One can still write tests that are flaky - some of the challenges that Selenium et al. experience are still present:

1. Selectors still need to be specific enough `.button` is great until you add another button.

2. Selectors can be _too_ specific in weird ways. Recently, I encountered a bug where reformatting my markup broke a test that detected if an element was present by checking for its contained text. Turns out the newlines that I inserted between the opening and closing tags were significant for the test. Thankfully, it's all JS, so one can appropriate work around.

3. For SPAs, it's harder to convey that a page/view load has completed. So, if I click a button and the entire viewport updates (potentially downloading a JS bundle in tandem), I either need to insert timeouts of arbitrary lengths, or explicitly call out to the the provided network API hooks. This can be annoying because now I not only need to worry about what I _can_ see when I navigate through my app, but also what I can't.

It's far better than what has come before, but it's not a silver bullet.