Washington wants to build an Artificial Intelligence Force and install a federal tsar to run it.
Every pundit in theBeltway is nodding along, talking about national security, modernization, and keeping pace with rival superpowers. It sounds decisive. It sounds proactive. It is also a recipe for a multi-billion-dollar disaster that will institutionalize corporate lobbying under the guise of public defense.
I have spent years watching government agencies try to procurement-manage software they do not understand. They do not need a new military branch. They do not need a political appointee dictating compute allocation from a mahogany desk. They need to get out of the way of the engineers who actually build the systems.
The lazy consensus says that centralized state direction can out-innovate decentralized capital markets if we just throw enough red tape and federal funding at the problem. History suggests the exact opposite.
The Myth of the Centralized Command
When politicians call for an artificial intelligence tsar, they imagine a modern Manhattan Project. They picture brilliant minds huddled in a secure bunker, backed by endless state resources, solving intelligence through sheer willpower and executive orders.
That mental model belongs in a museum.
Modern machine learning does not scale through bureaucratic hierarchy. It scales through data velocity, massive compute clusters, and private capital allocation that moves at silicon speed, not fiscal-year budgeting speed.
If you put a political appointee in charge of federal model training, you instantly introduce three fatal friction points:
- Procurement paralysis: By the time a government agency clears the paperwork to purchase a cluster of high-end GPUs, the hardware is already a generation behind what a private lab is deploying in Ohio or Texas.
- Regulatory capture: A federal AI office inevitably becomes a revolving door for legacy defense contractors who know how to fill out grant proposals but cannot ship a performant transformer model to save their lives.
- Ideological risk aversion: Bureaucrats avoid failure above all else. Cutting-edge model training requires breaking things, running experiments that fail ninety percent of the time, and burning capital on dead ends. Government procurement rewards the safe, mediocre compromise.
We saw this playbook run with cybersecurity. The federal government created layers of oversight, compliance frameworks, and advisory boards, yet critical infrastructure remains vulnerable because the people writing the policies cannot write a secure line of bash script. Adding an AI tsar to the mix will simply create a high-profile target for corporate lobbyists trying to lock in federal subsidies for their proprietary models.
Why the Department of Defense Gets This Wrong
The pitch for a dedicated military AI branch sounds terrifyingly simple: give the generals their own codebases to automate logistics, drone swarms, and intelligence analysis.
The reality of military software procurement is a graveyard of abandoned multi-billion-dollar initiatives.
The Pentagon does not suffer from a lack of visionaries; it suffers from an institutional immune system that rejects high-velocity commercial technology. When a private startup builds a breakthrough capability, the military does not adopt it. They subject it to a three-year vetting process, demand custom security clearances that take twelve months to process, and insist on legacy compliance standards that neuter the software's utility.
An artificial intelligence force run out of Washington will not attract the top ten percent of machine learning researchers. Why would a top-tier researcher leave a private lab pushing the frontier of reasoning capabilities to work on legacy government infrastructure under a political appointee who thinks parameters are political talking points?
They will not. You will end up with a department staffed by second-tier contractors and career paper-pushers managing PowerPoint decks about automated defense strategies while the actual technological leaps happen in private labs miles away from federal oversight.
The Real National Security Threat
Washington is looking at the wrong threat vector.
The danger is not that the United States lacks a unified command structure for algorithms. The danger is that excessive government intervention will strangle the open-source ecosystem and domestic startup culture that actually keeps the country competitive.
When politicians start talking about nationalizing model safety or creating state-sanctioned AI architectures, they are chasing shadows. They want a handle to pull so they can feel like they are steering the vehicle. But machine learning is not a steering wheel situation. It is an ocean current. You do not steer it; you learn how to swim in it or you drown.
Countries that try to centrally plan compute allocation and model weights will find themselves holding expensive, obsolete models while agile ecosystems iterate circles around them. Central planning failed to pick winners in steel, semiconductors, and consumer software. It will fail spectacularly in automated reasoning.
If Washington genuinely wants to secure technological dominance, the prescription is boring, unglamorous, and completely opposed to what political appointees want to hear:
- Deregulate power generation: Compute requires power. If local and federal zoning boards continue to block nuclear and natural gas grid expansions to feed data centers, no amount of executive orders will save American infrastructure.
- Reform immigration for talent: Stop talking about national security while deporting the very PhD graduates from foreign universities who build the architectures we rely on.
- Open public data sets: The federal government sits on petabytes of clean, structured data that could train incredible civic and logistical models. Instead of locking it behind bureaucratic clearance walls, open it up to researchers.
Instead of doing the hard work of clearing infrastructure bottlenecks and fixing immigration, politicians prefer the theatrical option. They want a catchy title, a prime-time announcement, and a tsar to put on television when things go wrong.
Stop waiting for Washington to code your future. They are too busy figuring out how to tax it.
