It's easiest if you use the thumbs up / down button if you have a particularly bad (or good) experience. I browse all traces with feedback daily and make notes for common problems or errors.

I actually just stated working on the next big update and all the feedback helps a lot!

This is actually an intentional way how it learns about limitations. It's impossible to tell the AI upfront everything you can and can't do in the product. So it speculates to some degree and then 'learns by doing'.

Not unlike a real user who first has to try and figure out if it works or not.

Yeah, ideally the assistant should not first suggest something it cannot do. But it also gave you a correct answer at the end - custom code really does not yet support files.

Isn't that how it'd work in most platforms?

I keep seeing users set up workflows in relay that load the same data again after an update. I assume that must be some learned old habit.
Completely unnecessary as the system tracks the state of all data to prevent that category of bugs.

Always feel like the UI has a permanently enabled debugger mode that adds all the unnecessary complexity.
We spent a lot of time thinking about how we can make simple things simple in relay.app. Give it a try if that sounds interesting.
(Feedback form should also work)

Tried to combine all my favorite things from 20 years of working with APIs :) Always happy to receive questions / feedback, here or per mail.

The only automation platform that knows the Google APIs from both sides :)

I think you can build a mute equivalent with a filter and a user label. The logic for the real mute label isn't much more complicated than 'Skip inbox'.