Definition
Plain language
An AI that operates a computer the way a person does — clicking, typing, and navigating apps to finish tasks.
As stated in the literature
An agent, typically built on a vision-language model, that perceives a desktop or browser via screenshots, DOM, or accessibility tree and acts through mouse and keyboard operations; evaluated on benchmarks like OSWorld and CUA-World.
Also called: computer-use agents, CUA, CUAs
Why it matters: It lets AI handle real software tasks people do every day, rather than only answering questions in a chat box.
For example, you could tell such an agent to open a spreadsheet, find last month's expenses, and email the total to your boss, and it would click and type its way through each step.
Heard on the show
“The standard desktop benchmark for computer-use agents is called OSWorld.”Episode 017 — When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers