The Autonomous Machine in Rock Creek Park
And What it Means for Drone Dominance Requirements
A 197-year-old gristmill illuminates the architecture of autonomy, the elemental core we still rely on today, and perspective about how to scale Drone Dominance.
Rock Creek Park led me to Peirce Mill today. A four-story gristmill built in Washington, D.C. in 1829, tucked into the park two miles north of the National Zoo.
Peirce Mill Gristmill, Cogs, Gears, and Engine of the Mill
We debate what “autonomy” means. Human-in-the-loop versus human-on-the-loop. Sensor-to-shooter timelines. Removing manual interventions from complex (and sometimes lethal) processes. And…we tend to consider autonomy a modern ambition, fueled by silicon, software, and ML.
Scout AI x Hendricks autonomous NOMAD® a UGV weighing under 650 pounds and delivering over 1,000 pounds of payload capacity. Powerful, low cost, enhances mobility and mission flexibility while reducing burden and extending endurance and maneuver options.
And yet, standing in the basement of Peirce Mill, I stared up up at a massive oak pit wheel meshing into a lantern gear. Inhaling dense oak wood, sweet, earthy compounds released over 200 years, inhaling cool, cave-like, hand cemented stone. Rich, loamy, residual, leather, stone, and wood.
Peirce Mill
In the midst of this reverie, intellect and awe converged. I was witnessing one of the earliest best-in-class autonomous systems ever built in America!
Let’s Resurrect Oliver Evans: The First Autonomous Engineer
Oliver Evans patented automated milling around 1790, roughly a decade after James Watt’s steam engine began reshaping British industry.
Although WATER POWER was the elemental core (we will revisit this), Evans was actually more interested in eliminating manual steps from a continuous operation than he was in the power generation itself.
The man, the myth, the legend, the stud, Mr. Oliver Evans
Milling used to be a batch process heavy on manual labor, and Evans noticed latency at every handoff. This led a vertically integrated system, from basement to attick.
The elemental core was GRAVITY and a single WATER wheel.
Gravity and Water powered every stage of the process from grinding, elevating, cooling, sifting, and sorting, with very little human intervention.
Isaac Peirce went on to build his mill to Oliver Evans’ specs. The result was a vertically integrated factory where water did the work of dozens of humans.
The Architecture: A Vertical Kill Chain?
(Un) Fortunately (?) , my brain sees a system and draws an analogy to defense tech. And it struck me that Peirce Mill’s architecture maps surprisingly well onto how we think about autonomous process chains.
Floor by floor here is the breakdown.
Gear Pit
The Power Plant: Exterior and Basement. Water from Rock Creek was diverted through a channel called a millrace and directed over a water wheel. The wheel’s main shaft entered the basement to drive a pit wheel, a large horizontal gear with wooden teeth called cogs. The pit wheel meshed with a smaller later gear, a cage of round wooden dowels, to convert horizontal to vertical rotation. This single power source drove every mechanism in the building. 60–70% of the energy turned the millstones and the rest powered the auxiliary automation.
Grinding and Input
Grinding and Input: Ground Floor. Farmers dumped their grist into a receiving hopper. The grain flowed down through wooden chutes to the basement, where a grain elevator lifted it back up and fed it between two massive round stones: a stationary bedstone and a rotating runner stone. The grain was crushed between them into meal autonomously.
Cogs of the Mill
The Attic: Cooling and Drying. The genius of Evans was alive in the attic. I could feel it buzzing. And let me tell you, those cogs were aesthetically pleasing.
Hopper Boy, Peirce Mill
Freshly ground meal is warm and moist. If you bag it immediately, it spoils. Traditional mills had boys raking meal across a floor by hand to air-dry it.
Evans automated this with a device he named the hopper boy, which is actually where the term “hopper boy” comes from. A round, spinning platform with mechanical rakes that spread the meal evenly across its surface to cool and dry. The meal entered hot from the elevator, and gravity slowly fed the cooled product down to the next stage.
Second Floor: Sorting and Grading. The dried meal descended to a bolter, a long, barrel-shaped cylinder set at an angle, fitted with graduated mesh screens. As the bolter rotated, the meal tumbled through increasingly fine screens.
One of the coolest design features IMO was the floor, where you can see cutouts that enable a modular redesign of the space:
Modular Mill Design
I was also enamored of this engraving, which my lovely tour guide Nick told me has adorned the column since 1937.
W.H. and E.M., 1937
Anyway, the entire system was connected by leather and canvas flat belts running over wooden pulleys, grain elevators carrying product between floors, and wooden chutes directing flow by gravity.
One water wheel. Four floors. Multiple sequential processes. Minimal human touch. Autonomy!
Autonomy Through the Ages
There are a few things about Peirce Mill that should resonate with those of us who are building autonomous systems today.
Single power source, distributed actuation. The water wheel is the mill’s engine, but its energy is distributed through a network of gears, shafts, and belts to actuate different mechanisms on different floors at different speeds. This is a mechanical precursor to the architectural challenge of distributed compute and power across autonomous platforms.
Autonomous Drones, Firestorm Labs
Vertical integration eliminates handoff latency. Evans’ entire insight was that the manual handoffs between milling steps — scooping, carrying, raking, sifting — were the bottleneck. Not the grinding itself. In defense autonomy, we talk about the same problem: the sensor-to-shooter timeline isn’t slow because sensors are slow or shooters are slow. It’s slow because of the handoffs in between. Evans solved his version of this problem in 1790 :)
Gravity as a design principle. The mill is deliberately built tall so that gravity does the work of moving material downward through each processing stage after the elevator lifts it to the top, and several stages like the sifters are built at downward sloping angles.
I must (briefly) interject that I observed the brilliance of this principle at play when we were designing steel, linear induction roller coasters at Premier Rides.
Steel Magnetic Launch Coaster, Elite Engineering, Premier Rides
The ideal coaster leverages magnetism upon launch and then wields the elemental core (gravity) to its advantage to sustain the ride. Gravity is free, reliable, ever-present and it is the primary transport mechanism. And great autonomous architectures do the same thing: they identify which forces or dynamics in their environment are free and reliable, and they design around them rather than fighting them.
Water: the original and enduring infrastructure. This might be the most important takeaway from Peirce Mill, and one I’m thinking about so much lately!
Potomac River next to the Mill
The entire system at the Mill runs on water, the medium for energy transference. And humans have been building autonomous systems on water since antiquity. Roman watermills ground grain across the empire. Medieval Europe was powered by tens of thousands of water wheels driving sawmills, fulling mills, and forge hammers. Water has never left the equation. We moved from water wheels to steam (water, heated), to hydroelectric dams (water, falling), to nuclear power (water, cooling). And now, as we scale the compute infrastructure that underpins modern AI and the autonomous systems built on it, water has returned to the center of the conversation. Water was the first infrastructure we built on, and water may be the constraint that ultimately governs how far we can scale.
Graceful degradation. The mill could operate at reduced capacity. If the bolter broke, you could still grind. If the hopper boy failed, you could rake by hand while the rest of the system kept running. The automation was additive and it enhanced a process that could still function manually.
Firestorm Labs xCell, Forward Deployed Additive Manufacturing
This is the same design philosophy behind human-on-the-loop autonomy: the system runs itself, but a human can step in at any stage without bringing down the whole operation.
Built to last. Peirce Mill ground grain commercially for nearly 70 years, from 1829 until the main shaft broke in 1897. It was restored to working order in the 1930s, and it still operates for demonstrations today. The oak gears, the stone construction, the fundamental mechanical logic endures. So yes, while attritable assets and drones mass are critical needs, I gotta say…it’s worth reflecting on machines that are designed for generations.
Advice on Drone Dominance from the Great Beyond (What Oliver Evans would tell us)
I left Peirce Mill thinking about a different program entirely: the Pentagon’s $1.1 billion Drone Dominance initiative, which aims to field over 300,000 domestically produced, attritable drones by 2028. Twenty-five vendors competed in the first Gauntlet at Fort Benning in February 2026. Delivery orders for 30,000 one-way attack drones are being placed now. Phase II narrows the field to as few as three vendors, with volumes scaling to hundreds of thousands of units.
Secretary Hegseth has explicitly called for moving away from the Pentagon’s decades-old preference for small numbers of exquisite, multi-million-dollar platforms in favor of mass, attrition, and speed. The Drone Dominance program website frames itself as “a commercial competition that leverages private capital, dropping prices while boosting lethality.” The target unit price for Phase I is $5,000.
But my spidey senses tingle at the tension between the philosophy and the requirements, or at least some of them.
Here are a few consolidated recommendations for how we should think about building autonomous systems at scale as it relates to Drone Dominance.
Design for the materials you have, not the materials you wish you had.
Oliver Evans didn’t wait for exotic alloys or precision-machined components. He built his system from oak, stone, leather, silk, and gravity, materials that were locally abundant, well-understood, and replaceable. The mill worked not because its components were exquisite, but because the system architecture was elegant.
UAS Build at the Scout AI HQ
We are not applying this lesson to drone requirements. NDAA compliance mandates are necessary and correct in that we cannot build a defense-critical industrial base on components sourced from adversary nations. But when compliance requirements at the bill-of-materials level force small drone companies to source globally for specialty components just to meet specifications that exceed what the mission actually demands, something has gone wrong. The requirements are shaping companies away from domestic prototyping and toward external sourcing, which is exactly the dependency the program was designed to break.
Group 1 through Group 3 drones, the small, tactical systems that Drone Dominance is targeting, should have requirements shaped around what American industry can build now, with materials available today, at price points that enable mass production. These are attritable systems. They are rounds of ammunition. The requirements should reflect that. ISR payloads. One-way attack. Simple, reliable, and cheap enough to lose. Don’t over-engineer the cogs when what matters is that the wheel keeps turning.
Acknowledge the supply chain bottlenecks — especially fiber and silicon — and design around them.
The fiber optic supply chain is a case study in how converging demands create strategic fragility. Fiber-optic tethered FPV drones have emerged as one of the most effective counter-jamming technologies on the battlefield — Ukraine has fielded them at scale to defeat Russian electronic warfare. The fiber of choice, G.657A2, is bend-insensitive and compact enough to spool on a drone. But global preform production is running at full capacity, and AI data center buildouts are consuming massive quantities of the same fiber. Corning signed a $6 billion fiber supply agreement with Meta alone. The price of a kilometer of standard single-mode fiber jumped 75% in January 2026. Specialty drone fiber prices are being revised weekly.
This is the water problem all over again. The same resource powering the mill is now needed to cool the data center. Fiber connects our networks and guides our munitions, and the supply chain wasn’t built for both at once. Rare earth metals for drone motors, carbon fiber for airframes, and NDAA-compliant batteries face similar pinch points. When we shape requirements for 300,000 drones, we are making industrial policy whether we intend to or not. The requirements need to be informed by what the domestic supply chain can actually deliver at volume, not by what a prototype can achieve in a lab.
Prioritize mass and swarming over individual platform sophistication for Groups 1–3.
The entire value proposition of attritable autonomy is cost imposition: a $2,300 drone that forces an adversary to expend a missile costing fifty times that amount, or at minimum to reveal their position, is a winning trade even if the drone is destroyed. That math only works at scale. Every dollar of additional per-unit cost — every exquisite sensor, every over-specified communication link, every gold-plated requirement — reduces the number of units you can field and weakens the cost-imposition logic.
Firestorm Labs HQ, rapidly manufactured drone capabilities
For Groups 1 through 3, the requirements conversation should center on: How many can we build? How fast can we replace them? Can they operate in denied environments with minimal communications — or none at all? Can they swarm effectively with simple coordination protocols? The Ukrainian experience has already shown that some of the most effective FPV drones fly their terminal phase with onboard AI and no communication link at all. No radio. No signal to jam. The autonomy is in the architecture, not in the individual platform’s sophistication. Evans understood this: the hopper boy didn’t need to be brilliant. It needed to be reliable, replaceable, and part of a system.
Treat cybersecurity as load-bearing structure, not as a coat of paint. When you’re fielding hundreds of thousands of networked autonomous systems, a cybersecurity failure is a loss of the entire capability. Security architecture needs to be designed in from the foundation, validated continuously, and treated as non-negotiable.
Let companies build, break, and iterate. Don’t require perfection before production. The Drone Dominance Gauntlet structure gets this right in principle: compete, demonstrate, deliver, repeat. But the broader acquisition culture still pulls toward locking requirements before vendors have learned what works at scale. The answer is not more exquisite requirements up front. It’s faster iteration cycles with simpler requirements, letting vendors learn by producing, not by PowerPointing.
Firestorm Labs Drone
Evans iterated too. His 1790 patent was a framework. Every mill built to his design adapted to local conditions, including the available water flow, the creek’s drop, the stone on hand.
The Drone Dominance program should enable the same thing: set the mission parameters, set the price ceiling, set the cybersecurity floor, and then let American companies engineer to the problem with what they have.
Re-Ground in the Elemental Core
The mill on Rock Creek still works because it was designed around an elemental core.
Water. Gravity. Locally available materials. Nearly two centuries later, have our design principles changed as significantly as we might think?
Build for the forces you have. Design around the materials that are available. Integrate vertically. Eliminate handoffs. Make the individual unit simple and replaceable, and put the intelligence in the architecture.
The next time someone tells you that autonomy is a 21st-century innovation, send them to Rock Creek Park on a Saturday. Admission is free. The museum interpreters (shout out Nick!) are excellent.
And the oak gears are still turning, powered by the same water that has been doing our work since before we had a word for engineering or autonomy.