We designed two novel fMRI paradigms to investigate how people self-optimise performance when managing competing demands. We hypothesised that the brain adopts distinct functional states to support different tasks, that switching between them involves a costly process of collapse and reconfiguration of the functional connectome, and that this process is optimised with practice. Accordingly, self-ordered switches (SOS) were associated with transient states of low-connectivity and high-activation. Individuals progressively improved their performance with practice. This learning behaviour was reflected by an ongoing redeployment of the neural resources supporting switching and routine behaviour.Furthermore, those who developed more structured behaviours also scored more points, showed a greater deepening of switching-dysconnectivity and a greater tuning of activity within dorsal frontoparietal cortex to switching events with practice. These results demonstrate that a fundamental property of human neurocognitive systems is concurrent self-optimisation to maximise behavioural outcomes and minimise the use of neural resources.Study 1: Feedback guided SOS. While in the MRI scanner, 16 healthy individuals performed SOS for 20 minutes while being guided by response feedback (Figure 2b-d). Visual discriminations were made on each trial using left or right button presses. Switches between tasks and rules were made using up or down button presses. Response feedback occurred after each trial and consisted of a set of hierarchically organised meters and an overall score.Each task had an overall progress meter, and two subordinal meters corresponding to the discrimination rules. To increment the task meters, both rule meters for that task had to progress, which occurred when a correct response was made for a corresponding trial. To score maximum points, both task meters had to progress. Response errors, or making too many responses on a particular rule or task, resulted in penalties that hindered progress (see