AI testing won't die anytime soon and this proves it
I've been doing freelance mobile dev for about 2 years and testing has always been the part I half ass because I'm usually the only person on the project.
Found this testing agent yesterday from another reddit post that was giving free credits to run your first couple of test cases. I wasn't going to use it on client work without seeing it handle something real first so I tried it on strava an app I use daily and it has a massive user base and actual complex flows. The intent I typed was:
"Open the Strava App. Go to the Record window, select Run as the activity and start recording. Tap on Lap to record one lap, then pause and finish the activity. Finally, name the activity and save it."
That's a multi step flow crossing tab navigation, activity type selection, recording state machine transitions, its not a simple open and tap test.
It opened Strava, found the record tab, verified it was on the right screen, selected run, tapped start, recorded a lap, paused, finished, named the activity and saved it. Each step shows the Al's reasoning like what it sees on screen, what it thinks it should do and what action it takes. It's reading the Ul the way you would if someone handed you a phone and said go record a run on strava without accessibility IDs or xpaths just pure screen level inference.
Now I'm thinking about what this looks like on actual client apps where I'm currently spending 2 to 3 hours manually tapping through flows before every release.