DO-178C Was Written for Deterministic Code. Shield AI Is Using It to Certify a Neural Network to Fly Alongside Manned Fighters.
How One Award Moved Shield AI From Prototype to Production
On June 17, 2026, Shield AI was awarded a U.S. Air Force production contract to implement Hivemind mission autonomy software for the Collaborative Combat Aircraft program. The award moves the San Diego–based company out of the prototype phase and into the industrial-scale work of building AI that flies alongside manned fighters.
The contract's structure matters as much as the win itself. The Air Force treats mission autonomy as a standalone capability, decoupled from any single airframe. Through the Autonomy Government Reference Architecture (A-GRA), the service can push software upgrades to Hivemind across multiple aircraft platforms without redesigning hardware. That means Shield AI's code isn't locked to one drone; it's positioned to run across the broader CCA ecosystem.
"Mission autonomy is a foundational capability for future airpower," said Christian Gutierrez, senior vice president of Hivemind at Shield AI. Under the award, the company will focus on collaborative combat autonomy: multiple unmanned aircraft operating together under human supervision, coordinating tactics and reducing operator workload at scale.
The production contract follows Shield AI's selection as a mission autonomy provider in February 2026, after a competitive evaluation supporting Technology Maturation and Risk Reduction efforts. Hivemind is already flying aboard Anduril's YFQ-44A as part of ongoing development work, proof the software has moved past simulation.
For a company founded in 2015 that built its reputation on the V-BAT and X-BAT aircraft, the award marks a strategic inflection point. Shield AI is now a production-grade software supplier to the Air Force, not just a venture-backed startup demoing autonomy in controlled test environments. The shift from R&D to manufacturing-scale delivery is exactly the transition that kills defense-tech companies that can't hire fast enough, and it's already visible in the roles the company is rushing to fill.
18 Openings in Seven Days: What Shield AI Is Actually Hiring For
The CCA production contract didn't just change Shield AI's revenue trajectory; it opened a hiring pipeline that has added 18 roles in the past week alone. The positions span four locations (San Diego, Washington DC, Dallas, and remote) and they reveal a company shifting from proving autonomy works to proving it can be manufactured, integrated, and sustained at scale.
The most telling openings cluster around three functions: flight test, autonomy software integration, and mission-autonomy engineering.
Flight test is the bottleneck, and Shield AI knows it. The company is hiring a Staff Engineer, Flight Test (R4559) in San Diego, a role that sits at the intersection of traditional experimental test engineering and the newer discipline of validating AI-driven flight behavior. This isn't a position that existed in most defense firms five years ago. The job requires someone who can design test plans for autonomous aircraft, execute them in live flight environments, and feed results back into the autonomy development loop, a cycle that compresses when you're moving from prototype to production. Shield AI is also hiring a V-BAT Air Vehicle Operator in Dallas, a role focused on operating the company's vertical-takeoff-and-landing platform during test and evaluation. Together, these postings signal that the company is staffing up for a sustained flight-test campaign, not a one-off demo.
Autonomy integration is where the software meets the airframe. The Staff Engineer, Software – Autonomous Aircraft Integration (R4241) role, also based in San Diego, is tasked with getting Hivemind running on customer UAV platforms. The LinkedIn posting describes the work as integrating, testing, and demonstrating Hivemind across a variety of unmanned aircraft, which means the person in this role spends as much time in the field with hardware as writing C++ at a desk. The Senior Autonomy Flight Software Integration and Test Engineer (R3682) listing, which carried a salary range of $129,000–$194,000 on BuiltIn, went further: it required five or more years of software development experience, direct familiarity with flight controllers like PX4 and ArduPilot, and a willingness to travel 30% of the time for live customer demonstrations. That posting has since closed, which likely means the role was filled, a data point that itself speaks to hiring velocity.
Mission autonomy and pilot integration round out the core. The Autonomous Pilot Integration team, referenced across multiple LinkedIn postings, builds autonomy solutions for what the company describes as "a wide range of CONOPs and mission sets" (military shorthand for concepts of operations). These aren't generic software roles. They require engineers who understand how a human pilot and an AI system share control of an aircraft in contested airspace, and who can translate that understanding into working code. The Staff Engineer, Autonomy Integration (R3492) listing on Shield AI's Lever board called for seven or more years of experience integrating autonomy algorithms into aerospace and defense systems.
