Concept
Give an agent a mouse, and it can use anything — including everything.
The agent capability of operating software through its GUI — reading the screen, moving the mouse, typing — rather than through APIs. The most general agent skill, and the one with the widest risk surface.
Review
Not yet scored
No ClawScore review has been published yet.
Coverage
2 stories
Latest: Jul 10, 2026
Sources
2 links
Official and reference links for deeper verification.
Discovery
Noindex
Readable here, but withheld from sitemap until coverage deepens.
Computer use is the capability for an agent to operate a computer the way a person does: it reads the screen, moves the mouse, and types, instead of calling an API. A browser agent is the common special case, driving a web browser to navigate and fill forms. The appeal is generality, since most software lacks a clean API but almost all of it has a GUI, so a GUI-driving agent can in principle use anything with no per-app integration.
That reach is also the risk. An agent with mouse, keyboard, and screen access has the broadest blast radius in agentics: it can act in any open app, see secrets it types, and take irreversible actions one click at a time, on a brittle substrate where a moved button becomes a misclick. On-screen content is untrusted input, so it is squarely inside the prompt-injection surface. Anthropic and OpenAI both shipped computer-using agents, which is what moved this from research demo to something operators have to threat-model (run it isolated, scope it, and gate the irreversible actions).
Meta's Muse Spark 1.1 is the first Spark model with an API, and it leads with tool calling and computer use, not benchmarks. The tell isn't the model. It's that Meta wants you building agents that act.
Agents are graduating from API calls to direct computer control. A new infrastructure layer is forming underneath them, and it's quietly rewriting what the word 'agent' means.
