It is 10pm and you are building something with AI that nobody asked you to build. A small app, a bot, a script that does one annoying thing for you. You have read that you need to "get into AI" before it gets into you, so you tinker after the kids are down, and some nights it clicks and some nights you wonder if any of it is actually leading anywhere or if you are just keeping busy while the market moves on without you.
Here is the part you have not been told. There is a job that pays you to do almost exactly that. Take a powerful AI model and make it actually work inside one real business, with all that business's mess. It is called a Forward Deployed Engineer. Over the last year it has gone from a title almost nobody used to one of the fastest-growing jobs in tech. And in 2026 it started showing up in Sydney.
Most of what is written about it is either breathless ("the hottest job in tech") or scary ("you need to be an ex-Palantir genius"). Both miss what the role actually is, and whether it is a real option for someone here. So let me give you the version I would give a client across the table.
A Forward Deployed Engineer is an engineer a company sends to sit with one customer and build a working version of its product for that customer's exact problem.
That is the whole idea. "Forward deployed" is an old military phrase. It means you work at the front, where the action is, not back at headquarters. So instead of building one feature for millions of users and never meeting any of them, you go and sit inside one customer's business and build whatever it takes to make the thing work for them.
Palantir, the company that started all this, put it in one clean line. A normal engineer builds one capability for many customers. A forward deployed engineer builds many capabilities for one customer. Same skill, flipped on its head.
The AI labs took that old idea and made it the job of the moment. At a company like OpenAI or Anthropic, a Forward Deployed Engineer goes into a big customer, a bank, a hospital network, a government department, and turns a general AI model into something that does a specific job inside that organisation. The model is the easy part. It already exists. The hard part is everything around it: the customer's old systems, their private data, their rules, the one workflow that would save them millions if you could only get the AI to do it reliably. That last stretch is the job.
If you want the shortest possible definition: a Forward Deployed Engineer is the person who takes AI the last mile, from "this demo is impressive" to "this is running in our business and we trust it."
Because the demos kept failing to become real.
In mid-2025, a study out of MIT looked at about 300 real company AI projects. Around 95% of the organisations got nothing back from them. Not a small return. Zero. The few that worked were not the ones with the best model. Everyone has the same models now. They were the ones who got the thing actually wired into how the business runs.
So a gap opened up. On one side, AI that can do remarkable things in a demo. On the other, companies that cannot get it to do one useful thing on a Tuesday with their own data. Somebody has to stand in that gap and close it. That somebody is a Forward Deployed Engineer.
The investors who fund these companies are blunt about why. One of them, the firm a16z, put it like this: a company buying AI is like your grandma getting a new phone. She wants to use it. She also needs someone to come over and set it up. The labs worked out that selling the phone is not enough. They have to send someone over. So they started hiring people to be that someone, by the thousand.
You can see it in the numbers. On one major job site, postings for this role went from about 640 in April 2025 to over 5,000 a year later. That is not a trend. That is a door opening.
Forget the slideware picture. Here is the real shape of it, in four phases.
Phase one: find the real problem. You sit with the people who do the work and the manager who is stressed about it, and you dig until you find the one bottleneck worth solving. The problem they describe first is rarely the one that matters.
Phase two: get unblocked. This is the part nobody warns you about, and it eats more time than the building. The data you need is locked behind a company login that will not let your tool in. It is scattered across old systems nobody has documented. The connection you need has no instructions. And the security and risk team has to sign off before you touch anything, which in a bank or a government department can take weeks. Half the job is getting past these walls without breaking what is behind them.
Phase three: build it and wire it in. Now you write the actual thing. You pull the customer's data in, point the AI at it, find where it gets things wrong, and build the screen a real person will use. Then you connect it to the systems they already run on, so it lives inside their working day instead of in a demo.
Phase four: hand it over, and feed it back. You teach the customer to use it and to trust it. Then you do the part that makes this a product job and not a consulting one. You go back to your own company and tell them what you had to build by hand, so the next version of the product does it for everyone automatically. Build for one customer, learn, make the product smarter. That loop is the whole point.
The title is hot, which means it is also getting slapped on jobs that are not really it. Here are the three real versions, so you can read a job ad and know what you are looking at.
The Palantir original. The classic one. You embed with a customer, often in government or defence, and build on top of Palantir's existing software. Heavy on getting things into real use, lighter on building the core technology from scratch.
The AI lab version. The one driving the hype. You embed with a big customer and build a working AI solution on top of a frontier model. This is the one most connected to "AI engineering," and the one most relevant if that is where you want to go.
The everyone-else version. Every AI startup and software company has now started using the title too. Some of these are the real thing. Many are a sales or setup role wearing a more exciting name. You tell the difference by one question: does this job actually write real, working code inside the customer's world, or does it mostly run demos and hand off? If it is demos and handoff, it is a sales engineer with a costume on. A true Forward Deployed Engineer builds.
