Computer Use

Computer use is the capability of an AI agent to operate a computer like a person: it reads the screen through screenshots, moves the cursor, clicks, types and scrolls in any software, including programs that expose no API.

Agents & automationcomputer useGUI agentscreen automation
§ 01DefinitionComputer Use

No jargon, no math

In plain words, what is it?

The agent works in a loop of screenshots. It captures the screen, the model reads the image the way you do, decides the next interface action (click this, type that, scroll), and a controller executes it through the operating system's accessibility or input layer. Then a new screenshot confirms what happened and the loop restarts. This is how you automate the software nobody can automate: the 2009 ERP, the customs portal, the supplier's booking site, none of which have an API. It breaks three ways: brittle layouts, because a moved button breaks the flow; latency, because every step is a round-trip of screenshot, model, action; and risk, because an agent that can click “save” can just as easily click “send”, so permissions and logs are not optional.

§ 02How it worksScreenshot, decide, click, verify, repeat
  1. 01

    See. Capture the screen and read it like a person.

  2. 02

    Decide. Choose the next interface action: click, type, scroll.

  3. 03

    Act. Execute it through the input layer.

  4. 04

    Verify. A new screenshot confirms the action landed.

  5. 05

    Repeat or stop. Loop until the task is done or a permission stops it.

§ 04

Want this running in your stack?

We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.