The study investigates the mechanisms by which command-like descending neurons (DNs) in *Drosophila* orchestrate complex behaviors. Command-like DNs, which are small sets of neurons, are known to drive specific behaviors but the underlying circuit mechanisms are unclear. The research reveals that these command-like DNs co-activate larger populations of DNs to generate behaviors requiring the active control of multiple body parts. Connectome analyses and experimental manipulations show that this recruitment is mediated by direct excitatory connections between command-like DNs and interconnected DNs in the brain. The study also finds that behaviors driven by DNs with many downstream partners require network co-activation, while those driven by DNs with few partners do not. These DN networks form excitatory clusters associated with distinct actions and inhibit each other, suggesting a mechanism where behaviors are generated by combining multiple motor subroutines through the recruitment of increasingly large DN networks. This framework reconciles the two models of DN control: command-like DNs drive complete behaviors by recruiting additional DN populations, which combine and coordinate multiple motor subroutines.The study investigates the mechanisms by which command-like descending neurons (DNs) in *Drosophila* orchestrate complex behaviors. Command-like DNs, which are small sets of neurons, are known to drive specific behaviors but the underlying circuit mechanisms are unclear. The research reveals that these command-like DNs co-activate larger populations of DNs to generate behaviors requiring the active control of multiple body parts. Connectome analyses and experimental manipulations show that this recruitment is mediated by direct excitatory connections between command-like DNs and interconnected DNs in the brain. The study also finds that behaviors driven by DNs with many downstream partners require network co-activation, while those driven by DNs with few partners do not. These DN networks form excitatory clusters associated with distinct actions and inhibit each other, suggesting a mechanism where behaviors are generated by combining multiple motor subroutines through the recruitment of increasingly large DN networks. This framework reconciles the two models of DN control: command-like DNs drive complete behaviors by recruiting additional DN populations, which combine and coordinate multiple motor subroutines.