That last distinction matters enough to slow down on. A sales engineer helps win the deal, then leaves. A Forward Deployed Engineer shows up after the deal is signed and stays until the thing is live and working. One sells the meal. The other is in the kitchen until it is on the plate.
And do not confuse it with a consultant. A consultant bills for the hours and hands you a slide deck, or a pile of code you now have to keep alive yourself. A Forward Deployed Engineer works for the software company. Their real motive is to get the product running so well inside your business that you keep paying for it, and to carry what they learn back into the product. Different paycheck, different motive, different result.
I read a stack of these job ads so you do not have to. Strip out the buzzwords and the same short list comes up again and again.
The experience bar runs wide. Palantir will take someone barely a year out of university. The AI labs tend to want five or more years. But notice what almost none of them lead with. A PhD. Years of deep machine learning research. A maths background that makes your head hurt. This is not, at its heart, a research job. It is a building job, pointed at a customer.
I have sat on the hiring side for thirteen years. So let me tell you what actually separates the people who get these roles from the people who look great on paper and never get the call.
Almost every engineer can show one of two things. Either "I can build" or "I can work with customers." Very few can show both, with AI in the middle.
That overlap is the entire job. A brilliant builder who has never sat with a customer reads as a risk: great, but will they survive a room with a frustrated client in it. A polished customer person who cannot actually ship working code reads as the other risk: lovely to deal with, but who is doing the building. The person who gets hired has clearly done both. They took an AI tool the whole way into a real business, with a real person on the other side, and made it work.
Here is the part that stayed with me from all those years reading applications. The people who got passed over were almost never the least capable. They were the ones whose proof did not match the job. They had the certificate from the AI course. They did not have the one story that showed them doing the actual work. And a certificate and a story are not close. One says "I learned about this." The other says "I have already done your job once."
I had a client, a delivery lead, who had spent months getting an AI tool live inside a large, heavily regulated organisation. It was the work a Forward Deployed Engineer does, even though nobody there called it that. They wanted AI to read sensitive case material and point staff to the rules that applied. The catch: in a place like that, one confident wrong answer is not a glitch, it is a liability. A single model was never going to be trusted with it.
So here is what he built. He ran a second AI model whose only job was to check the first one's answer before any person acted on it, and he kept the whole thing on systems the organisation controlled instead of sending sensitive material outside. The second model was not a clever trick. It was the thing that won the room. It gave the risk team a reason to say yes, because now there was a check on the machine instead of blind faith in it.
When he first told me about it, he had buried it. Three lines, near the bottom of his CV, under what he thought of as his "real" work. I made him move it to the top and tell it as what it was: the messy brief, the pushback from the risk team, the second model, the day it went live. That is forward deployed engineering. He just did not know the title existed, and he had nearly hidden the most valuable thing he owned.
I am not going to sell you a role without telling you what it costs.
You are the person in the middle. The customer wants it fixed yesterday. Your own engineers want it built properly so it does not collapse in six months. You stand between the two and take the pressure from both. On a good week that is the best seat in the building. On a bad week it is the reason Forward Deployed Engineers burn out, and plenty do.
There is travel, and it swings a lot by company. Some roles are mostly remote with the odd on-site week. Others, including some at the big labs, can put you on a plane up to half the time. Read that line in the ad before you say yes, not after.
And the work is rarely tidy. You will spend more of your life than you expect chasing access, waiting on a password, sitting through a security sign-off, and rebuilding something because the data was messier than anyone admitted. If you need clean problems and a predictable week, this is not your role. If you would rather own a real outcome than a clean ticket, it might be the best job you ever have.
Honest answer: yes, but it is early, and I am not going to pretend otherwise.
The title is live here. As of mid-2026, real roles have been posted in Sydney and Melbourne. Palantir has had a presence here for years, mostly Sydney and Canberra, much of it government work that needs a security clearance. OpenAI advertised the role in Sydney earlier in 2026. Databricks, Salesforce and the consulting firm Deloitte have all put their own versions out across the major cities. And at least one homegrown Australian startup, the Sydney support-AI company Lorikeet, is hiring for it too.
But let me be straight about the size of it. We are talking about a handful of roles, not a flood. New Zealand has effectively none advertised under this exact name yet. The big local names you would expect, the leading Aussie tech companies and the major banks, are not using this title much yet, even though they are doing plenty of AI work.
So if you are waiting for "Forward Deployed Engineer" jobs to appear by the hundred on Seek before you act, you will be waiting a while, and you will be late. Here is the move instead. The same work is being hired under less obvious names right now: Applied AI Engineer, AI Solutions Architect, AI delivery roles. Same job, calmer label. That is where the local volume actually is today. Target those, and you are building for the bigger title as it arrives.
