Ok, I set an AI task going, an as I watched its machinations I realised I had missed out a critical piece of information. I watched as it changed about half a dozen unnecessary forms and made loads of server side changes.
Now, I wanted to interrupt it, but - is it better to let it finish and ask it to undo than stop it and ask it to undo? Or does it make no difference?
For non-Anvil tasks, I usually start a new feature with a clean working tree. That makes going back easy, since I can just use Git to revert to the last commit. So sometimes, even when I see things going in the wrong direction, I let it continue to explore other options. Then I revert and decide whether there’s anything worth keeping or remembering from that failed experiment.
With Anvil apps, it’s different. The constant commit amending makes it almost impossible to go back to a specific point in time.
I usually work locally on my computer using Anvil Sync. When I notice a turn going in the wrong direction, I stop it immediately and ask it to undo what it did. It usually does a pretty good job.
When working online, though, the longer you wait, the greater the chance it will miss some small details when reverting its changes.
I’m not sure whether it’s possible to stop a turn from the online UI (I’m on my phone right now).