I have an old Amiga game disk image (*.adf) that I want to extract some files from. I asked Claude to do it, and it tried to use 'unadf', couldn't find it, and proceeded to just write a Python program to extract files from an Amiga disk image.

It worked. Wild.

iPadOS 26 gestures at the top of the screen aren’t clicking for me.

When a window is at the top of the screen, if I want to move it, about half the time my swipe is interpreted as showing the menu bar. There’s no gesture to remove the menu bar so it’s a 5 second penalty while I wait for it to hide.

And how do I scroll to the top of a table view in a windowed app?

You can't effectively use current LLMs to write code you couldn't write yourself.

The problems you create won't be immediately evident, but they are there.

People seem surprised that getting an LLM to write your essay requires less engagement with the subject matter and less "thinking". Not sure why, that seems pretty obvious.

If I hire a contractor to build a deck, I'm not gaining much in the way of construction skills. But I am getting my deck built.

Context matters. In an educational context, students should not be using LLMs. But if the context is more like a consultation then an LLM may be appropriate.

If Messages were just a database, and the Messages "App" were English text instructions on how to display the messages and let users interact with them, a sufficiently advanced LLM could give you the functionality of Messages.

Imagine a tool call that did Observation style tracking of the data that the tools accessed in the system, so when a new message arrives, the UI could just update.

The GUI for LLMs hasn't been invented yet. We can provide an LLM with a graphical environment and a set of frameworks and conventions that the LLM can use to interact with the user in a way that feels familiar and friendly.

Apple is in an amazingly good place to do this, but I don't know if they have the vision. It's a pretty radical change.

A fork of iOS that was "llm-first" on a new hardware device (to completely break compatibility and expectations) could be revolutionary.