Defense Primes Are Losing Young Systems Engineers to Startups
The talent migration reshaping program engineering on both sides of the defense industry divide
Three years ago, an engineer leaving Northrop Grumman or Raytheon for a defense startup meant leaving defense entirely—accepting that most VC-backed companies were not cleared, not connected, and not viable for anyone who wanted to work on programs that mattered. That calculus has inverted. The defense startup ecosystem now includes Anduril, Shield AI, Joby, Saildrone, L3Harris spinouts, and dozens of smaller Series B and C companies working on directed energy, autonomous systems, and space. They hold facility clearances. They win OTAs. Several are now prime contractors on programs of record.
The engineers know this. And they are leaving.
This is not a slow trickle. Engineering directors at multiple large primes, speaking without attribution, describe exit interview patterns that have changed sharply since 2024. Engineers with three to six years of experience—enough to be genuinely productive on complex programs, not enough to be trapped by pension vesting—are departing in clusters. When one leaves for a startup, two more follow within six months. The social proof is compounding.
What is actually being transferred, what is being lost, and what this means for the programs that depend on experienced systems engineering workforces at the prime level requires more precision than the usual talent shortage narrative provides.
What Is Actually Driving the Migration
The three drivers are real, but they are not equally weighted.
Compensation is the stated reason on exit interviews. It is partly true. Defense startup equity packages for engineers with five years of experience and a TS/SCI clearance are material. Cleared engineers are scarce enough that some Series C companies are offering refreshes that match or exceed what a staff engineer at a prime earns in total comp. But pure cash compensation at primes is not catastrophically below market. Base salaries for systems engineers in major defense corridors—Huntsville, Northern Virginia, San Diego—have moved significantly since 2022. Compensation is a threshold condition, not the primary driver.
Culture and autonomy are the heavier weights. Engineers in their late twenties at large primes describe environments where their functional contributions are real but their decision authority is minimal. They write sections of ICDs. They participate in working groups. They track action items. The actual architecture decisions are made by people fifteen years their senior, often through processes that have not materially changed since the 1990s. At a 400-person defense startup, a systems engineer four years out of school may own the interface definition between two major subsystems and have direct access to the chief engineer. The feedback loop between work and consequence is compressed by an order of magnitude.
Tooling is the factor that engineering leadership at primes is most reluctant to discuss publicly, because it directly implicates procurement decisions made by people who still work there. The dominant requirements management environment at major primes remains IBM DOORS or DOORS Next, supplemented by manual RTM spreadsheets, and governed by configuration management processes that can require multi-week cycle times for a requirements change to propagate through the system and reach review. Engineers who entered the workforce using modern collaborative tools find this environment genuinely difficult to work in—not philosophically objectionable, functionally painful. A requirements change that should take an afternoon takes three weeks. A traceability question that should take a query takes a day of manual cross-referencing.
Defense startups are not uniformly better here—some run on equally chaotic combinations of Confluence pages and shared drives—but the better-run ones have adopted modern requirements platforms that offer graph-based traceability, AI-assisted impact analysis, and real-time collaboration. The contrast is visible to any engineer who has worked in both environments.
What Institutional Knowledge Is Being Lost
This is where the conversation needs more precision than it typically gets.
The systems engineering knowledge that experienced primes carry is heavily procedural. It lives in how to navigate a CDR with a DoD customer who has specific and sometimes idiosyncratic expectations. It lives in understanding which DID requirements are actually enforced and which are satisfied with a one-paragraph compliance statement. It lives in knowing how to write an ICD that a subcontractor in Wichita will actually be able to implement, given that subcontractor’s specific organizational constraints. It lives in understanding the political geometry of a program—which government stakeholders own which requirements, which ones are immovable, which ones can be negotiated.
This knowledge is almost entirely undocumented. It is transmitted through proximity and repetition over years. When an engineer with four years of prime experience leaves, they take with them a partial apprenticeship that they have not finished. They understand the vocabulary and the process scaffolding. They do not yet carry the hard-won judgment about where the process is important and where it is theater.
The institutional knowledge that is not being lost is technical. Modern defense systems engineers are trained in model-based systems engineering to a depth that many of their senior counterparts are not. They understand graph-based dependency modeling. They can write system behavior in SysML or Cameo. They understand software-hardware interfaces at a level that reflects how modern defense systems are actually built, with heavy embedded software content and complex RF and EO/IR integration. This knowledge transfers intact to startups and is genuinely valuable there.
The net result: primes lose engineers who are becoming capable and have not yet peaked. Startups gain engineers who understand systems architecture but will need to learn government program execution through direct experience. Both ends of this exchange have structural gaps.
How Primes Are Adapting—and Why It Is Not Enough
The major primes have responded with three strategies, with varying levels of seriousness.
The first is compensation adjustment. RTX, LMT, and NOC have all moved base salary bands for cleared systems engineers, in some cases substantially. This has reduced the pure cash gap but has not closed the equity gap or addressed the non-monetary drivers.
