In June, Google integrated computer use into Gemini 3.5 Flash, letting developers build agents that can "see, reason and take action across desktop, mobile and browser environments", with examples such as continuous software testing and knowledge work. Agents that operate interfaces the way a person does open up automation for systems that never had an API.
Where Computer Use Helps
Many businesses still rely on older portals, supplier websites and internal tools with no integration options. An agent that can fill in forms, copy information between systems or run through a test script can take over that repetitive work. For SaaS products, it also enables features like automated onboarding checks or end-to-end testing that behaves like a real user.
Where It Needs Care
Clicking through screens is slower and less predictable than calling an API. Layouts change, pop-ups appear and pages load at different speeds. An agent acting inside real accounts can also make real mistakes, such as submitting the wrong form or changing the wrong record.
Guardrails We Recommend
- Prefer an API whenever one exists, and use computer use only for the gaps.
- Give agents their own accounts with the least access they need.
- Require human approval for anything irreversible, such as payments, deletions or sending messages.
- Log every action with screenshots so you can review what happened.
- Start with read-only tasks, then expand once the agent proves reliable.
The Takeaway
Computer use is a powerful bridge to systems without integrations. Treated like a new team member, with limited access, supervision and a clear record of its work, it can remove hours of manual effort safely.