In 2026 a software engineer leaving a big-tech job for a defense startup walks into a loop that looks familiar for about two rounds, then stops. The coding screen is normal. The systems round is normal. Then a recruiter asks whether you’re a US person, an interviewer spends an hour on interrupt latency and failure modes, and someone asks, without softening it, whether you’re comfortable building weapons. That last stretch is the part nobody rehearses, and it is where candidates who breeze through the technical rounds get filtered out.
The lane itself is new in scale. A decade ago defense engineering mostly meant the established primes. Now a cohort of venture-backed companies is hiring fast against real contracts, and the loop they run borrows from startup speed on top of the sector’s old constraints. If you want the specifics of any one company, the cluster this piece sits on top of has individual guides; what follows is the shared machinery.
The shared loop
Across the companies this cluster covers, the pattern is fairly consistent, and it is shorter than a FAANG onsite. Figure on four stages. A recruiter screen covers role fit, logistics, and the eligibility questions below, often with two or three quick technical questions in your discipline. A coding or systems round tests whether you can actually build: for software roles this is frequently closer to production engineering than a LeetCode grind, and for systems roles it is a design conversation about something the company builds. A domain or embedded/real-time deep-dive goes deep on the thing you’d own, with an interviewer who will push on your answers rather than accept them. A behavioral round, weighted toward ownership and mission, closes it out.
Two companies make a useful contrast for what “new primes ask” means. Anduril runs a loop built for speed and bias-to-action, closer to a product startup than a government contractor, while the established defense primes like Lockheed, Northrop, and Raytheon run a slower, more process-driven loop where the clearance you already hold can matter as much as the interview. Startups like Shield AI and Saronic sit on the fast end: fewer rounds, more depth per round, and a strong preference for engineers who have shipped hardware or real-time software rather than studied it.
Clearances and ITAR, stated plainly
This is the part that is specific to the sector, and the part candidates get wrong most often. Start with the distinction that decides whether a role is even open to you. A security clearance requires US citizenship. The ITAR US-person bar is wider: under the export-control definition (22 CFR 120.62), a US person is a citizen, a lawful permanent resident, or a protected individual such as an asylee or refugee. So a green-card holder can satisfy the ITAR requirement that gates most startup roles, but cannot hold a clearance. Someone on an H-1B or F-1 visa clears neither, which is why “must be a US person” in a job posting usually ends the conversation for candidates who need sponsorship.
The second thing people get wrong: you almost never get cleared before you are hired. A clearance has to be sponsored by an employer with a demonstrated need, so the normal sequence is a conditional offer, then sponsorship, then a wait measured in months. You can start on unclassified work right away with no clearance, and an interim Secret or Top Secret, when granted, lets you touch classified work before the full investigation closes. If a posting says “active clearance preferred,” that is a genuine advantage for people already cleared, and a reason the primes value clearance portability so highly.
The tiers themselves are worth knowing cold, because interviewers do ask which you hold and what you’re eligible for.
| Clearance tiers for US defense-tech engineering roles, 2026 | What it covers | Who qualifies / where it shows up | End-to-end processing (rough) | Before you’re hired? |
|---|---|---|---|---|
| None, ITAR US-person only | No classified access, but the role still touches export-controlled technical data | Most startup software and hardware roles (postings say “must be a US person”): citizen, green-card holder, asylee, or refugee. This ITAR bar does not by itself make you clearance-eligible | N/A | You must already qualify as an ITAR US person |
| Secret | Access to information whose disclosure would cause serious damage to national security | Common baseline for cleared engineering work on defense programs; requires US citizenship | Roughly 3 to 6 months | No; sponsored after a conditional offer, sometimes with interim access |
| Top Secret | Higher standard, with a deeper background investigation | Sensitive programs and senior architect or lead roles; requires US citizenship | Roughly 6 to 12 months | No; employer sponsors, interim TS sometimes granted |
| TS/SCI | Top Secret plus access to specific compartments; SCI is a layer, not a higher level | Intelligence-adjacent programs; requires US citizenship, and may add a counterintelligence or full-scope polygraph at the high end | Roughly 9 to 18 months, including SCI steps | No; sponsored, and any polygraph is agency- or program-driven |
Those processing windows are rough vendor estimates; the one official anchor is DCSA’s Q1 FY2026 reporting, which put end-to-end Top Secret processing near 227 days (per ClearanceJobs). Clearance eligibility is citizenship-only at every tier, which is the gate the ITAR US-person status in the first row does not clear.
