The Department of War's FY27 budget proposes $54.6 billion for the Defense Autonomous Warfare Group (DAWG). DAWG in its current form is a coordinating organization rather than a unified command. It consolidated programs that previously sat under Replicator and adjacent efforts and provides a budget line and a coordination function, but it does not yet have the authorities of a unified command, the institutional independence of a separate service, or the force structure that would let it operate autonomous warfare capability rather than coordinate it. The institutional question worth working through is not whether DAWG works in its current form. It is whether the Autonomous Warfare Command (AWC) that DAWG eventually evolves into or is replaced by will work, and what the structural conditions are that determine the answer.
Standing up AWC on paper is a SECWAR and Congressional action. Making it actually work is a different problem. The structural questions that will determine whether AWC produces a real autonomous warfare capability or a coordinating body the services route around within a decade are not yet being asked publicly, and the answers will matter more than the budget number that has dominated the trade press coverage so far.
How Does the Transfer Actually Happen?
The first is wholesale program transfer by SECWAR directive, which is what happened with the Space Force absorbing Air Force Space Command. The Secretary issues a memo, the program transfers, and the gaining organization assumes responsibility for the budget, personnel, contracts, and mission. This is the cleanest model on paper but the slowest in practice because every transferred program has dependencies (sustainment contracts, classified networks, training pipelines, range access, foreign release agreements) that must be renegotiated or replicated. Five years after the Space Force stood up, it is still untangling shared infrastructure with the Air Force. Wholesale transfer also creates immediate political opposition from the losing service, which will lobby Congress to slow or reverse the transfer through appropriations language.
The second is the SOCOM model, in which the gaining command receives its own funding line (MFP-11), develops command-peculiar capabilities directly, and pulls forces from the services for employment but does not own the underlying service forces. SOCOM has Special Operations Forces assigned from each service component (Army Special Operations Command, Naval Special Warfare Command, MARSOC, Air Force Special Operations Command), but those forces are still recruited, trained, and largely sustained by the parent services. SOCOM owns the SOF-peculiar acquisition (specialized weapons, specialized aircraft modifications, specialized training facilities) but not the service-common base. This works for special operations because the SOF community is small and clearly separable from the conventional force. It works less well for autonomous systems because autonomy is now embedded in conventional formations at every echelon. There is no clean separation between an autonomous force and a conventional infantry squad with FPV drones.
The third is what could be called the standards-and-orchestrator model, in which AWC owns the integrating layer (architectures, software pipelines, testing infrastructure, common controllers, interoperability standards) but the platforms and operators remain organic to the services. This is closer to how DARPA operates on the development side, or how STRATCOM operates on the employment side for nuclear forces (it commands forces it does not own). The advantage is that it preserves service ownership of the warfighter while consolidating the parts of autonomy that genuinely benefit from joint integration. The disadvantage is that without a forcing function, services will find ways to route around the standards and continue to build their own.
The best outcome is a hybrid. AWC receives a Replicator-style budget line for AW-peculiar capabilities (orchestrator software, common controllers, attritable platform procurement at scale, AI/ML model development), the services retain their major program-of-record platforms (CCA, MASC, RCV), and Congress adjudicates the boundary every year through appropriations language. The goal is to avoid producing four decades of MFP-11-style boundary disputes if SECWAR simply follows the historical SOCOM trajectory.
There is also a timing dimension that must be examined. If the $54.6 billion is entirely new money through reconciliation, the services may tolerate it temporarily because it does not come directly from their accounts. But reconciliation funding is one-time. If the AWC budget becomes recurring baseline, it competes directly with CCA, MASC, JIATF-401, RCV, and every other service modernization priority in the annual budget cycle. The services will fight AWC hardest not when it stands up but when its first baseline budget request lands in the PPBE process. Without an MFP-11 equivalent that shields AWC funding from service interference, the command will be slowly starved through the normal appropriations cycle regardless of what its founding authorities say on paper.
The Pentagon will have to answer one question when AWC stands up. Will the services be told to turn over specific programs, or are they just ordered to coordinate? The answer determines whether AWC is a real command or a coordinating body. Right now the public messaging around DAWG is deliberately ambiguous on this point, and that ambiguity is allowing the services to maintain the status quo. The transition to AWC is the point at which the ambiguity has to resolve.
The Questions Nobody Is Asking
Most of the public discussion around DAWG either celebrates the budget number or laments the bureaucratic burden. The harder questions, which apply to AWC rather than to DAWG in its current form, are the ones that will determine whether autonomous warfare gets a real command or another layer of coordination on top of the service stovepipes.
