Shield AI builds AI pilots for military aircraft. Founded in 2015 by brothers Brandon Tseng (a former Navy SEAL) and Ryan Tseng along with Andrew Reiter, the company is headquartered in San Diego with major offices in the Dallas area and around Washington, DC. Its two flagship products are Hivemind, autonomy software that flies aircraft without GPS or comms, and the V-BAT vertical-takeoff drone, with the jet-powered X-BAT in development. The business is on a tear: after a $240M round at a $5.3B valuation in March 2025, Shield AI reportedly raised about $2B in early 2026 (a Series G led by Advent International and JPMorgan’s strategic investment group, plus $500M in preferred equity from Blackstone) at a $12.7B valuation, and in mid-2026 won a US Air Force production contract to supply Hivemind as the mission autonomy for the Collaborative Combat Aircraft program. That context shapes the interview: this is safety-critical, real-time software running on actual aircraft, so the bar for C++ discipline, systems thinking, and mission motivation is higher than at a typical startup, and citizenship requirements filter the pipeline before you ever talk to an engineer.
Process
Most first-person accounts and aggregator writeups describe roughly five stages, though details vary by team, so treat this as the typical shape rather than a guarantee. It starts with a recruiter screen covering background, motivation, work authorization, and clearance eligibility. Next comes a hiring manager conversation that goes deeper on your resume and how your experience maps to the team, whether that’s autonomy, perception, GNC, simulation, or platform software. The technical round follows: for most software roles this is a live coding session in the LeetCode-medium range, usually expected in C++, and some candidates also report a code review exercise where you critique realistic code rather than write it. Certain teams add a take-home; one ML candidate described a project scoped at about four hours that later became the subject of a presentation. The team panel is a set of interviews with future teammates mixing coding, domain questions, and system design. The final stage is an onsite or virtual onsite that often includes presenting your take-home or past work to a panel, plus conversations with senior leaders; some candidates report meeting executives as high as the CTO. Glassdoor pegs the average process at around 26 days, and most people report three to six weeks from application to offer. For the design conversations, our system design archive covers the fundamentals that come up here.
What they actually ask
- Modern C++ in depth: RAII, smart pointers, move semantics, const-correctness, undefined behavior, and threading. Autonomy and flight software teams treat C++ fluency as table stakes, and reports consistently rank it above Python for these roles.
- Algorithm problems around the LeetCode-medium level, often with a robotics flavor: graph search and path planning, geometry, and data structure choices under memory or latency budgets. Practice these in C++, since that is likely the language you’ll be asked to use.
- GNC and state estimation for the relevant teams: Kalman filters and sensor fusion, coordinate frames and quaternions, PID and control loop behavior, and what happens when GPS is denied, which is Hivemind’s whole premise.
- ML and perception questions for those tracks: model selection tradeoffs, designing a perception or ML pipeline for a constrained airborne compute platform, and how you validate models when failure means losing an aircraft.
- A code review or debugging exercise: several accounts mention being handed imperfect code and asked to find the problems, which tests taste and rigor rather than memorized patterns.
- Behavioral rounds probing mission fit: why defense, why Shield AI, how you handle ambiguity and long hours, and examples of ownership under pressure. Prepare structured stories; our behavioral guide covers the STAR format that works here.
Levels and comp (2026)
Levels.fyi data on Shield AI is thin, so treat these as directional medians rather than offers you can bank on: roughly $108K total comp for Software Engineer I, $128K for SWE II, around $170K for Senior, $250K for Lead, and $270K or more for Staff. Job postings tend to list wide base ranges, and San Diego and DC-area numbers run below Bay Area big tech, with equity carrying the upside story. That equity is private stock in a company now valued at $12.7B after several rapid markups, which cuts both ways: significant paper appreciation already happened, and the recent $500M in preferred financing sits above common shareholders. Ask directly about the current 409A price versus the preferred price, vesting schedule, refresh policy, exercise windows, and whether the company has run tender offers. Nothing here is unusual for defense tech, but walk in knowing the total comp conversation is more about equity belief than cash.
Prep priorities
Put C++ first. Whatever your target team, sharpen memory management, move semantics, concurrency, and the standard library, and do all your practice problems in C++ rather than Python. Second, grind medium-difficulty algorithms with emphasis on graphs, search, and geometry; the coding interview archive and our coding patterns cheat sheet are good starting points. Third, review the math for your specialty: Kalman filters and rigid-body dynamics for GNC roles, training and deployment tradeoffs for ML roles, schedulers and determinism for embedded and flight software roles. Fourth, prepare a tight technical presentation of a past project, since the final round often centers on presenting your work to a panel that includes senior people; practice defending design decisions out loud. Finally, rehearse your mission answer. Interviewers ask why you want to build weapons-adjacent autonomy, and a vague answer lands badly in a company that screens hard for conviction.
Frequently Asked Questions
Do I need a security clearance or US citizenship to interview at Shield AI? Nearly all engineering postings require US citizenship because roles must be clearance-eligible, and this gets confirmed at the recruiter screen. You generally do not need an active clearance to be hired, but the company sponsors clearances for roles that need them, and holding one already can speed things up.
Is there a take-home assignment? Sometimes. It varies by team and role. Some software candidates report only live coding plus a code review exercise, while some ML and autonomy candidates describe a scoped take-home, around four hours of expected effort, that they later presented at the final round. Ask your recruiter what your specific loop includes.
How hard is it compared to FAANG interviews? The pure algorithm bar is somewhat lower, mostly LeetCode-medium territory, but the domain bar is higher. You will be probed on C++ internals and on robotics, controls, ML, or real-time systems depending on the team, and the final presentation round has no FAANG equivalent. Glassdoor rates the difficulty around 3.2 out of 5.
Can I work remotely? Mostly no. Shield AI is heavily in-person across San Diego, the Dallas area, and the DC region, which fits the classified and hardware-adjacent nature of the work. A few roles show flexibility, but assume onsite by default and confirm the expectation early in the process.
Similar company guides
Prepping for Shield AI? Put it to work:
