Dozens of mouse pointers clicking the screen together—is this where Computer use should go?

AlexZ posted a demo on X. In the screen, dozens of mouse pointers move and click simultaneously, with the caption 'Computer use needs to evolve into mouse swarms.' original post link

Computer use is about letting AI look at the screen, move the mouse, click and type to operate software. It is general because any software with a UI can be touched; it is slow because every step goes through screenshot, recognition, decision, execution. Duplicating the mouse pointer dozens of times is visually striking, and no wonder some comments say 'So twisted, I love it' and 'This is so cool.' But the comment section quickly split into two camps.

Wilf Lin's reply is direct: no application is designed for multiple mouse pointers. If somewhere truly needs multiple pointers for efficiency, it should first be made usable in the command line. This judgment isn't gentle, but it hits the GUI's underlying assumption. Window focus, clipboard, input methods, popups, file locks, and login state are mostly designed for a single user and single focus. Multiple pointers are not the same as adding threads; instead, they blur state ownership. PG puts it more bluntly: this will add a million ways for existing apps to break.

bays wong proposes another direction: if there are enough tokens, spin up multiple agents, each agent gets one mouse, and do whatever. AlexZ replies, then it would be very slow. There is a real issue here: multi-agent parallel GUI operation has costs in scheduling, isolation, conflict handling, and result verification. LIA jokes that 90% of the pointers in the video aren't doing anything at all. This joke also reminds people that visual clustering and task parallelism are two different things.

Some also ask, if it all needs to be this fast, why not use code. thirk says that strictly speaking, this is slower and worse than writing code; Jim asks how it's done, since most of the time coding is faster than using a mouse. Dr. Dapeng directly says it looks lively but is useless, and can work for making short videos to farm traffic. These voices aren't pleasant, but they are closer to engineering reality than 'cool.' The value of Computer use lies in generality, in scenarios where you can't hook APIs, there is no command line, and you can only click the UI. Its shortcomings are also obvious: slow, fragile, and hard to observe.

Aurumo says that if it's like this, why still need a mouse. Michael prefers AI-native operations and thinks computer use is still too unsuitable for AI. SeanG's question goes further: when speed gets so fast that humans can't keep up, can the frontend also be removed? These questions are more worth looking at than a few extra cursors. If mouse swarms are just a demo, they provide a visual metaphor. To actually make a product, first answer who is responsible for the clipboard, who handles CAPTCHAs, whose command wins when multiple agents grab the same button, and how to roll back after an error. It looks satisfying, but the first engineering question is who owns the clipboard.