Google began rolling out a computer-use mode for its Gemini Mac app on November 19, letting the assistant operate local files, desktop software, and websites for the user. Michael Friedman, a group product manager for the Gemini app, announced it on Google’s blog.

The agent works in a separate window out of sight, with a preview inside the chat, so the user keeps the cursor and screen. Google says users pick which apps Gemini may touch, can follow each step, stop it mid-task, and sign off on important actions first. Its example jobs include batch color-grading photos, pulling figures from PDFs into shipping quote forms, assembling a Slides deck from dashboard metrics, and preparing resale listings from a folder of photos.

Google’s post offers no accuracy figures or failure rates, and it does not define which actions count as important enough to need approval. That definition decides how much of the chore stays with the human. The feature is limited to English and U.S. users for now.

Teams weighing it should hand it one low-stakes recurring task first and count how often the approval prompts fire.

Google, “Let Gemini tackle repetitive chores on your Mac,” blog.google, November 19, 2026.