Wet T-shirts on the antennas
The Wall Street Journal's account of Project Convergence, published October 10, is the most detailed public look yet at how the Army's new equipment behaved in the field. In July, about 9,700 infantry soldiers fought a ten-day mock battle at Fort Irwin against the opposing force standing in for the People's Liberation Army. The temperature approached 110 degrees.
Servers overheated. Batteries died and could not be recharged. Soldiers soaked T-shirts in ice water and draped them over antennas. One drone overheated and had to land. Another took about ten minutes to get airborne after a long wait to charge its battery, where Ukrainian crews launch a comparable aircraft in under a minute. Drones lost signal behind the mountains. Soldiers flew in limited windows, at half the maximum flight time to save batteries and avoid losing aircraft, and without autonomy or AI assistance.
The failures were mostly in the support layer
Read the list again and most of it has little to do with whether a drone can fly. It is power, heat, charging, launch readiness, links in broken terrain, and fear of losing an aircraft the unit will have to account for. Those are the jobs around the airframe that we described in The drone arrived before the pilot and in our paper on running an attritable force.
One scene captures it. A swarm of five small drones had to turn back when three Black Hawks carrying senior officials flew into the same airspace. Nobody did anything wrong. The drones' tasking had no picture of the crewed aircraft, so the safe answer was to cancel the whole sortie.
The software gains were real
The same reporting has good news, and it points the same direction. One battalion commander said new data-sharing tools cut the time to plan and communicate missions by two-thirds. Satellite connectivity let him reach an operations center across the base without putting up masts. The Army's chief of staff's technology officer described the exercise as a way to fact-check vendors' sales pitches, and that is the right use of a test like this.
The improvement came from moving information faster between people who had a link. The next step, as the Army itself put it, is making communications and data integration reliable enough that AI can help soldiers in remote and austere conditions, behind the mountain where the link drops.
Software that assumes heat, scarce power, and no link
EdgeLance is built for exactly those conditions. The mission lives on each device, so a node behind a ridge keeps its map, tasks, and AI and syncs over the team's mesh when a peer comes into view. Inference runs on the hardware the unit carries, staged through mission packs that are checked against each device's memory and power budget before step-off, so a hot, half-charged laptop is not asked to run more than it can.
Crewed aircraft belong on the same picture as the drones. Our multi-domain tracker takes in ADS-B and other aircraft reports alongside drone telemetry, and the tasking layer only offers work to assets that can do it. Airspace control measures as hard constraints on every task are on our 2027 roadmap, so that when helicopters enter a sector the affected tasks move and the rest of the swarm keeps flying. Every asset sits on one roster with its battery and health, so launch readiness is visible before the commander asks for it.
Test it the way Fort Irwin did
Brig. Gen. Daniel Hibner said it would be dangerous to hand everything over to technology, and that war is still a human endeavor. We agree, and it is the reason the support layer matters: soldiers will only rely on drones and AI they can keep charged, keep connected, and keep track of. Units evaluating edge software should copy Project Convergence's conditions on purpose. Run it hot, pull the link behind terrain, drain the batteries, and put crewed aircraft in the airspace, then see what still works.