There seems to be a lot of experience behind this presentation. But I wonder, how does all this G-testing compare to Thompson sampling? (Which was not mentioned

There seems to be a lot of experience behind this presentation. But I wonder, how does all this G-testing compare to Thompson sampling? (Which was not mentioned at all in the presentation.) For example, is there any drawback with adapting the A/B probabilities on-the-fly, instead of having a hard decision point?