Many engineered and biological systems must trade off performance and energy use, and the brain is no exception. Over long time-scales, evidence suggests that energy use is kept approximately stable by homeostatic firing-rate set-points, resulting in stable average performance levels. On shorter timescales, however, this stability can and should be deviated from, such as during attention when both neural activity and performance often increase (e.g., in predator situations, or when attending to small sensory details). It remains unclear how this fundamental performance-energy trade-off can be achieved in neural circuits of the brain. Here we show that any spiking network with linear readouts is subject to a trade-off between total spike count (energy use), and decoded signal error (performance). However, standard network models have no control over this trade-off as energy use varies, e.g., with input strength, and precision typically remains fixed. To remedy this, we first formulate a cost function which explicitly trades off performance and energy use on different neural and temporal scales. We then derive a spiking network model from this cost function, and show that it adaptively controls target activity levels. The network utilizes several known activity control mechanisms for this --- threshold adaptation and feedback inhibition --- and elucidates their potential function within neural circuits. Finally, using geometric intuition, we demonstrate how these mechanisms in turn regulate coding precision, and thereby performance. Overall, this work addresses a key energy-coding trade-off which is often overlooked in network studies, and unifies work on homeostatic set points, attentional signals, and inhibitory feedback.