Hybrid quantum-mechanics/molecular-mechanics (QM/MM) simulations are popular tools for the simulation of extended atomistic systems, in which the atoms in a core region of interest are treated with a QM calculator and the surrounding atoms are treated with an empirical potential. Recently, a number of atomistic machine-learning (ML) tools have emerged that provide functional forms capable of reproducing the output of more expensive electronic-structure calculations; such ML tools are intriguing candidates for the MM calculator in QM/MM schemes. Here, we suggest that these ML potentials provide several natural advantages when employed in such a scheme. In particular, they may allow for newer, simpler QM/MM frameworks while also avoiding the need for extensive training sets to produce the ML potential. The drawbacks of employing ML potentials in QM/MM schemes are also outlined, which are primarily based on the added complexity to the algorithm of training and re-training ML models. Finally, two simple illustrative examples are provided which show the power of adding a retraining step to such "QM/ML" algorithms.