One correction worth making to a common assumption: the polygraph is not a standard part of a defense interview. Most TS/SCI roles at contractors and startups never involve one. Counterintelligence and full-scope polygraphs are tied to specific agencies and programs at the high end, and you will know because the posting says so, not because an interviewer springs it on you.
Software and systems roles get their own rounds
Most people reading this are not embedded engineers. A big-tech leaver usually comes from web, backend, distributed systems, or platform work, and companies like Anduril and Shield AI hire heavily for exactly that: command-and-control and mission software, data and cloud infrastructure, and backend platform teams. The systems-design round is where a familiar skill gets reshaped. A FAANG prompt optimizes for scale and availability inside a healthy datacenter; a defense prompt often inverts the assumptions. Design a command-and-control service that keeps working over a contested, intermittently connected link with a hard latency budget, degrades predictably when a node drops off the network, isolates failures so one bad sensor feed cannot cascade, and deploys to an air-gapped environment with no cloud to phone home to. The data structures are the same; the constraints you optimize against are not.
Embedded, real-time, and safety-critical topics
For anything touching the vehicle or the flight computer, the deep-dive round goes where web interviews rarely do. Expect questions on real-time scheduling and why a missed deadline is a failure rather than a slowdown, priority inversion and how a mutex protocol prevents it, interrupt latency and what runs in an interrupt handler versus a task, watchdog timers and fault response, and why flight-critical code avoids heap allocation after startup. Hardware-in-the-loop testing comes up constantly, because you cannot push a patch to something that has already launched. Sensor fusion basics and deterministic timing round it out.
The specific flavor depends on what the company builds, which is why the deep-dive is where the cluster guides earn their keep. A hypersonics shop like Castelion or an aircraft company like Hermeus will push on guidance, control, and thermal and structural limits. A drone maker like Neros Technologies cares about flight-controller firmware, link reliability, and manufacturing at volume, and a broader weapons-hardware company like Mach Industries blends propulsion, structures, and embedded software in one loop. The through-line is that interviewers want to see you reason from first principles about systems that fail hard, not recite definitions.
The space-adjacent companies in this lane run the same embedded bar with an orbital-mechanics layer on top. True Anomaly works on space security and proximity operations, Muon Space builds Earth-observation satellites, and Impulse Space builds in-space maneuvering vehicles. For those, orbit determination, guidance and control, and flight software join the embedded questions above.
Cyber and autonomy roles screen differently again. A company like Horizon3.ai working on autonomous offensive security probes exploitation, detection, and attacker tradecraft rather than control loops. On the autonomy and ML side, the round moves away from datacenter training and toward the vehicle: expect questions on edge and on-vehicle inference under tight compute and power budgets, the sim-to-real gap and why a policy that works in simulation fails on real hardware, deploying and updating models on constrained or disconnected targets, and the mission-data pipelines that pull sensor logs back off a platform and into retraining. Clearance requirements often attach to the roles serving federal customers.
Mission-fit: the round that actually filters
The behavioral round in defense is not a culture-fit formality. Two things are being checked. The first is ownership under pressure, the same screen any serious hardware company runs: bring a project you drove through a setback where the outcome was in doubt, and talk about the decision you made with incomplete information and what you changed after it went wrong. A specific recovery lands better than a polished win.
The second is the one candidates underprepare: whether you want to build this. The question about being comfortable working on weapons gets asked directly at many of these companies, and the wrong move is to dodge it or to perform enthusiasm you don’t feel. A candid answer, grounded in how you actually think about deterrence, national security, or the specific problem the company works on, lands far better than a rehearsed line. A generic “I just love hard engineering problems” reads as evasive to people who weighed the ethics and signed up anyway; they want to know you did the same thinking before you applied. Have a reason you picked this company over an adjacent one, and make it specific to what they build.
The technical rounds reward depth you have or you don’t, and the clearance math is largely out of your hands once you qualify. The round you can most improve with preparation is the last one, and it is the round most candidates treat as an afterthought. Walk in having decided, for your own reasons, why this work and why now, and the rest of the loop is a fairer fight than it looks from the outside.
Practice the behavioral round:
