Thoughts

---

The United States is very good at buying unmanned systems. The harder work is commanding them after the assumptions that make the demo look easy are gone.

That is the problem most of the charts keep sliding past.

Ukraine did not invent the drone. What it built, under fire, is a working habit of finding something, handing it to an effector, and not losing the thread. 

Open reporting on Delta keeps coming back to a number a battalion S3 actually cares about. The gap between a unit calling in a Russian position and the first rounds landing used to eat twenty minutes. Commanders now talk about three to five. That is not a software aesthetic. That is a staff process that stopped waiting on a perfect picture. Army Secretary Driscoll has said in public that the U.S. is still chasing that class of tool. The remark was not theater.

Copying the map is the tempting mistake. Delta works because enough video, enough radios, and enough determined people still reach a command post. A serious opponent will treat that post, that bird, and that satellite as opening targets, not scenery. GPS will lie or drop out. A command node that looked solid in an exercise will be the first thing on the list. If unmanned mass needs an untouched sanctuary to remain a force, it is still remote control. Expensive remote control.

NGC2 is the Army trying to grow up about this. After Capstone 6 they said the prototype phase was over and the scaling work had begun. That is real progress, and it should be treated that way. A common data layer across divisions is necessary. It will not, by itself, answer the harder question. A shared picture that only lives while the transport layer lives is an improved version of the last war’s command post. The test that almost never makes the demo is simple, and a little rude: kill the supervisor node. Deny GPS for an hour. Inject one teammate that is lying. Issue a task that steps outside the last published intent. Does the company still have a commander, or does it have aircraft waiting for a stick?

A lot of autonomy programs fail that test in spirit before they fail it on the range. They optimize the happy path. Leader-follower looks clean until the leader is gone. Swarm looks clean until the mesh is junk and one bad airframe starts herding the rest. Computer vision looks clean until the other side has had two years to study the model in the open. The formation does not need another confidence score. It needs the software to stop when the order is out of bounds, keep a relative picture when the constellation is dark, and elect a new lead before the package turns into debris.

The human piece sits under the tech and does not get enough airtime. Airframes are arriving faster than clear, local authority to halt a bad machine decision. If the only person who can veto that decision sits in a building that is no longer on the net, the formation is running on hope. The countries that get this right will treat doctrine as something that compiles into the runtime, not a file on a share drive.

None of this asks the United States to become Ukraine. Different force, different law, different appetite for attrition. It does ask us to stop treating platform count as command. Replicator-style mass is a procurement idea. Command of that mass, after the network is the target, is an operational one. Those are not the same budget line. They should not be the same slide.

Dominance in unmanned conflict will not go to whoever fields the most airframes in PE 2028. It will go to whoever can still task, refuse, and re-form a mixed team when the satellite is gone, the post is gone, and the only honest picture left is the one nearby vehicles still share with each other. Everything else is inventory.