The geographic spread tells its own story. San Diego remains the engineering hub, the flight test and integration roles are anchored there, close to the company's headquarters and the Navy and Marine Corps test ranges that make Southern California a natural fit for defense flight operations. Washington DC roles skew toward government relations, security, and business development, reflecting the customer-facing side of a production contract. Dallas, where Shield AI operates a V-BAT production facility, is where the air vehicle operators and production-adjacent engineers land. Remote roles, like the Principal Technical Product Manager and Principal Designer positions added this week with salaries ranging from $220,000 to $340,000, suggest the company is willing to pay a premium for senior talent regardless of location.
The 18 roles added in the past week are a snapshot, not the full picture. But the pattern is clear: Shield AI is hiring for the hard middle of defense autonomy, the integration, test, and production work that separates a working prototype from a program of record. That's where the CCA contract pressure is felt most acutely, and it's where the talent market is tightest.
Why San Diego — Not Silicon Valley — Became Ground Zero for Defense AI
When Ryan Tseng founded Shield AI in San Diego in 2015, Silicon Valley wouldn't return his calls. "Ten years ago, Silicon Valley had turned its back on the defense sector," Tseng told the San Diego Union-Tribune. "It was hard to get meetings. We didn't get anybody to write us a check. Everybody thought we were crazy." A decade later, Shield AI is valued at $5.3 billion, employs roughly 950 people globally, and just closed a $240 million Series F round led by L3Harris and Hanwha Asset Management. The company's headquarters, and the bulk of its engineering workforce, sits in San Diego, not Palo Alto.
The geography is not accidental. San Diego offers something Silicon Valley cannot: proximity to the military test ranges, operational installations, and classified flight-test infrastructure that defense-autonomy engineers need to do their jobs. Shield AI's Hivemind software has been validated on F-16s, MQ-20s, MQM-178s, and its own V-BAT drones, testing that requires access to restricted airspace and military ranges concentrated along the Southern California and Gulf Coast corridors. The company's V-BAT vertical-takeoff drone, which Fast Company reported uses AI to fly itself without GPS or communications, was developed and iterated in a region where engineers can move from lab to live test range in hours, not weeks.
San Diego's military footprint is dense. Naval Base San Diego is the largest surface combatant fleet concentration area in the Navy. Marine Corps Air Station Miramar, Naval Air Station North Island, and the vast offshore and desert test ranges operated out of Naval Air Weapons Station China Lake are all within reach. For Shield AI's flight-test engineers, the company is actively hiring Staff Flight Test Engineers and UAS Autonomy Test Directors in San Diego, that proximity is the job. You cannot certify an AI pilot for combat from a WeWork in Menlo Park.
The talent market reflects this. Shield AI's senior positions span San Diego, Washington DC, San Francisco, Dallas, and remote locations. But the engineering core remains anchored in San Diego, where the company's Principal Technical Product Manager and Principal Designer roles (both carrying $220,000–$340,000 salary ranges) list San Diego as a primary hub. The city has become a cluster where defense-AI engineers, military test pilots, and autonomy software developers co-locate because the work demands it.
This is a broader pattern. As the War & Wealth analysis of defense-tech hubs noted, innovation in military technology is no longer confined to traditional industrial centers, but the new clusters form where the testing and operational infrastructure exists, not just where venture capital flows. Southern California's defense corridor, stretching from San Diego through Orange County (where Anduril is based) to the high desert, offers a concentration of ranges, depots, and program offices that no amount of Silicon Valley funding can replicate.
The irony is sharp. Silicon Valley's venture capital now pours into defense tech, over $130 billion into U.S. aerospace and defense startups in the last four years, according to the War & Wealth report, but the engineers building and testing the actual systems are increasingly working 400 miles south, in a city the Valley once ignored. Tseng's 50-year mission started where the runways and restricted airspace are. The talent followed.
The Hardest Problem in Defense AI Isn't the AI
The hardest part of building an AI combat aircraft isn't the AI. It's proving to the Air Force that the AI won't kill anyone, including the pilot it's supposed to protect.
Shield AI's Hivemind software has already demonstrated autonomous flight on multiple airframes, including mid-mission adaptability tests on Anduril's Fury prototype in February 2026. But flight demos and production certification are different problems. One earns headlines. The other earns a contract. The gap between them is where most defense-AI companies stall, and it's the reason Shield AI's current hiring push is weighted so heavily toward test engineers and safety-of-flight specialists rather than pure machine-learning researchers.
The DO-178C Problem
The standard that governs safety-critical software in airborne systems is DO-178C, managed under the FAA's Aircraft Certification Service. It defines a rigorous development lifecycle with five Design Assurance Levels (DALs), from DAL-E (no safety effect) to DAL-A (catastrophic failure would cause loss of aircraft and life). Flight-control and mission-critical avionics software typically falls at DAL-A, the highest bar.
DO-178C was written for deterministic software: code that behaves the same way every time given the same inputs, with every requirement traceable from specification to implementation to test. AI models break this model fundamentally. Their behavior emerges from training data, not from human-written specifications. A neural network doesn't have "requirements" in the DO-178C sense; it has weights.
A 2025 paper presented at an AIAA conference put it directly: "The integration of Artificial Intelligence in safety-critical aerospace systems has introduced new challenges in ensuring compliance with regulatory standards. DO-178C, the de facto standard for software development in the aerospace industry, provides guidelines for the development of safety-critical software. However, its applicability to AI software is not straightforward."
The industry hasn't settled on a replacement. EASA, the FAA, and EUROCAE/RTCA are all working on AI-specific certification frameworks, but nothing is finalized. That leaves defense contractors in an uncomfortable position: they need to certify AI systems now against standards that weren't designed for them.
How Shield AI Is Architecting Around the Gap
Shield AI's approach, visible in both its public technical materials and its job postings, relies on architectural separation. Hivemind doesn't replace the aircraft's certified flight-control software. It sits above it, issuing high-level commands while a safety layer called MM-RTA (Mission Manager — Real Time Assurance) continuously monitors those commands and intervenes if it detects an unsafe condition like a potential ground or air collision.
This partition strategy is becoming the standard architecture for AI integration in defense platforms. The AI operates as a separate subsystem with defined, governance-certified interfaces to the certified avionics. The certification claim shifts from "this model is safe in all contexts" to "this model, in this configuration, operating within these parameters, has been assessed and approved."
It's a pragmatic workaround, but it creates its own engineering burden. Every interface between Hivemind and the flight-control system needs to be specified, tested, and verified. Every failure mode the MM-RTA layer needs to catch needs to be enumerated. The safety case, the documented argument that the system is acceptably safe, grows enormously.
What the Job Postings Reveal
This is where Shield AI's hiring becomes legible. The company's open roles for autonomous-flight test engineers aren't asking for people who can train better models. They're asking for people who can build the evidence that a model is safe enough to fly.
That means engineers who understand DO-178C's traceability requirements, the demand that every requirement be traceable to the code that implements it, and every line of code be traceable back to a requirement. It means engineers who can design test campaigns that satisfy DAL-A objectives for systems that include non-deterministic components. It means people who can write the certification artifacts, the plans, the standards compliance matrices, the verification and validation reports, that a Designated Engineering Representative will review before signing off.
The administrative weight of this work is enormous. A Certidian analysis of aerospace certification workflows found that up to 60% of certification effort is pure administrative overhead: traceability mapping, document standardization, inconsistency detection, and documentation generation. A Tier 1 aerospace supplier with over 5,000 engineers reported reclaiming 12,000 engineering hours annually by addressing just the documentation burden. The B-21 Raider program reportedly cut its software certification time by 50% through process reform, a data point that underscores how much room for improvement exists and how much pressure programs are under to move faster.
