Over the past decade the United States has invested substantial economic resources in protecting its troops against improvised explosive devices (IEDs). Yet we know little about the impact of these investments on combat tactics and soldier safety. We introduce a model of insurgent learning where combatants adapt during an asymmetric war using defensive and offensive technological innovation. We test comparative statics of the model using declassified military records on individual IED explosions in Afghanistan from 2006 to 2014. Consistent with insurgent learning, we show that detonation and casualty rates did not decline during this period. This microlevel evidence is also consistent with the qualitative historical record from other substate conflicts. We conclude by decomposing variable input costs for defensive and offensive innovation presented in military documents.