In 1940 the Army Wrote the Jeep's Specification and Asked 135 Companies to Build It. Today a Drone Maker Decides What to Build, Pays for It With Investors' Money, and Waits for the Army to Buy.

In July 1940 the U.S. Army sent a written specification for a small four-wheel-drive scout car to 135 American automotive companies and gave them eleven days to bid.^[1] In 2026 a defense company valued at $61 billion builds what it chooses, with investors' money, and sells to the military afterward.^[11] Between those two facts is a ladder of four steps, and every step answers the same question: who writes down what the thing has to be?

The answer moved from the buyer, to a shared conversation, to a finder, to the builder. Each move shortened the distance between a need and a machine in the field. That distance is what a smaller player can win on.

Step 1: The government writes the specification. The Jeep, 1940.

The Army's specification was exact. It called for four-wheel drive, a weight limit of 1,300 pounds, a payload of 660 pounds including a .30-caliber machine gun, room for three, and a 75-inch wheelbase.^[1] Bids were due July 22.^[1] Bantam delivered the first prototype that September. Willys and Ford followed in November.^[1][2] The Army chose the Willys design in 1941, mainly because its engine produced 105 foot-pounds of torque against Bantam's 83 and Ford's 85, then had Ford build to the Willys drawings so that one factory's accident could not stop the supply.^[1] More than 640,000 were built.^[1] In the summer of 1945 Willys began selling a civilian version, the CJ-2A, and the line kept the CJ name until the 1986 model year.^[1]

The buyer decided what a scout car was. The companies competed on how well they could build to it. (The longer story is in DoAyni's earlier piece on the Jeep and the tank plant.)

Step 2: The government states the problem. The builders write the numbers. OXCART, 1957 to 1962.

When the CIA needed a successor to the U-2, it did not send out a detailed specification. CIA's own historians say Lockheed and Convair "were informed of the general requirement, and their designers set to work on the problem without as yet receiving any contract or funds from the government."^[3] The general requirement was to survive Soviet air defenses by flying very fast, very high, with a small radar return.^[3][4] Lockheed's Kelly Johnson proposed a cruise of about Mach 3 above 90,000 feet.^[4]

It was a real competition. A panel scored Lockheed's A-12 against Convair's design on a common table of speed, range, altitude, radar cross-section, cost and first-flight date.^[3][4] CIA's representatives at first preferred Convair's design because its radar signature was smaller. Lockheed won because its specifications were slightly better, its projected cost was significantly lower, and it "had produced the U-2 on time and under budget."^[4] Conditions came with the award: show a smaller radar return by January 1, 1960, and accept a clause for "periodic reevaluation of costs."^[4]

Lockheed's freedom was in how it worked, not in what the aircraft had to do. Johnson's log records the CIA's acceptance that Lockheed's "method of doing business will be identical to that of the U-2."^[4] The A-12 first flew in April 1962.^[3][4] The SR-71 was the Air Force's two-seat follow-on version, and it first flew on December 22, 1964.^[5][6]

Step 3: The government finds what already exists. In-Q-Tel, 1999.

By 1999 the best technology was being built in commercial companies that had never heard of a government requirement. The CIA's answer was to set up In-Q-Tel as "an independent, not-for-profit strategic investment firm."^[7] Its stated purpose is to bridge the gap between the technology needs of the intelligence and defense communities and the private sector.^[7] It lists a dozen government partners, from the CIA, NSA and NGA to the Army, Space Force and the FBI.^[7] Its own page puts technology development agreements at between $500,000 and $3 million and equity-only investments at roughly $250,000 to $500,000.^[7]

Nobody writes a specification at this step. A need is stated, the market answers, and a finder shortens the path to the customer. The CIA's own authority is unusually broad: funds available to it "may be expended without regard to the provisions of law and regulations relating to the expenditure of Government funds" for confidential, extraordinary or emergency purposes.^[8] In-Q-Tel, though, is a separate nonprofit, and the statutes do not name it.^[7][8] (The author worked at In-Q-Tel from 2003 to 2016.)

Step 4: The builder writes the need. The government buys. 2017 on.

Anduril Industries, founded in 2017, has raised more than $11 billion, including a $5 billion round at a $61 billion valuation in May 2026.^[11] The company says it funds research itself "to cut development timelines and move faster than traditional industry, avoiding the delays and cost overruns typical of cost plus contracts."^[9] Its founder said it designed, built and demonstrated its Roadrunner drone with its own funding for about two years before naming a customer.^[10]

It has also entered the formal system. In January 2022 it beat 11 other bidders for a Special Operations Command counter-drone integration contract worth nearly $1 billion.^[13] In June 2026 the Air Force awarded production contracts for the first three lots of its Collaborative Combat Aircraft program to Anduril's FQ-44A and General Atomics' FQ-42A, a program that began as a government solicitation. Orders for the first lot follow once the 2027 budget is approved.^[12] In September 2025 Australia committed A$1.7 billion to its Ghost Shark undersea drone, after the government had co-funded prototypes from 2022.^[14] The honest version of step 4 is not that the company never meets a government requirement. It is that it starts without one, and arrives at one with a product that already runs. That is also the Jeep's arc: a prototype becomes a standardized contract.

What the short loop buys

War is where the shortened loop gets tested. A New York Times chart of munitions costs from the 2026 U.S.-Iran war, reproduced by Scott Galloway, an NYU professor and serial entrepreneur, in his newsletter, puts an Iranian Shahed drone at about $35,000, against interceptors that cost millions: roughly $4 million for a Patriot PAC-3 and $2.1 million for an SM-2.^[20] Unit prices are not the cost of a kill, and the Shahed's own cost estimates range from $20,000 to $50,000.^[20] But the same chart shows the answer arriving: an interceptor drone called Merops at about $15,000, which the Army bought 13,000 of in roughly eight days.^[16] The system is American-built and was first used in Ukraine in 2024, so "Ukraine-tested," not "Ukraine-made."^[16]

In Ukraine the Center for Strategic and International Studies estimates that the ratio of Russian to Ukrainian casualties, killed, wounded and missing, rose to nearly 8 to 1 in the first half of 2026, up from between 2 to 1 and 3 to 1 for much of the war, "in large part due to Ukraine's use of drones."^[15] The pattern holds in all three domains. In the air, the Pentagon's Drone Dominance program bought about 30,000 one-way attack drones in March 2026 from 11 vendors at about $5,000 each.^[19] At sea, satellite imagery showed damage to Russian frigates at Novorossiysk in August after an attack with sea drones, jet-powered drones and missiles.^[17] On the ground, Ukraine's defense minister set a goal of contracting 25,000 ground robots in the first half of 2026.^[18] Copying runs both ways: Russia led on fiber-optic drones at scale, and the Shahed concept is Iran's.^[21] Faster innovation keeps a smaller player in the game. It does not decide the game.

Why does this matter?

The Jeep went home with the soldiers who drove it, and by the summer of 1945 it was a farm truck.^[1] The same thing is beginning with machines that decide for themselves. War is paying for the development, and the lesson of the four steps is where the money flows: toward whoever shortens the path between a need and a working machine. The question to ask about any new capability, military or civilian, is not only what it can do. It is who wrote down what it had to be, and how long that took.

This piece is one part of a longer argument. It is the companion to Osparna's "Who Holds the Pen, and Who Can Wield It", which asks the same question of the technology the government pays for and does not own. See also DoAyni's earlier pieces on the Jeep and the tank plant, on how a Korean War pilot's decision loop beat a better aircraft, and on the CIA's investment arm and the Pentagon's.