A forward deployed engineer builds software inside a customer's organisation, on the customer's data and systems, until the thing works for them, then hands it over running. It is a software engineering job with the customer in the room: part engineer, part consultant, part support lead. In the 29 forward deployed engineer postings we counted in September 2026, 76% asked for customer or product sense, against 28% of AI job ads overall. That one number is the job.
The title comes from Palantir, which sent engineers to sit with its government and enterprise customers rather than ship a product and wait for tickets. AI companies adopted it because their products have the same problem: a model is easy to demo and hard to make work on one company's messy records, permissions and workflows. Someone has to go and do that part.
What does a forward deployed engineer do day to day?
Four things, in roughly this order on any engagement:
- Discovery. Sit with the people who have the problem, find the one worth solving, and work out what data exists, who owns it and what the security team will allow. The output is a written scope, not code.
- Getting the data out. Most of the first weeks go here: extracting from a CRM, a warehouse, a document store or a folder of PDFs, cleaning it, and building a pipeline that reruns without breaking.
- Building the workflow. An AI system that does the customer's task on the customer's records, with review points where a person can check and override it, deployed where they can use it.
- Proving it and handing it over. Evals on the customer's own cases, a results report a sceptical stakeholder will believe, deployment inside their cloud with monitoring, and documentation their team can run without you.
Then you move to the next customer. The rhythm is weeks, not quarters, and the measure is whether the customer's team is still using the system after you leave.
What do forward deployed engineer job ads ask for?
We re-ran our job-ad sample on 29 September 2026: 2,610 public postings from Hacker News "Who is hiring?" (July to September 2026), Remotive and Arbeitnow. 29 were forward deployed engineer roles, 16 from US and UK startups on Hacker News and 13 from European employers. Among those 29:
| What the ad mentions | FDE ads (29) | All AI ads (315, 19 Sep) |
|---|---|---|
| Customers or product sense | 76% | 28% |
| Production or shipped systems | 62% | 39% |
| Python | 48% | 38% |
| Agents or agentic systems | 48% | 39% |
| System design or architecture | 38% | 23% |
| Evals | 31% | 23% |
| TypeScript or JavaScript | 24% | 19% |
| A portfolio, GitHub or side projects | 24% | 17% |
| A degree | 17% | 12% |
Two things stand out. The customer is named in three ads out of four, and the ads ask for shipped systems more often than for any language. And the role is more on-site than AI engineering generally: 14 of the 29 said on-site, 12 remote and 9 hybrid (some offered more than one). You are expected to be where the customer is.
The small print: 29 is a small sample, it over-represents startups, and Australia barely appears in these sources. The counts file is public, and the method is in the script beside it.
How much does a forward deployed engineer earn?
Where the ads stated pay, US base salaries ran from US$160,000 to US$270,000. Seven of the 29 printed a figure we could read: US$160k to $220k, $160k to $230k, $160k to $240k, $200k to $270k and one at $250k plus, all US startups with equity on top, and one London role at £70k to £90k. Those are senior-leaning startup roles in expensive cities; they are not a national average and we have no Australian figures from this sample. What they do show is that the market prices the role alongside senior software engineering, not below it as "customer support with code".
What skills do you need to become a forward deployed engineer?
Five, and the last two are the ones most engineers skip:
- Python and SQL, well enough to pull data out of a system you have never seen and trust the result.
- Data pipelines that rerun idempotently, because the customer's data will change under you.
- LLM workflows: tool use, structured output and the review points that let a person override the model.
- Evals on someone else's cases. Being able to say "it handles 91 of your 100 cases, and here are the nine it does not" is what turns a demo into a contract.
- Writing and handover. A scope a security team will sign, a results report for a stakeholder who does not care about models, and documentation that lets their team run it without calling you.
Our forward deployed engineer role page lists the day-to-day in more detail.
Can you become a forward deployed engineer without experience?
Not straight out of nothing. This is not an entry-level job: only five of the 29 ads stated a number of years, with a median of six, and most of the rest read as mid-level or senior. The usual route in is sideways, from a role that already has one half of the job:
- Software or data engineers who want more contact with customers.
- Solutions engineers, sales engineers and technical consultants who want to build the thing rather than scope it.
- Technical support leads who already debug customer systems and can code.
If you can program and have shipped something, the gap is usually the customer-facing half plus evals, and that is trainable in months. If you cannot program yet, learn that first; the FDE role sits on top of it.
How do you prove you can do the job?
With a portfolio that looks like an engagement, not a tutorial. A hiring manager for this role wants to see the whole arc on one realistic customer problem: the written scope, a pipeline that reruns, the workflow running on real records with its review points, an eval with a baseline, a deployment with monitoring and a runbook, and the handover pack. Six artefacts, each one something an employer can open.
That is how Square 1's Forward Deployed Engineer Bootcamp is built: twelve weeks, live on Zoom with one instructor, about 15 hours a week, six blocks that each end in one of those artefacts deployed and a gate you must pass, and a recorded viva defending the deployment. The entry bar is Python, SQL and having shipped something before. How the cohort runs has the dates and seats. If you want to check where you stand first, the free forward deployed engineer skill check takes about three minutes.
Is forward deployed engineering a good career in 2026?
It is a distinct, named job now, not a nickname: 16 of the 315 AI ads we counted on 19 September were titled forward deployed engineer, more than were titled ML engineer (13). The work exists because deploying AI inside real organisations is the hard part, and that is not going away soon. The trade-offs are real too: travel or on-site weeks, context switching between customers, and being the person in the room when something breaks. If you like the customer as much as the code, it is the best-paid way to use both.