GlossaryAgentsIntermediate

Computer Use

Also written: AI computer interaction, AI system control, Computer-use agents, AI agents for computer use, Software agents

AI interacting with software and operating systems like a human user.

In plain terms

Computer use means an AI can operate software on a computer. Think of it like an AI using a mouse and keyboard to open applications, click buttons, and type information.

For example, an AI with computer use capabilities could open a web browser, navigate to a specific website, fill out a form, and then save the information to a document.

The expert version

Computer use refers to an AI's ability to interact with and control computer systems, including graphical user interfaces (GUIs) and application programming interfaces (APIs). This capability allows AI agents to perform tasks that typically require human interaction with software, such as navigating operating systems, launching applications, and manipulating data within them.

This functionality is often achieved through a combination of vision models to interpret the screen, natural language understanding to process instructions, and action execution modules that translate AI decisions into user inputs or API calls. Tradeoffs include the complexity of simulating human interaction, potential security risks, and the computational cost of real-time environment interaction.

Why it matters now

Frontier AI labs are focusing on computer use because AI agents need to interact with live software to be truly useful for work. Recent research and product announcements show a push towards AI that can operate applications, perform multi-step tasks, and be safely overseen within digital environments. This development is key to moving AI from generating text to actively performing actions.

Where it shows up 6 items, newest first
How each company uses it
Anthropic Claude
Anthropic is advancing Claude's computer use capabilities through acquisitions and research into AI agent oversight and safety.
Microsoft Copilot
Microsoft is developing deep, evolving environments to train computer-use agents for better real-world performance.
OpenAI ChatGPT
OpenAI is integrating advanced agent capabilities into its latest models, enhancing their ability to perform work tasks.
Related terms

AI agentsRobotic Process AutomationHuman-Computer InteractionTool UseTask AutomationOrchestration

Often confused with
[object Object]
[object Object]
[object Object]
Origin

The concept of AI controlling computers has roots in early AI research and robotics, evolving with the development of more sophisticated AI agents and automation tools. The term gained prominence with the rise of AI agents capable of interacting with GUIs.

Try this about 5 minutes

Try asking an AI assistant like ChatGPT, Copilot, or Claude to perform a multi-step task that involves opening an application, finding information, and saving it. For instance, ask it to 'find the current weather for London and write it in a text file.' Observe how it navigates the necessary steps.

Try it: Playbook entries that use thisPlaybook →

The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.

Questions people ask

Can AI really use a computer like a person?
Yes, advanced AI can now simulate human interaction with computers by opening applications, clicking buttons, and typing, though the complexity of tasks it can handle is still developing.
What's the difference between AI using software and just running a script?
Scripts are pre-programmed sequences of commands, while AI computer use involves understanding context, making decisions, and adapting to changes within the software environment, much like a human would.

First seen on Subvolts Feb 25, 2026. Updated Oct 19, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.