Forward deployed engineer.
Four founding roles in a new unit, built from zero. You’d own the outcome inside one of Australia’s biggest operations, end-to-end — discovery to deployed, frontier AI in your hands. The autonomy, the ownership and the exposure are real — with 26 years of Kinetic IT trust behind you from day one.
The work.
You embed inside one of Australia’s biggest operations — an airline, a utility, a government department — and you find the friction that stands between skilled people and their best work: the manual reconciliations, the status chased and cross-checked across half a dozen systems, the hard-won know-how nobody has had time to write down.
Then you make it disappear. Frontier AI at your fingertips — agentic systems that reason, plan and act, custom agent harnesses, autonomous agents that hold an operation’s history and wake themselves on a heartbeat — shipped into their environment in days, not quarters. You sit next to the people who use it and watch the three-hour task become three minutes.
What’s worth building, you work out in the field — that’s the whole role. Nobody hands you a spec, because nobody has one. You work alongside the people who run the operation, surface the use cases that are both valuable and critical, and stand up proofs of concept — then let impact decide: double down on the ones that prove clear, immediate value. The winners move into production, with the rest of Kinetic IT behind you — and what they teach you feeds the platform, so the next deployment of the same outcome lands faster. Workflow by workflow, the operation runs differently than it has in years.
The brief, in four words: find friction, drive value. The mission is simple and large: close the gap between what frontier AI can already do and what the operations that run this country actually run on.
These operations are sophisticated because they have to be. The friction is rarely because people are careless; it is because critical systems accumulate constraints over years. The FDE’s job is to respect the operation deeply enough to change it safely.
The stack.
“Frontier AI in your hands” is the literal job — so here’s the kit behind it. Every model worth using, all three clouds at their AI frontier, a full realtime-voice pipeline, and a laptop built to run models locally. Provisioned the week you start.
You build in TypeScript, React and Node by default — fast to ship, proven, boring in the best way. But you deploy into the customer’s world: their Python, their Java, their .NET, their data platform, their ServiceNow, their Microsoft estate. The default gets you moving on day one; meeting them in the stack they already run is the job.
Major frontier and open-weight model families, through enterprise-approved routes. You benchmark aggressively, but customer security policy decides what can touch customer data.
Not just their plumbing — their model platforms. Whichever a customer already runs, you build where they live.
Phone-call-capable agents, end to end. The whole pipeline is yours from week one.
The harness around the model — durability, memory, retrieval, and the tracing to trust it in production.
A live URL in an afternoon: branch-per-PR databases, managed auth, transactional email, CI that ships.
64GB of unified memory runs the full dev stack plus mid-size open models (7–32B) locally — private by default, zero marginal cost, offline when you need it.
It’s all here from day one — Claude Max on every engineer’s machine, every frontier API behind the things you build. And because the point of a sandbox is cycle time, there’s no procurement queue: when a project needs a tool, it’s pre-authorised and live the same day.
One rule sits above the kit: no tech religion. The stack serves the customer’s outcome — never the other way around. Value decides.
And this kit is the sandbox — the first of three layers. It’s where value gets proven, not where pilots live: what works is generalised into the platform and lands in the customer’s approved environment — their cloud, their identity, their governance. Two clocks run in parallel: value by Friday, and the promotion path mapped from day one — so the security conversation never starts from a blank sheet after the demo. How that works is in the operating model.
What the day-to-day demands.
We’ll tell you this in these terms in every conversation we have, because the right person leans in and the wrong person self-selects out — and that’s the cheapest filter either of us has.
TRAVEL, FOR REAL
It’s national: roughly a week at a time, multiple times a quarter per account, more during go-lives — wherever the customer’s operation runs. Pilot wrap-ups and reviews in person. The relationships, the speed, and ultimately the value all run through presence. The other side of the deal: between trips there’s real flexibility — we care about the value you drive, not where you sit.
TRUST IS THE ASSET
Kinetic IT has spent the last 26 years earning unprecedented trust inside these operations. Everything you ship either enhances that trust or erodes it — we only do the former. You’ll earn your own the same way it was built: by delivering, reliably, inside someone else’s critical environment. The deliverable after a bad meeting is a better product, not a complaint.
ZERO TO ONE, IN DAYS
Discovery to first working demo in days. Production software in the customer’s environment, with their data, their security review, their constraints. Everyone can see whether it’s moving.
BEING ON POINT
When something breaks on the eve of a go-live, the first response is yours. You know the customer, the workflow, and the system. Around launches, you are backed by named platform, security, account and leadership support — but ownership stays with you.
AT SCALE, REPEATEDLY
Doing things that don’t scale — at the next account, and the one after that. The platform compounding underneath you doesn’t shrink the mandate; it grows the leverage.
What we measure.
Two numbers, no theatre.
Operational value per account.
Customer outcome value comes first: cycle time reduced, backlog cleared, cost removed, risk lowered, quality improved, hours returned to frontline teams. Renewal and expansion are lagging commercial proof that the value is real.
Product leverage.
Is each subsequent deployment of the same outcome easier? Time-to-deploy at the next account, and the platform share of each deployment — rising, relentlessly. It’s the single best indicator that gravel is becoming pavement.
Requirements, plainly.
Eight things we need to see. The traits behind them are unpacked in the FDE role conversation.
- Strong software engineering fundamentals: you can architect, build, and ship production systems alone, across the stack, in unfamiliar environments. Comfort in enterprise stacks like Microsoft and ServiceNow is a big plus.
- Hands-on depth with modern AI — LLMs, agents, retrieval, evals — and the judgement to tell the thing that demos well from the thing that actually works in production.
- Value-first judgement — no tech religion. You’re genuinely excited by the frontier and fluent in the latest tooling — and just as happy solving the problem with something vanilla, or in a stack that wasn’t your first choice, when that’s what delivers the customer’s outcome fastest.
- Evidence of shipped outcomes under ambiguity. Not projects. Outcomes: something that worked, that mattered to someone, that you drove against resistance.
- An instinct for productisation. You spot the reusable capability behind the specific fix, and you build so the next deployment is faster than the last.
- Confidence in customer-facing communication: you can support discovery, run workshops, explain technical options, hold a room of sceptical executives and earn trust with frontline operators — in the same day. Genuine, not salesy.
- A good understanding of enterprise data, identity, security and governance considerations, including secure data access, privacy, observability, guardrails and human-in-the-loop controls.
- Travel: national, on-site with your account, multiple times a quarter — with real flexibility between trips. If that’s a dealbreaker, this is the wrong role.
What we don’t require: a specific degree, a big-company logo, or deep domain expertise in any one industry. Domain knowledge can be trained.
Five situations. Two answers each.
No score. No percentage. The site doesn’t decide — you do. That’s the whole design. The traits these situations test are unpacked in the FDE role conversation — this is the applied version.
Monday is go-live. The version that works is narrow and rough — hard-coded edges, thin tests — but it’s safe and you can see exactly what it’s doing. The version you’d be proud of needs three more weeks.
How we’ll evaluate you.
Most companies hide the rubric. Here’s ours — all eight signals, published. It is hard to fake, because we triangulate the stories with a live working session, specifics, references, and what you actually built.
Shipped outcomes under ambiguity
“Tell us about something you shipped where nobody told you what to build.”
LISTENING FOR · How you found the real problem versus the stated one. What you cut to hit the timeline. Who used it, what changed, what number moved.
Ownership depth
“Tell us about a time something was failing and it wasn’t your job to fix it.”
LISTENING FOR · The right person can’t tell this story without smiling. The wrong one explains whose job it actually was.
Customer empathy vs salesiness
“What did the customer ask for — and what did you actually build?”
LISTENING FOR · Genuine empathy is specific: a user by name, by routine, by what made their job miserable. “Stakeholders” and “buy-in” are the wrong answer.
Constructive pushback
“Tell us about telling a customer — or your own product team — they were wrong.”
LISTENING FOR · Push back pointedly, then commit. All deference is a fail. All combat is a fail.
Pain translated into product
“What’s the worst a customer has ever treated you — and what did you change because of it?”
LISTENING FOR · The changelog after the story, not the survival. Contempt for past customers is disqualifying, however good the rest of the interview.
The splinter
We won’t ask for it directly — performed answers are easy.
LISTENING FOR · It shows up through biography: why you left, what failure still bothers you, what you built that nobody asked for. The real ones cost you something.
Speed under a real constraint
A live working session: messy context, unclear ask, hard deadline. Modern AI tooling encouraged.
LISTENING FOR · The order of operations. Interrogate the problem, find the valuable version, scope ruthlessly, build. We grade the choices and the working prototype — not the polish.
Cross-domain curiosity
“Tell us about a domain you went deep on that wasn’t yours. What did insiders miss?”
LISTENING FOR · The second and third question. The conference attended for fun. Red flag: every deep dive in your life was assigned.