A customer requested the ability to click a button to automatically add all the masks for a flaky screenshot test. This would help in managing and fixing screenshot test failures by automating the masking process.
A few weeks ago I posted that Screenshotbot is now a no-AI company (https://lnkd.in/eXCZ2kvu) I wanted to give an example of the practical implications of this. Today I intentionally chose not to implement a feature request by a customer. Here's the feature request: they wanted the ability to click a button and automatically add all the masks for a flaky screenshot test (https://lnkd.in/eXGuWDVU) Seems like a straightforward feature. Doesn't need too much architectural changes, since the change is quite isolated. There's definitely some algorithmic challenges, but seems solvable. (Maybe k-means with a cost associated with over-selected pixels?) It's probably a well studied problem, and Claude could probably implement it in one go. Initially, my gut reaction was that this was a good feature. But the lack of Claude slowed me down. And that gave me time to realize that we actually don't encourage excessive use of masks. It's a great tool to quickly reduce noise for your team from flaky screenshot tests, but it's hard to maintain: if the layout shifts, you have to move all the masks; or if you clone a screenshot test, you also have to clone the masks. Adding the feature would just add a new surface that I would have to maintain, with limited utility, and in particular for a behavior that we actively discourage. Sometimes not writing code is the right answer.