The DoD is the Umpire

Fix the arsenal, don’t break the institution

Shyam Sankar’s “Steal the Ball, Kill the Umpire” metaphor is as seductive as it is wrong. To win the next war the U.S. need simply break the rules, move fast, innovate recklessly. Like many in tech Shyam conflated disruption with effectiveness and romanticizes wartime chaos as a model for security policy. In doing so, Sankar underestimates the scale, complexity, and necessity of the Department of Defense (DoD) and the layers of systems necessary to train an effective fighting force all while overhyping AI's readiness.

Someone has to protect the umpire (ChatGPT generated)

I love the OSS. I think I have read 15 books on them, I am a member of the OSS Society. I am going to say something that should not be controversial. The OSS did not win WWII. Sankar invokes “Wild Bill” Donovan as the archetype of successful subversion, which he was, but again, he didn’t win the war. He was a cog, just like Churchill’s SOE, in a much larger machine. That machine’s lifeblood was the War Department’s ability to coordinate and lead the U.S.’s industrial and logistical might.

During WWII, the U.S. produced:

• 296,000 aircraft

• 86,000 tanks

• 64,000 landing craft

• Over 2.6 million machine guns

The U.S. built 2,710 Liberty ships and over 5,000 other merchant vessels between 1941–1945. 7,710 ships to move the ammunition, food, clothing, and troops across the Atlantic and Pacific oceans. The U.S. built at least one Liberty ship in four days! This was the circulatory system that kept the U.S. and allies in the fight, save Britain and Russia, resupplied the Chinese Nationalist fighting the Japanese.

The production miracle didn’t emerge from rogue innovation. It came from centralized coordination, public-private collaboration, and disciplined command structures. Marshall, a man of systems, was much more instrumental to the war effort than Donovan. For all intents and purposes the War Department was the umpire. The umpire is the reason we won. Ford did not make tanks and planes because it wanted to, it made them because the War Department and the War Production Board forced them to. Sankar’s idea that we need to "kill the umpire" forgets who the umpire is in this case.

The DoD is not perfect. God knows I saw my share of bureaucratic inefficiencies over my career. Contractors and Primes do not help, major programs have overrun cost, and the DoD’s tech inertia is real. However, the core role of the DoD’s is to form the institutional backbone of American defense. That is irreplaceable. If you “kill the umpire,” who governs strategy, force projection, nuclear command and control, multilateral defense alliances, and industrial mobilization? Palantir? The DoD is the only is the only organization that can coordinate across agencies, technologies, geographic boundaries, and alliances. War is still about who can bring enough mass to bear to force their enemy from the field of battle. That core principle cannot be reduced to a hackathon. To believe otherwise is a Silicon Valley fantasy.

Sankar assumes AI can replace the legacy systems in defense infrastructure. It’s a horrifying thought. AI, especially LLMs, is not a thinking machine. It parrots the next word based on patterns, not meaning.

• LLMs don’t reason.

• LLMs can’t verify truth.

• LLMs hallucinate facts.

How do you trust such systems in war? In 1986, Stanislav Petrov, a Russian Colonel, prevented a nuclear exchange with the U.S. by ignoring protocol and not “returning” fire when U.S.S.R missile early warning malfunctioned. 1979 and 1980 the U.S. had three similar close calls because on human and machine errors. What happens when an AI system hallucinates a missile trajectory? What if that AI doubles down on its mistake, as current leading edge AIs are prone to do? To be clear: AI is a powerful tool, but tools need operators. Without human oversight and institutional checks, it’s a liability. The idea that we are going to remove cognitive load to the point of automating some of the core aspects of national defense should worry everyone.

Sankar's AI-centric vision also assumes perfect access to bandwidth, power, and data integrity. That’s a fatal error. Since at least the Gulf War, China and Russia built robust capabilities to contest the electromagnetic spectrum including:

• GPS jamming and spoofing

• Targeted disruption of satellite communications

• Integrated EW doctrine at the operational and tactical level

Modern, sensor-fused, networked systems rely on vulnerable links. No matter how smart they are, they need access to data to enable their decisions. An AI-enabled system denied access to its cloud-based data backend becomes a $150 million paperweight. Technology doesn’t dominate in a vacuum. It survives or dies based on its robustness in degraded environments. That’s exactly where China and Russia plan to fight.

Its not the Umpire, It’s the Scoreboard

Yes, the DoD has problems. The solution isn’t to “kill” it. The solution is to retool it with urgency and humility. We need to:

1. Invest in Existing but Proven Systems
   Leverage systems that work under duress: Aegis, HIMARS, and EW-hardened platforms all the way to Assured-Position Navigation and Timing end user devices. Don't throw systems that away chasing the latest untested LLM framework or networked death machine.

2. Rebuild the Arsenal of Democracy
   Reshore critical manufacturing (like legacy chips). Expand munitions plants. Build more dry docks. This requires a purchaser to make these industries profitable. Then, focus on logistics. Our defense is only as good as our logistics. As

   Tony Stark
   pointed out earlier this week, our logistics are falling behind and the ships and open-sea sub tenders we need are not high technology.

3. Field-Test AI in Reality, Not Sandboxes
   Training matters. AI and autonomous systems absolutely have a role, but they must be pressure-tested under jamming, cyber denial, and real operational friction before replacing manned systems. This is also true of U.S. units. How do we develop measures and controls for AI hallucinations? By training, in hard environments constantly. In this, tech needs to get into users hands faster and, if it fails, die just as quickly. The DoD should make every training event a contested environments with enemy EW and signals. This requires lifting restrictions on training and make it easier for units to train in these environments.

4. Create Layered, Redundant Command Systems
   Assume loss of data link, exactly when you need it. Murphy’s Law will apply to our machines too. Design platforms that degrade gracefully rather than collapse when isolated.

It’s odd. A few decades ago techno-optimists believed technology would lift humans up, remove the need for war, and fight misinformation. The truth is, while we have seen amazing technological advancements, the more pie eyed visions have not borne out. Shyam Sankar’s argument strikes me as similarly naive. It takes one valid observation, that the DoD needs to move faster, and runs off a cliff with it. To win the next war, we don’t need to kill the umpire. We need to train better players, modernize the rules, and make sure the game itself still serves its purpose. The DoD doesn’t need to be dismantled. Smashing the old system without a new, tested one to replace it will weaken the military. Instead, it needs to be rearmed with new tools and the wisdom that built the arsenal of democracy.