The last decade has seen a growing number of connections between statistical physics phase transitions and the theory of computation. Techniques from spin glasses have transformed the understanding of random constraint satisfaction problems while phase transitions play the central role in the efficiency of a wide class of MCMC algorithms. I will survey recent developments in these areas and describe new results on the complexity of counting independent sets.