Where the workflow shifted
Google Gemini Computer Use puts screenshot observation, browser/mobile/desktop environments, suggested UI actions, safety policies, and prompt-injection detection into one workflow.
A browser agent needs to know what it can see, what it may click, and when it should stop for human approval.
Tool names are not outcomes
The signal matters when it changes how a team ships, reviews, or recovers work, not when it only names another tool.
Check permissions and failure
- Add target page, allowed actions, blocked actions, screenshot cadence, confirmation points, and exit criteria to each browser task
- Keep the test narrow: one low-risk task or tool entry before connecting permissions, logs, failure handling, and human takeover to production
What still needs proof
Without boundaries, an agent can treat prompt injection, wrong pages, or sensitive operations as normal steps. Keep the original source open so the announcement, the evidence, and this site's interpretation stay separate.