Does Any Centralized Model Work at AI Speed?
Every organizational model under discussion is a Cold War or post-9/11 construct. SOCOM, Cyber Command, Space Force, USSTRATCOM, JIATF-401, all of them are built on quarterly or annual planning cycles, multi-year acquisition pathways, and personnel rotation cycles measured in years. Autonomy moves at software speed. Hivemind, Sidekick, and Prism iterate weekly. A model architecture trained Monday can be flying Friday. Can any joint command, with its inherent layering of approvals and consensus-building, operate at the cadence the technology demands? SOCOM took years to refine MFP-11 governance. Cyber Command took eight years from establishment to elevation as a unified command. Space Force is still fighting for separation five years in. If the institutional time constant for new joint commands is measured in years and the technology time constant is measured in days, the structural mismatch may be the most important problem nobody is solving for. AWC could be designed correctly and still arrive too late to matter.
Does Centralization Drive Innovation or Stifle It?
The honest answer from SOCOM is both, but unevenly. SOCOM has been extraordinarily innovative on the demand-pull side. It identifies operational requirements faster than the services and uses its acquisition authorities to field solutions in months that conventional forces would take years to acquire. It has also been notoriously dependent on services for almost everything supporting that innovation: logistics, base support, recruiting, medical, sustainment, professional military education. The SOCOM model concentrates demand-pull innovation in a small organization with operator-driven requirements while leaving supply-side innovation (platforms, capabilities, base infrastructure) to the services. For autonomous warfare, this might map cleanly. AWC could own the operator-driven demand signal (what behaviors do we need an autonomous swarm to exhibit, what mission profiles do we need an orchestrator to support) while the services continue to own the platforms those behaviors fly on. Or it might map badly, because for autonomous warfare the platform is increasingly a commodity and the autonomy is the capability. If AWC owns the autonomy stack and services own the platforms, the services may find themselves owning the less important half of the system.
Where Does the Talent Actually Go?
The best autonomy engineers in the country already work at Anduril, Shield AI, Palantir Technologies, Applied Intuition, Scale AI, and a dozen smaller firms. Most of them are not coming back into government regardless of which command structure exists. AWC cannot hire them at GS pay, cannot offer them startup equity, and cannot promise them the operational tempo they get in industry. The realistic talent pool is the same talent pool the services are already drawing from: military officers and NCOs with autonomy operational experience, civilian acquisition professionals, and a small number of contractors. If that pool is finite, then AWC either pulls from the services and weakens their internal autonomy expertise, or builds a parallel workforce that takes a decade to mature. The Space Force is dealing with this exact problem and has not solved it.
What Is the Industrial Base Effect?
Centralizing the autonomous warfare buyer might be good for industry by creating a single coherent demand signal with consistent requirements. Or it might consolidate procurement into a few large primes and crush the small innovative companies that have driven the recent surge. The history of joint commands suggests the latter. SOCOM, despite its reputation for innovation, increasingly buys from a small number of trusted vendors with long-standing SOF relationships. If AWC follows that pattern, the most innovative autonomy startups may find themselves locked out by primes who can navigate joint command acquisition processes that startups cannot. The current Replicator-era model of attritable mass procurement that DAWG inherited was specifically designed to bring in non-traditional vendors. A SOCOM-style joint command could work against that.
How Does the Classification Architecture Survive Consolidation?
Air Force CCA autonomy is classified at one level. Navy USV autonomy is classified at another. Army RCV autonomy sits on a third set of networks. Marine FPV operations cross all of these depending on the deployment environment. Joint commands historically struggle for years to reconcile classification differences across services (CYBERCOM is a current example). If AWC attempts to integrate autonomous systems across all four services, it will encounter a classification architecture that was not designed to support cross-service autonomous operations.
Who Runs the Algorithm, Who Owns It, and Who Updates It?
When a Marine squad employs an FPV drone running Hivemind autonomy, those are three different questions with three different answers. The Marine runs the drone at the point of use. Shield AI owns the underlying software as intellectual property and pushes behavior model updates from its engineering team, potentially overnight. The Department would hold the deployment authority that determines when and where that model is cleared to fly. No previous weapon system forced this separation. A rifle is owned by the service, maintained by the service, and employed by the service member, and the manufacturer's role ends at delivery. Autonomous software inverts that relationship, because the vendor remains in the loop indefinitely while the software keeps learning, keeps updating, and keeps depending on vendor infrastructure for retraining and deployment.
