Brian Schimpf and the Art of Building Without a Reverse Gear

At mile nine, the robot chose the wall. It was October 2005 in the Mojave Desert, and Cornell's entry in the DARPA Grand Challenge had begun the morning with 132 miles of bad road ahead. The Spider was a low, black military vehicle made autonomous with lidar, computers and enough improvised cooling to survive the heat. Brian Schimpf and fellow Cornell student Matt Grimm had helped assemble the team. Now their machine detected something that post-race inspection could not find, turned into concrete and stopped.

The indignity came with a technical punch line: the team had not implemented reverse. Time had run out. The Spider could struggle, turn and complain in the mute language of motors, but it could not back away from its mistake. It finished twentieth.

A tidy founder myth would make this the instant Schimpf understood everything. Real engineering is less obliging. He and the team simply went back to work. Schimpf delayed graduation. Cornell raised a $1 million grant for the next contest, and in 2007 a converted Chevrolet Tahoe named Skynet navigated intersections, obeyed traffic laws, merged with other vehicles and became one of six university-affiliated cars to finish the DARPA Urban Challenge.

A childhood in the long shadow of Kodak

Schimpf grew up around Rochester, New York, in a family of engineers. His mother and stepfather worked at Eastman Kodak. He did, too, writing code there during a summer job after ninth grade. The setting mattered. Kodak had made a city around technical competence, then became the corporate parable everyone knows. For an engineer coming of age nearby, invention and institutional decay occupied the same landscape.

Cornell was 90 minutes away through the Finger Lakes. Schimpf studied operations research and industrial engineering, a discipline concerned with decisions under constraint. The DARPA competitions gave that abstraction wheels, sensors and consequences. The robot did not care how persuasive the presentation had been. It either found the road or it did not.

The friendship built around those cars endured. Grimm would later call himself the “pointy haired business guy” and Schimpf the computer scientist. Their paths separated briefly after Cornell, then converged at Palantir. Schimpf joined its Washington office in 2007 as an early engineer. There he met Trae Stephens, another future Anduril co-founder, and Kori, an engineer who became his wife.

Palantir taught the code to meet the customer

Palantir sent small teams into difficult institutions. Engineers worked beside users, watched early software collide with reality, and revised it until it became useful. Schimpf served government organizations in intelligence, defense and law enforcement. He helped build Foundry, Palantir's commercial data platform, and eventually led engineering and product organizations as a director.

The decade changed his job. He arrived as a software engineer and discovered, as he later put it, an aptitude for management and strategy. Technical judgment widened into organizational judgment: which problems deserve scarce attention, who should own them, and how much disorder a growing company can tolerate before velocity turns into confusion.

One principle survived the transition intact: ambitious people do better work when they own the outcome. Schimpf's version is not a beanbag-chair theory of empowerment. Ownership includes late nights, server racks, failed deployments and the embarrassing discovery that a basic feature never made the deadline. The reward is that the work belongs to the person doing it, consequences included.

The CEO who had to be told he was the CEO

By 2017, Stephens, Grimm, Palmer Luckey and Joe Chen were assembling a company around a contrarian premise: defense hardware should be led by software, funded ahead of demand and sold as a working product rather than a promise attached to a cost-plus contract. Nearly all of the first dozen employees attended an early recruiting presentation. Schimpf did not. Stephens nevertheless told him that he was the CEO. Schimpf, alone among the future founders, was not yet convinced.

His friends wondered why anyone would leave a senior role at a rising Palantir. Schimpf saw the other side of the wager: an enormous problem, a founding team with complementary skills and credible financing. “There is no bigger swing,” he later said. Anduril began in 2017 with Schimpf as chief executive.

The first product, Sentry, was an autonomous surveillance tower. The early prototype used desktop graphics processors with liquid cooling so they could function in desert heat on solar power. It took about three months to get a prototype working, roughly six months to reach a pilot, and two and a half years to scale. Schimpf wrote significant portions of the early software for Sentry and the Ghost unmanned aircraft while serving as CEO.

This was management by muddy boots. An early colleague remembered driving all-terrain vehicles with Schimpf through the mud to reach a test site in Apple Valley. The symbolism required no company off-site: the chief executive was not above hauling himself up the hill to see whether the tower worked.

Speed is a theory of knowledge

Schimpf's case for speed is easily mistaken for impatience. It is closer to epistemology. A team that runs one expensive test must make the test precious. Preparation expands, risk tolerance shrinks and every choice accumulates political weight. A team that can test repeatedly learns more often. Each result improves the next attempt. “Going fast has an insane quality of its own,” he has said, because learning compounds.

That logic shapes Anduril's products and its economics. Lattice, the company's software platform, connects sensors and autonomous systems so people can understand what is happening and assign tasks without manually piloting every machine. Hardware teams borrow from commercial supply chains where possible. Schimpf has described a low-cost cruise missile whose fuselage uses a process associated with acrylic bathtubs and whose fuel tank draws on techniques used for plastic toys. The comedy is part of the point. A component does not become strategically serious by acquiring a four-year lead time.

His public argument is that the United States has concentrated too much money in a small number of exquisite, expensive and difficult-to-replace systems. He wants larger numbers of cheaper, adaptable systems that can be produced quickly and shared with allies. He also insists that humans should have better context for consequential decisions, not simply more data flashing before them.

That position carries moral weight that ordinary enterprise software does not. A failed update is not an inconvenient outage. Schimpf has repeated a special-operations customer's warning to engineers that code may sit between a hostile drone and the person it is trying to kill. The feedback loop must be fast, but it cannot be casual.

Can a test team become an institution?

Anduril now operates across air, land, sea, space and command software. By late 2025, Schimpf said it had 15 business lines and expected 7,000 employees by year-end. In 2026 it won its first NATO contract for Lattice, secured a production contract in the U.S. Air Force's collaborative combat aircraft program and moved Barracuda toward large-scale production. The garage experiment has acquired factories.

Scale produces the problem that Schimpf has spent his career diagnosing in others. Process appears for good reasons. Reliability, safety and public money demand it. Then process becomes habitat, the inhabitants defend the habitat, and the organization forgets that its purpose was to make something work. One senior Anduril leader has quoted Schimpf saying the company should remain a couple of clicks more chaotic than a sane person might prefer, because excess order can smother invention.

His daily life contains a small clue to how he manages the flood. He told Axios that he gets surprisingly few emails because Anduril is not an email company, but thousands of Slack messages. He considers high-volume information consumption a personal strength. Mornings are quieter. He has described waking around 6:30 or 7 with his two young children, eating breakfast with them, then heading to work.

The central test is no longer whether Schimpf can help a small team improvise a machine in the desert. It is whether he can keep thousands of people close to reality while contracts, factories and hierarchy pull them away from it. Cornell's Spider had no reverse gear. A mature company needs one. It needs the ability to admit a wrong turn, back out and try another route without turning caution into paralysis.

Two decades after mile nine, Schimpf's career still follows the same course markers: difficult systems, real users, short feedback loops and an unusual tolerance for finding out that the machine is wrong. The wall was never the interesting part. The return trip was.