I'm not saying it "won't mutate", I'm saying it's too simplistic to expect it's average mutation to be as effective at skirting our cell's defenses as the average influenza bug, that's been honed by 1000's of cycles through the entire population, to be near the sweet point of it's genetic search-space to be close to as many viable variants as possible.
This particular virus hasn't even been through the entire population once, so my argument is that it's extremely unlikely to be genetically near the potential sweet spot I'm talking about. For example if you bolt any old virus together that just spreads, then it's possible and more likely that if you make a couple of substitutions then you get some defective variant that is just hobbled. Not that influenza can't be randomly hobbled to, it's just that a virus that's been through the population 1000 times having to deal with humans over and over is far less likely to be a few substitutions away from gibberish.
Repeated selection would push any genome towards the mean, since genomes near the mean have the highest number of possible offspring who are viable. But this process would require the virus to have gone through the entire population more than once otherwise it's not actually selecting for ones that deal with an "aware" immune system, at all. Remember, influenza is already evolved to deal with humans who are riddled with influenza anti-bodies, then it was honed by 1000s of cycles back and forth through the same populations. That kind of thing really shapes a virus.
This specific coronavirus is currently only optimizing for dealing with "naive" immune systems which lack any relevant or similar anti-bodies at all, so it's not facing any of the selection pressures that have shaped influenza in a multi-generational sense. So, it's strongly optimizing, but for exploiting a resources which is by definition finite - humans with naive immune systems never exposed to this particular bug before. Once those run out, it's going to be very far from the optimal position.
If you consider influenza again, consider that search spaces for things like a genome, which is a string of data, are extremely large so the search space needs to be culled for future searches to be effective. Whatever the influenza genome is today, is shaped by previous information. There could be really crappy variants in the search space, and with that being so vast, merely not being as close to them in terms of information theory would give you a huge advantage for every future mutation's likelihood of turning decent.
Just bolting some random novel virus bits together may well not work so well at all purposes, it might spread well the first time but not be able to generate sufficiently different variants - in the right ways - to make a successful second run.