Skip to content

AI interacting with screens: what it means for the office

Date of the official announcement: 22 October 2024 · Published: 5 October 2026 · Reading time: 3 min · Official source: Anthropic

Professionals analysing an automated administrative process on screen
In short
  • Anthropic introduced the experimental 'Computer Use' capability in Claude 3.5 Sonnet to interact with graphical interfaces like a human user.
  • It makes it possible to automate repetitive tasks in desktop programs or closed portals without complex APIs.
  • It requires strict security measures, isolated environments and human review of every critical action or any action involving sensitive data.

What screen interaction is and how it works

Anthropic presented significant advances in the ability of artificial intelligence models to interact directly with user interfaces on personal computers. With this feature, experimentally named Computer Use, the model interprets screenshots, moves the cursor, clicks buttons and types text into form fields as a human operator would.

The technical proposal aims to let agents carry out sequences of administrative steps across everyday applications without having to program complex API integrations for every program involved.

“The ability to interact with software through the screen opens ways to connect closed systems, but human supervision remains essential for any action that affects data or finances.”

Practical application in offices and professional firms

For companies and professional firms that work every day with local desktop programs or closed government portals (such as invoicing platforms, appointment systems or notary and tax systems), this technology opens the possibility of cutting hours of manual data entry.

Tasks with clear rules —such as downloading files from a portal, checking data in a spreadsheet and consolidating a weekly report— can be structured through assistants that follow step-by-step instructions under constant supervision.

Essential governance and security limits

This advance calls for a serious governance warning: handing control of the mouse to an AI model in processes involving money transactions, payments or sensitive medical and personal information requires isolated environments and strict human review at every critical step.

Our take at MR Mentoría: practical automation isn't about replacing people, but about freeing them from mechanical copy and paste. Trying these tools in controlled environments, with limited permissions and clear usage policies, is the sensible way to increase efficiency without compromising the business's security.

Published by MR Mentoría. If you want to set up safe automation workflows for your team, write to us: WhatsApp o [email protected].