I think testing via browser automation is fantastic, API-driven web browsers are just amazing tools, but it's also the source of a lot of the flakiness because of the inherent difficulties determining when a page is completely in a state that it can do what you're expecting of it. Playwright improves a lot on this over Puppeteer but it's imperfect, I often end up with a wait for a selector, load state, or function evaluating when it's actually okay to proceed and sometimes I'll use a combination to really make sure, because what it's really waiting for is not just achieving some state but what actually happens after it does.