Computer Use
What vendors mean: AI that can use any software like a person. What it actually means: an agent that looks at a screen and moves the mouse and keyboard — impressive, still slow and error-prone, and a real security surface.
The term
Computer use (sometimes “computer-using agent”) is an AI that operates a graphical interface the way a human does: it takes a screenshot, decides where to click or type, acts, and repeats. It lets AI drive software that has no API — legacy systems, web apps, internal tools — without integration work.
Why it matters to you
This is the promised bridge to work trapped in old, un-integrable systems: the swivel-chair data entry, the report pulled from a portal every Monday. Where an API exists, use the API — it’s faster and more reliable. Computer use earns its keep exactly where no API exists.
Two cautions. It’s still comparatively slow and makes mistakes on unfamiliar screens, and an agent that can click anything a user can click is a genuine security and permissions question.
What that means in practice: Pilot computer use on a bounded, low-stakes, high-repetition task in a system you can’t otherwise integrate — and keep a human approving anything irreversible. Judge it on reliability over a week, not on a demo. The magic wears off; the error rate is what you’ll live with.