For Shield AI, the production contract compresses the timeline. The company isn't just building a prototype that flies once at a test range. It's building a system that needs to be certified, produced at scale, and maintained across an operational fleet. Every month of certification delay is a month the Air Force doesn't have its collaborative combat aircraft.
The Human-Machine Teaming Layer
There's a second certification challenge that's less about software and more about trust. The CCA concept depends on AI-piloted drones operating in coordination with manned fighters, receiving commands from a human pilot, adapting to changing conditions, and making tactical decisions within defined boundaries. The human needs to trust the AI enough to rely on it. The Air Force needs to trust the human-AI team enough to deploy it in contested environments.
Certifying that interaction is a problem DO-178C doesn't address at all. It requires defining what "acceptable autonomy" looks like in combat, testing edge cases where the AI's judgment might diverge from what a human pilot would do, and building the telemetry and logging infrastructure to reconstruct and analyze every decision after a mission.
This is the work that doesn't map neatly to any existing standard, and the work that makes Shield AI's San Diego hub, with its proximity to Navy and Marine Corps test infrastructure, more than just a convenient office location. The engineers building these systems need to fly them, watch them fail, and document what happened. That requires runways, ranges, and relationships with military test organizations that you can't replicate on a video call.
The talent that can do all of this, bridge the gap between machine-learning research and DAL-A certification, build safety architectures for non-deterministic systems, and write certification evidence that survives DER scrutiny, is the scarcest resource in defense AI right now. Shield AI's hiring surge is the visible symptom. The underlying disease is a certification framework straining under the weight of a technology it was never designed to govern.
Three Companies, One Talent Pool: The CCA Software Fight
Shield AI's hiring push doesn't exist in a vacuum. On June 17, 2026, the Air Force awarded CCA production contracts to two airframe manufacturers, General Atomics for the FQ-42A and Anduril for the FQ-44A, while simultaneously setting up a separate, parallel competition among six vendors to supply the autonomy software that would actually fly them. Shield AI was one of those six, and one of three (alongside Anduril and RTX Collins Aerospace) picked to enter the first round of head-to-head software competition. The result is a fight for a narrow pool of engineers who understand both military aviation and machine learning, and the bidding war is already visible in the job postings.
Anduril added 160 roles in the past week alone, spanning manufacturing test engineers in Rhode Island to quality intelligence data engineers in Costa Mesa. These aren't prototype-stage headcounts. Anduril's listings for NDT inspectors, metrology engineers, and hydraulic assembly technicians in Santa Ana point to a company tooling up for sustained manufacturing. Shield AI's senior product and design roles suggest it's building out the teams that will integrate Hivemind onto production airframes at scale.
The structure of the CCA program itself is intensifying the competition. The Air Force deliberately decoupled hardware from software through the A-GRA, meaning the FQ-42A and FQ-44A airframes can accept autonomy stacks from any of the six vendors in the pool. General Atomics builds the FQ-42A, but its drone could run software from Shield AI, Collins, or Anduril. Anduril builds the FQ-44A, but Shield AI's Hivemind is the autonomy system currently paired with it. Every vendor is simultaneously a potential partner and a potential rival to every other. That architectural choice means the talent needed, engineers who can make an AI pilot meet military safety-of-flight standards on someone else's hardware, is the same talent all six companies are chasing.
The Air Force plans to narrow the software field from three to one or two vendors after an initial six-month performance period, with a final downselect to a single primary autonomy provider expected by summer 2027. Until then, Anduril, Shield AI, and Collins are in direct competition, and the other three vendors in the pool, Lockheed Martin, Northrop Grumman, and General Atomics itself, remain eligible to receive software contracts at any point over the six-year baseline. No one can afford to slow their hiring.
General Atomics, the incumbent with decades of Predator and Reaper production behind it, is hiring from a position of manufacturing maturity. Its president David Alexander said "manufacturing is already well underway" when the contract was announced. Anduril, which called the award "the first time that a new company has won a fighter aircraft program since the 1970s," is scaling its Arsenal-1 facility in Ohio. Shield AI doesn't build airframes at all, which means its entire bet rests on winning the software competition and licensing Hivemind across multiple platforms. That's a narrower path, and it demands engineers who can deliver combat-ready autonomy on someone else's schedule.
The Air Force wants over 150 CCAs by the end of the decade and roughly 1,000 across the program's lifetime. Nine vendors are already competing for Increment 2. Every increment, every lot, every software refresh cycle will require engineers who can certify autonomous behavior in safety-critical flight systems, a skill set that barely existed five years ago and that no single company's talent pipeline can supply. The production contracts didn't just launch a manufacturing program. They launched a labor market.
Inside Shield AI's 223 Open Roles: A Production Roadmap in Plain Sight
Shield AI's careers page lists 223 open positions. A close read of those postings, their seniority levels, location tags, and the specific language used, reveals a company that has moved past the prototype phase and is now scrambling to staff the unglamorous middle layer of defense manufacturing: the people who turn a working demo into something the Air Force can actually buy.
The Dallas factory floor is the priority. The most telling signal is where the roles are. Dallas, Texas, not San Diego, is where Shield AI is concentrating its production hires. The Director of Production listing is based there, with a salary range of $190,000–$290,000. The posting calls for someone to "architect and execute forward-looking production plans" and "drive aircraft production capabilities, implementing systems, processes, and staffing to support rapid growth." That is not R&D language. That is a company building a factory.
Dallas also shows up for roles like Receiving Quality Inspector, Technical Program Manager for Manufacturing, Senior Asset Management Specialist, and Manager of Propulsion (Fuel System). These are supply-chain and floor-operations positions. Shield AI's Frisco facility is becoming the manufacturing hub for V-BAT and, eventually, X-BAT airframes, and the company needs the staff to run it.
San Diego still owns the autonomy brain. The San Diego postings skew toward the core software and aerostructures work. A Staff Aerostructures Design Engineer role is based there, along with an Engineer I, Electromechanical position. The company's headquarters at 600 W Broadway remains the center for Hivemind development and the higher-level aircraft design work. But the mix of roles in San Diego also includes a Staff Hardware Recruiter, a sign that the hiring machine itself is being scaled up, not just the engineering headcount.
Seniority levels tell the scaling story. The open roles span from Engineer I to Director, but the concentration is in the middle: Managers, Senior Managers, and Staff-level engineers. Shield AI is not primarily hiring junior talent or C-suite executives. It is filling the layer that sits between the senior architects who designed Hivemind and the technicians who will bolt airframes together on the Dallas floor. The Associate Engineer, Manufacturing role in Dallas, listed at $85,000–$127,000 per year with a requirement for two or more years of experience in aircraft manufacturing, is a template for the kind of hire the company needs hundreds of: someone who has worked in aerospace production before and can operate in a classified, fast-moving environment.
Remote and distributed roles point to enterprise and government-facing growth. A Principal Technical Product Manager role lists San Diego, Washington DC, San Francisco, Dallas, and Remote as possible locations, with a salary range of $220,000–$330,000. A Principal Designer role carries the same geographic flexibility and a $230,000–$340,000 range. These are senior positions that don't need to be co-located with hardware, they cover the program management, customer-facing, and enterprise software layers that grow once a defense company moves from selling prototypes to managing production contracts and international partnerships.
The Washington DC metro area postings, including a Senior Sales Ops Admin and a Senior Director of Finance Transformation, reflect the government-relations and compliance overhead that comes with a major Air Force production contract. Shield AI needs people who can navigate Pentagon procurement processes, not just write autonomy code.
What's missing is as revealing as what's posted. There are relatively few pure AI-research roles visible in the current listings. The hiring is weighted toward manufacturing engineering, quality, supply chain, program management, and business development. That tracks with the company's stated transition: Hivemind is going into production for the CCA program, and the bottleneck is no longer the autonomy algorithm, it's building the aircraft, certifying them, and delivering them at scale. The company that existed six months ago, a defense-tech startup optimizing for demo milestones, is becoming something else: a manufacturer with a payroll problem.
The Air Force Wanted Competition. It's Getting a Labor Market.
