|
Forward Deployed Engineer. So fetch right now.
This title has become a catchall for about six different jobs, and I think most companies posting it know that and don't care. What they want is someone technical enough to write the code and ship the thing, and personable enough to sit in a room with a customer, figure out what that customer is fumbling toward, and go build it without a product manager translating for them.
The bag you're looking for is here
FDE postings on Indeed went from 643 in April 2025 to 5,330 in April 2026. That's a 729% jump year over year. Salesforce publicly committed to hiring a thousand of them.
In the San Francisco Bay Area, salary ranges anywhere from $160K to $300K.
So what is an FDE?
Beats me. But it's not actually a new title. Palantir coined it back in the mid-2000s to describe engineers who embedded directly with intelligence agency customers because those customers couldn't tell you what they needed through a normal sales process. This new wave of FDE is a little different from Palantir's version, but it's not exactly a new job either.
From my view as a recruiter (I'm using this title super loosely), here's what companies actually mean by it.
Btw, Parsity partnered with a recruiting platform, rounds.so, where I have semi-exclusive access to roles in and around tech-centric cities like New York, Austin, and the San Francisco Bay Area, where the majority of these roles are being hired.
The skills they're asking for are basically everything we teach at Parsity (and I don't care if that's a plug — that's honestly the entire reason I made this program. It wasn't just to teach you cool stuff. It's to prepare you for these types of roles).
- Build and deploy production agents (not some chatbot you built on Scrimba. No disrespect to Scrimba)
- Build RAG pipelines that consider retrieval quality, metadata, reranking, chunk size, and which vector store you reach for and why
- Write custom software on top of the AI platform the company sells, inside the customer's environment
- Translate business problems into working code by talking to non-technical people, and potentially sell them on other solutions from your company
Retell AI is a voice AI company hiring aggressively in the Bay Area, and I've used their product extensively (I built an AI voice agent for a small mobile mechanic shop with it). What they want is what I'd call applied AI knowledge.
A chatbot that connects to third-party data, maybe with a human in the loop somewhere. You understand RAG well enough to have opinions about it. Then they want you to be able to embed with their company and deploy their systems into other companies, anywhere from a phone company doing customer service to an enterprise taxi fleet.
You can't hand an enterprise customer a subscription and say good luck.
They're not paying just for the software, they're paying for the implementation, which is notoriously difficult to get right with most of these tools. AI systems are nondeterministic, and they require a type of knowledge and skill that isn't evenly distributed yet. This isn't like asking your engineers to build something with React, which is easy to hire for.
There just aren't that many people who've actually built agents in production, and of the few who have, almost nobody can say they've done it for more than two years. This is where you should be focusing your time and energy right now, while the competition is still this thin.
Bad news first
Of the dozen or so roles I'm looking at right now, zero are remote. Most require travel. And the salaries reflect it: the majority sit well over $200K.
Now the good news. The years of experience required is lower than you'd think. The technical bar isn't low, but it's not prohibitively high either. Anywhere from one to five years of experience is enough for most of these. Companies aren't looking for the most technical person in the room. It's a tradeoff between being technical, being personable, and having actual applied AI knowledge. That tradeoff is your way in.
According to my YouTube comments, everything I just described is dead simple.
Anybody could do it.
They built a chatbot last weekend.
How foolish of me not to consider that. I guess it's strange, then, that the finder's fee alone on these roles runs up to $40,000, stacked on top of a base salary north of $200,000. That's a lot of money to spend filling roles that are supposedly this easy, and some of them get almost no qualified candidates after weeks of searching.
How to get one
Postings run 3 to 5 years at established platforms, 5+ at the frontier labs, with flexibility for people coming from data engineering. Nobody cares about your degree for the most part. The one consistent hard gate is having shipped something to production and being able to talk about what broke.
Then build the primitives: RAG over vector databases and SQL, an agent that calls tools, evals, observability. This is not rocket science. If you're a full stack developer you're most of the way there already.
The last part is the one everybody skips. Practice explaining it out loud to someone who doesn't code. The most-listed skills in 2026 FDE postings are no longer Python and SQL, they're customer discovery and problem decomposition.
What you can do right now
Honestly, I think the AI bubble bursts at some point. I also think the technology sticks around, agents and workflow automation keep spreading into every industry outside of tech, and the demand for people who can do both sides of this job outlasts whatever we end up calling it.
So here's what you actually do about it.
Pick a free model, or put five dollars into an OpenAI or Anthropic API, that's plenty.
Read Anthropic's paper on "Building Effective Agents" to understand common patterns for building agents/workflows.
Then build something real this weekend: an agent in your CI/CD pipeline that reviews a PR diff and reports back a risk level or one that can translate natural text to a SQL query, or a voice agent on a platform like Retell AI that pulls someone's public info from LinkedIn or a company page and calls them with a sales pitch on voice AI (for legal reasons, don't actually deploy this | 
