I spent today building the marketing sections for a batch of my Amazon listings. The kind with a header, a three-step layout, a little tools list under the cover. Boring, repetitive, exactly the sort of thing you hand to a machine. So I pointed a browser bot at Amazon's editor and let it work.
Most of it went fine, just not how I expected. The editor is built out of web components with everything buried in shadow DOM, so half my usual tricks did nothing. The text boxes turned out to be Draft.js editors, the same rich-text widget a lot of sites use, and they ignore you if you try to set their value directly. They only react to real keystrokes. Click the field, type like a person, and it takes. I learned that the slow way, by watching three successful fills leave the box empty.
Then I hit the images. The upload control is a file input hidden inside the shadow DOM, and clicking the dropzone pops the native macOS file picker. That picker isn't a web page. It's the operating system, and no browser automation touches it. Mine, yours, nobody's. Hard stop. I could have burned two hours faking a drag-and-drop event into a component that was built on purpose to not trust one.
I didn't. I dropped the approved images into a folder named for each book, left every draft open to the right spot, and wrote down the exact step: drag these four in, hit save. Sixty seconds of human work. The point of the job was finished drafts, not a hands-off robot. Those are different goals, and mixing them up is how you lose a day.
This is the trap with automation. The first 95 percent of a task is usually easy to give to a machine. The last 5 percent is where the real mess lives: a native dialog, a captcha, a human approval, some control an engineer deliberately locked down. That 5 percent eats more time than everything else combined if you let it, and you end up with a brittle script built to dodge one minute of clicking.
So I run a quick check before I automate anything now. How often does this actually happen, and how long does it take by hand. A one-minute job I do once a year doesn't need a script. It needs me to do it and move on. Save the automation for the thing you'll run a thousand times, and don't be too proud to drag a file yourself.