The second is internal venture and rapid prototyping structures. Lockheed’s Skunk Works remains the canonical example of an internal fast-track organization. Others have created internal innovation units with smaller teams, faster timelines, and ostensibly more engineer ownership. The engineering culture within these units is genuinely different. The problem is scale: these organizations absorb dozens of engineers, not thousands, and they are often perceived by engineers on the outside as boutique exceptions that will not affect the mainstream career path.
The third is tooling modernization. This is the most promising avenue and the most underinvested. A handful of primes are actively piloting modern requirements management and model-based systems engineering tools in place of or alongside DOORS. The barrier is not budget—these tools are not expensive relative to program budgets—it is organizational. Changing the requirements toolchain on an active program requires convincing configuration management, the contracting officer, and the government customer. It often requires data migration from DOORS-formatted databases. The change management overhead is real, and the engineers who would benefit from the change are often the least empowered to advocate for it.
What Startups Are Gaining and What They Still Lack
Defense startups that are hiring from primes are acquiring something genuinely scarce: engineers who understand how to write requirements that survive government review, who have sat through a PDR and a CDR, who know what a TEMP looks like and why it matters. That knowledge compresses the learning curve on government program execution significantly. A startup that wins its first LRIP contract with several prime-experienced systems engineers on staff will navigate the transition differently than one staffed entirely with engineers who have never interfaced with a contracting officer.
What startups still lack is institutional memory about long-cycle programs. A defense startup working on an autonomous air system may be four years into its existence. A platform program at a prime may be fifteen years in. The understanding of how requirements drift over a decade-long development cycle, how to manage the accumulating technical debt in a specification baseline, how to reconstruct rationale for design decisions made eight years ago—this does not come from hiring four-year prime veterans. It requires a program to have run long enough to accumulate its own institutional memory, which most defense startups have not yet done.
The tooling environment at leading defense startups is a genuine differentiator. Modern requirements platforms built for systems engineering—those offering real-time traceability, AI-assisted change impact analysis, and direct integration with model-based design environments—are far more likely to be deployed at a 500-person defense company founded in 2020 than at a 50,000-person prime with a DOORS investment stretching back to the Clinton administration. Tools like Flow Engineering, which approach requirements as a connected graph rather than a document, are appearing in startups’ systems engineering workflows precisely because startups can make tooling decisions without navigating decade-old procurement infrastructure. The result is that some startups are developing stronger requirements traceability practices than the primes their engineers left—not despite being smaller, but partly because of it.
What This Means for DoD Programs
The DoD’s exposure here is concentrated and specific.
Large, multi-decade platform programs—DDG(X), T-7A fielding, Next Generation Air Dominance, Ground Based Strategic Deterrent—depend on stable, experienced systems engineering workforces at the prime contractor level. These programs have requirements baselines with thousands of individual requirements, complex interface control documents involving dozens of subcontractors, and government customers with institutional expectations about how reviews are conducted and how technical baseline changes are documented. When a prime’s experienced systems engineering workforce erodes, these programs absorb the consequences first. CDR readiness slips. ICD change cycles slow. Technical authority on the government side cannot compensate for eroded technical depth on the contractor side.
Acquisition offices have been slow to update how they assess contractor engineering capability during source selections. Resumes and org charts remain the primary evidence of workforce depth. There is no equivalent of a financial audit for engineering workforce quality. A prime can represent a systems engineering organization of 400 people on paper while managing an actual attrition rate that has reduced the functional depth of that organization by 30 percent over two years. Government customers will not see this until a program review.
The secondary exposure is on OTA contracts, where the risk calculus is different but the stakes are rising. DoD has deliberately pushed money toward non-traditional contractors and startups through OTA vehicles. Some of these organizations are acquiring systems engineering capability faster than the government’s risk models anticipated. Others are discovering, after contract award, that winning an OTA requires different competencies than executing one. The distribution of outcomes is widening, and acquisition offices do not yet have good visibility into which startups are on which side of that distribution.
Honest Assessment
The migration is not a crisis with a single solution. It is a structural realignment that reflects genuine differences in what large organizations can offer versus what they cannot.
Primes will retain engineers who want the exposure that only a multi-decade platform program provides: the experience of working on a system that will be in service for forty years, the technical depth that comes from watching a full development lifecycle. These engineers exist. They are not the majority of a generation that grew up building things that shipped.
Startups will continue to attract engineers who want to see the consequence of their work on a shorter horizon and who are willing to accept that they will learn government program execution the hard way. Some of those startups will succeed and will build their own institutional memory over time. Several will not, and the engineers at those companies will eventually return to primes or flow into the next startup.
The programs that should concern acquisition leadership most are those that have no good startup equivalent—programs too large, too classified, or too operationally sensitive to be replicated in a smaller organization. For those programs, the primes’ ability to retain and develop systems engineering depth is a national security input, not a commercial workforce issue. Treating it as the latter is the current default. That default needs to change.