. The role generally flows from engineering, not toward it.

Her actual recommendation — take the best pure engineering job you can get, do customer-facing and integration work within that role to build the instincts, then move into FDE with experience behind you — is probably right for most people. It's also, notably, not what gets engagement on career content. "Get a normal engineering job first" doesn't have the same ring as "hottest job in tech."

That gap between the honest answer and the shareable answer is worth sitting with. The FDE role is real, the demand behind it is real, and for some people at some career stages it's genuinely the right move. But the version of this story that's been circulating is optimized for clicks, not for the actual distribution of people who should be pursuing it.

Wyss, to her credit, says exactly that by the end of her video. She just has to get through nine minutes of setup first.

Dev Kapoor covers open source software, developer communities, and the politics of code for Buzzrag.