On money, I will only tell you what is real. In the United States, the long-running version of this role at Palantir sits around 208,000 US dollars a year on the public salary trackers, and the AI labs pay more than that, a lot of it in company shares. In Australia, almost none of the job ads list a salary yet, so anyone quoting you a precise Australian number is guessing. The honest read: treat it like a senior engineer or AI engineer role here, which lands somewhere around 180,000 to 250,000 dollars total package and up, with the lab roles likely higher. When the local market matures, those numbers will sharpen. Right now they are a range, not a promise.
You do not need permission and you do not need a new degree. You need the one thing the job screens for: proof you can take AI the last mile for a real user. Build that, and the title comes to you.
A few moves, in order.
Build one real thing, for one real person who is not you. Not another tutorial. Find someone with an actual problem and get an AI tool genuinely working for them. One finished, used, slightly-messy real project beats ten polished course certificates. That project is your whole pitch.
If you cannot picture one, here are three that count:
Pick one. Build it for the actual person who has the problem. Get it used, even by one team.
Put it where it can be seen. This is the part most people get wrong, and it is the cheapest to fix. Your proof has to be obvious in the first few seconds of your CV and your LinkedIn, in the words these roles are actually searched with. The work being invisible is the same as the work not existing.
Everything above is the plan. The two places it usually breaks are the same two I watch capable people lose on: their proof is invisible to recruiter search, and it is pointed at the wrong roles. That is the gap we built Careersy AI to close. Here is how it helps with this exact move.
None of it invents experience you do not have. It takes the real thing you built and makes it legible to the people deciding. That is the whole game, and it is the part capable people lose on most.
You can run it on your own profile with three free credits and no card.
Start with the project. A real one, for a real person, finished. Then make sure the right people can see it. The title is new, the demand is real, and for once the thing the market wants is something you can actually go and build this month.
A Forward Deployed Engineer is an engineer a company sends to work directly with one customer and build a working version of its product for that customer's specific problem. In AI, it means taking a general AI model and making it actually work inside one real business, with that business's own systems, data and rules. The short version: they take AI the last mile, from an impressive demo to something running and trusted inside a company.
They overlap but they are not the same. An AI engineer is a broad term for someone who builds with AI. A Forward Deployed Engineer is a specific kind of AI engineer who does that work embedded with a customer, owning the job from the first conversation to the day it goes live. If you are moving into AI engineering and you are good with people as well as code, the forward deployed path is one of the most realistic doors in, because it values getting things working in the real world over deep research.
No. Across the job ads, almost none lead with a PhD or deep machine learning research. This is a building job pointed at a customer, not a research job. What they screen for is proof you can build real, working software, sit in front of a customer, and stay calm when things are messy. A strong, finished AI project for a real user is worth more than another certificate.
In the United States, the long-running version of the role at Palantir sits around 208,000 US dollars a year on public salary trackers, and the AI labs pay more, much of it in company shares. In Australia, most job ads do not list a salary yet, so precise local figures are guesses. A fair read is to treat it like a senior engineer or AI engineer role here, roughly 180,000 to 250,000 dollars total package and up, with the AI lab roles likely higher.
Yes, but it is early. As of mid-2026, real roles have been advertised in Sydney and Melbourne by Palantir, OpenAI, Databricks, Salesforce and Deloitte, plus the Sydney startup Lorikeet. It is a handful of roles, not a flood, and New Zealand has very few under this exact name. The same work is also being hired under plainer titles like Applied AI Engineer and AI Solutions Architect, which is where most of the local volume sits right now.
A solutions engineer or sales engineer mostly helps win the deal, with demos and proofs of concept, then hands off once the customer signs. A Forward Deployed Engineer shows up after the deal is signed and writes real, working code inside the customer's systems until the thing is live. One helps sell it. The other builds it. If a job titled "Forward Deployed Engineer" is mostly demos and handoffs, it is really a sales engineering role using a hotter name.
Build one real AI project for a real person who is not you, and finish it. Then make that proof obvious in the first few seconds of your CV and LinkedIn, in the words these roles are searched with. Target the live roles, including the under-the-radar Applied AI titles, and reach the people doing the hiring directly rather than applying cold. The job screens for evidence you can take AI the last mile, so the fastest path is to go and create that evidence.
Eli Gunduz is the founder of Careersy and a current Principal ANZ Tech Recruiter, with 13 years inside Australian and New Zealand tech hiring. He built Careersy after watching capable people get filtered out of jobs they could do, for reasons no one ever explained to them.
Everyone gives you job-search tactics. Nobody warns you what three months of silence does to your head. Three concrete moves, from someone who read the applications, to take back the part the market works hardest to steal: a bit of control.
5 min read
Deep Research that remembers your prep and forgets your stale facts. Careersy AI Lite, an entry point for everyone. And the week's smaller fixes. The Debrief points at the room, not the company.
5 min read
