I believe AI is rarely used in mainstream apps, but it could be, and I think slowly will be.
If the information an app's AI must learn arises within the app, from user interaction or error, it'd be smart if the program could log that kind of information and then look for patterns in the logs. It could profile users to see ehat tasks are done most often, how many steps are needed. Then when it recognizes that task recurring, it could ask the user if they wanted it to execute a macro that did the following [then it presents then with a list of the steps, allowing them to edit as needed]. Then it executes the 'macro' that it learned from observing the user.
Another use of AI is error detection, not only in the software, but in user error when the software was used inefficiently, redundantly, or improperly. If the software were designed such that it was given a set of models of user tasks (like AI plans), it could observe users in the way they achieve known tasks, and offer suggestions or ask for confirmation that imminent unusual outcomes are intended.
And of course, AI could be used extensively in user interface design, on devices, web sites, or apps. Some of this, like voice recognition, is entering the mainstream of daily use just now. As conversations with apps that can add their own data and models of tasks/concepts/domains develop further, the need for AI inside the app will only grow.
There are a ton of ways that AI could be used in apps. A few of these have started to arise in mobile devices and their apps, usually in fusion of user mobility with external web-based databases (e.g. GPS and maps), but IMO it's been slow.