The Air Force's Collaborative Combat Aircraft program has crossed a line that defense-AI companies have been approaching for years. In June 2026, the service awarded engineering, manufacturing development, and production contracts to General Atomics and Anduril for the first increment of CCAs, the FQ-42A and FQ-44A, with RTX's Collins Aerospace and Shield AI supplying the mission autonomy software for both platforms. Three companies are now in production. The question is no longer whether AI-piloted aircraft will fly alongside manned fighters. It is whether the defense industrial base can hire fast enough to build and certify them.
The scale of what's coming is hard to overstate. The Air Force has said it wants to field at least 1,000 Increment 1 CCAs before the end of the decade. That's not a prototype fleet. That's a force structure, one that demands manufacturing lines, test campaigns, software iteration cycles, and the engineers to run all of it. Anduril added 160 roles in the past week alone, spanning manufacturing test, quality engineering, and metrology positions across its Rhode Island and Southern California facilities. Shield AI added 18 roles in the same window, including a Principal Technical Product Manager position with a salary range of $220,000–$330,000, a signal of how aggressively the company is competing for senior talent that can bridge autonomy software and production-scale program management.
What makes this hiring surge structurally different from past defense buildups is the Air Force's deliberate architecture. The service built the CCA program around the Autonomy Government Reference Architecture, a modular open-systems standard that decouples mission software from airframe hardware. Col. Timothy Helfrich, the Air Force's portfolio acquisition executive for fighters and advanced aircraft, framed it plainly: the goal is to avoid locking into a single vendor, creating instead "a competitive ecosystem where the best algorithms can be deployed rapidly to the warfighter on any A-GRA compliant platform."
That design choice has workforce consequences. It means the Air Force isn't just buying drones from two companies. It's creating a market in which Collins Aerospace's Sidekick software flies on General Atomics' FQ-42A, Shield AI's autonomy stack is being integrated onto Anduril's FQ-44A, and both software vendors must maintain teams that can support, update, and certify their code across multiple airframes simultaneously. The engineering roles this demands, autonomy integration, safety-of-flight verification, cross-platform software sustainment, didn't exist as a category five years ago. Now they're among the fastest-growing job families on defense-AI career boards.
The talent war is also exposing a geographic fault line. The traditional defense primes, Lockheed Martin, Northrop Grumman, Boeing, built their workforces around hubs in the Washington DC corridor, Fort Worth, and Southern California's traditional aerospace corridor. The CCA production ecosystem is different. Anduril's manufacturing and quality roles cluster in Costa Mesa, Santa Ana, and Quonset Point, Rhode Island. Shield AI's engineering core sits in San Diego, with satellite teams in Dallas and Washington DC. General Atomics operates out of its long-standing Poway, California facility. None of these are the zip codes that defense recruiters have spent decades targeting.
The Air Force's own timeline is compressing all of this. General Atomics' YFQ-42A began flight tests in August 2025; Anduril's YFQ-44A followed in October. By February 2026, both prototypes had successfully integrated third-party autonomy software through A-GRA and logged semiautonomous flight hours, the YFQ-42A for more than four hours under ground-operator command. Jason Levin, Anduril's senior vice president of engineering, called it "a meaningful step towards fielding a real operational capability by the end of the decade." The service expects a final production decision this year.
For engineers watching from the outside, the CCA program is creating something the defense sector hasn't offered in a generation: a career track that sits at the intersection of AI research, flight-critical systems engineering, and production manufacturing, with the budget and institutional backing to sustain it for the next decade. The old model, in which autonomy work lived in labs and never reached a production line, is over. The new model demands people who can ship.
The bottleneck is that the pool of engineers who understand both machine learning and the certification standards for military flight software, DO-178C, DAL-A, the whole apparatus of safety-critical avionics, is narrow and getting narrower as commercial AI companies compete for the same talent. Every role Anduril and Shield AI fill is a role the other can't. Every month of hiring delay pushes the production timeline.
The Air Force wanted competition. It's getting one, not just between airframe vendors, but for the people who make the whole thing fly.
Working in frontier tech? Zero G Talent tracks the openings: browse frontier tech jobs, openings at Anduril Industries and Shield AI, and the people building the field.