Title 10, Title 32, and Title 50 authorities were written for a world where ownership, maintenance, and employment collapsed into a single chain of custody, and they do not describe the autonomous software reality. The practical questions that flow from that break are the ones nobody has answered. Who is accountable when the algorithm fails on a mission, the operator who employed it, the vendor who wrote it, or the command that authorized its use? Who has the authority to push an emergency update to deployed autonomous systems mid-conflict, and through what pipeline? Who decides whether a model update is safe enough to certify for combat, and at what tempo can that decision cycle run? Who can deny an adversary the ability to exploit a known vulnerability in a model already fielded on hundreds of drones? None of the current organizational models, from service ownership to SOCOM-style joint command to Space Force-style service-like authority, has a clean answer to any of these. AWC will have to.
What About Coalition Operations?
Every realistic future fight is coalition. AUKUS, NATO, the Ukraine support coalition, autonomous systems will operate alongside allied autonomous systems. Each service has its own foreign release authorities, its own ITAR exemption history, and its own established working relationships with partner militaries. The Air Force has decades of NATO interoperability experience with manned aircraft that informs how it approaches CCA exports. The Navy has its Five Eyes maritime surveillance integration. The Army has bilateral arrangements with dozens of partner ground forces. AWC would need to either inherit all of those relationships or rebuild them, and the regulatory machinery for joint command foreign release is not well-developed. SOCOM has had to build its own foreign engagement infrastructure over thirty years and still operates largely through service component channels for partner force training.
What If the SOCOM Model Is Actually Wrong for This?
The implicit assumption in most of the public discussion is that the SOCOM model is the template for AWC. SOCOM exists because special operations require specialized units operating under unified command for low-density, high-skill missions. Autonomous systems are the opposite. They are becoming ubiquitous across every formation, every echelon, every domain. Centralizing ownership of a ubiquitous capability in a specialized command is a categorical mismatch. Every infantry platoon will have drones, every ship will deploy USVs, and every fighter squadron will fly with autonomous wingmen. A model that worked for a few thousand SOF operators may not work for a million-person joint force using autonomy in routine operations. The closer analog might not be SOCOM at all but something like the historical role of the Defense Logistics Agency: a service-supporting enterprise that owns the standards, the supply chain, and the integrating infrastructure, but does not employ the capability operationally.
The living logistics and living weapons framework applies directly here. The question is not who owns the drones but who owns the feedback loop that makes them better: the software update cycle, the operational data analysis, the performance testing, the industrial base management, and the sustainment infrastructure. If AWC controls that loop, it controls the pace of autonomous warfare innovation regardless of which service badge is on the operator's uniform. If nobody controls the loop, each service develops its own standards, its own software ecosystem, its own sustainment model, and its own testing protocols, and the joint force discovers on the first day of conflict that its autonomous systems cannot talk to each other.
That outcome is not hypothetical. It is what happened before Eagle Claw, before Goldwater-Nichols, and before every other organizational reform that Congress had to force because the services would not coordinate voluntarily. The question for AWC is whether autonomous warfare can learn from that history or whether it must repeat it.
The Forcing Function That Does Not Yet Exist
The transition from DAWG to AWC does not yet have a catastrophic failure to make the status quo indefensible. It does not yet have a Congressional mandate that overrides service resistance. What it has is a budget number, a growing threat in the Pacific, an accelerating technological trajectory that Ukraine has validated in combat, and the institutional memory of what happens every time the Pentagon is left to manage its own organizational transformation.
The harder questions above (transfer mechanism, AI tempo, talent, industrial base, classification, algorithm ownership, coalition integration, the SOCOM template itself) will determine whether AWC produces a real autonomous warfare capability or a coordinating body that the services will route around within a decade. None of these questions have public answers yet. None of them are being seriously addressed in the trade press or the unclassified policy discussion. All of them will be determinative.
The companies, investors, and operators trying to position for what AWC produces should be working on these questions now, not after the standup memo lands. The structural decisions made in the first eighteen months of the command will set the trajectory for what autonomous warfare looks like in the joint force for the next twenty years. The trajectory we are on says the structural decisions will be made under time pressure, with incomplete information, by a command that does not yet exist, against service resistance that has not yet fully mobilized. That is the worst possible decision-making environment for choices that matter this much.
The forcing function will arrive eventually. Either the United States and its allies will face a contested-environment fight that exposes the gaps in the current service-by-service approach, or Congress will move ahead of that fight and impose the structure that the services will not produce voluntarily. The first path is the one that played out before SOCOM. The second is the one that played out for the Space Force. Either way, autonomous warfare will eventually get its own command. The variables are how long it takes, how much institutional friction is burned in the process, and how much capability is missed in the gap between when the structural questions should have been answered and when they actually are.