From Replicator to Reality: Why Autonomous Swarm Programs Keep Falling Short
The Pentagon has spent billions on autonomous swarm programs. The hardware works. The software works. The doctrine does not. Here is why.
Replicator promised thousands of autonomous systems by summer 2025. Hundreds materialized. Swarm Forge promises validated swarm packages in ninety days. The demonstrations are real. The gap between demonstration and deployment is also real — and it is not a technology gap.
The problem is doctrine. Every swarm program to date has optimized for autonomous capability: can the drones avoid collision? Can they hand off targets? Can they execute without GPS? These are necessary but insufficient conditions. A swarm that can fly autonomously but cannot maintain mission coherence under attrition is not a formation. It is a collection of individually capable agents.
What separates a swarm from a formation is shared intent. When a drone is lost, does the remaining formation adapt to preserve the mission effect, or does it simply continue executing individual tasking that no longer adds up to a coherent whole? The answer depends on whether the system encodes doctrine at the architecture level, or merely adds it as an afterthought.
KhanBMS encodes doctrine as structure. The decimal hierarchy ensures that every tier carries sufficient intent to continue executing without real-time re-authorization. Attrition does not fragment the formation because the formation was never dependent on continuous micromanagement. The Mongols solved this problem eight centuries ago. Modern programs are rediscovering why it matters.