Sorry what point are you making exactly? There are plenty of apps that make API calls to LLMs. I don’t think that’s the problem statement.

> Today on my Linux machine I can prompt an agent and have it run shell commands directly. Until Apple is able to reproduce this on their devices they're behind the curve

You can do that on a Mac today as well:

https://steipete.me/posts/2025/claude-code-is-my-computer

I don't think running shell commands is the achilles heel of Apple's ecosystem. I am quite sure most Apple customers are happy not to be running shell commands; they've done pretty good job so far without any shell access ability (unless you're strictly speaking about MacOS — and here too most of their customers have no doubt never launched Terminal).

The agent part may be important — but that's also the thing most likely to land in a future xOS.

The agent needs a shell. Canned integrations are doomed to be underwhelming.

Apple themselves published a recent paper pointing out that LLMs are very bad at sticking to algorithms (similar to how they're bad at arithmetic.) So if you want an agent to do anything complex it needs to generate a program to do it.