Research fields like pervasive computing are showing that the interactions between components in large-scale, mobile, and open systems are highly affected by unpredictability: self-organising techniques are increasingly adopted to conceive infrastructures that can manage such interactions in a robust and adaptive way—with performance being relegated to a secondary ob jective. Accordingly, in this paper we discuss the framework of self-organising coordination: coordination media spread over the network are in charge of managing interactions with each other and with agents solely according to local criteria, making interesting and fruitful global properties of the resulting system appearing by emergence. Differently from the standard setting of coordination, here coordination rules are intrinsically and necessarily stochastic: they need to be probabilistic in order to tackle fairness and to support key fluctuation-like mechanisms, and they also need to be timed in order to support fine balance between the coordination activity and agents behaviour. We show that the TuCSoN coordination infrastructure can be used as a platform for enacting self-organising coordination; this framework is put to test on two cases: the collective sort adaptive tuple distribution mechanism, and an engine implementation for supporting chemical-like coordination reactions.