Google is rolling out Gemini task-automation features in the US and several other countries, connecting the assistant to more than 60 apps. The launch pairs everyday actions such as scheduling and messaging with a hardware-isolated virtual machine architecture for agent tasks, putting the question of what an assistant can do alongside where that work runs.
Google DeepMind Watch analysis
What happened
The features are powered by Google’s Gemini 4 Hydrogen model. Google says the updated assistant can work with integrations for phone calls, calendar scheduling, passwords, shopping and messaging, helping users handle complex errands. Its announcement says the rollout covers the US and several other countries; it does not name those other countries in the supplied details.
Google also says it is introducing a hardware-isolated virtual machine architecture through Google Cloud’s Agent Compute Engine to protect user data during agentic tasks. That is Google’s stated security purpose, not evidence here of an independent assessment of the protection.
Why it matters
An assistant that can reach across calls, calendars, passwords and shopping is more than a chat window with an unusually busy diary. App connections make useful multi-step errands possible, but they also make permissions and the handling of user data central to the product, not fine print for later.
Google’s announcement gives a concrete security detail: tasks will use a hardware-isolated virtual machine architecture. It does not explain the controls users can manage, which apps are available in each country, or how the feature behaves when an action needs confirmation. Those details will matter as much as the app count.
Our read
This is a meaningful move from answering questions to taking actions across services. The 60-plus integrations make the ambition tangible; the isolated execution environment is a welcome part of the pitch. But “connected to your apps” is not, by itself, a user-control policy. Before handing an assistant the keys to a busy digital life, people should be able to see what it can access and when it needs their say-so.
What to watch
- Which countries and apps are included as the rollout expands.
- What permissions users grant, and which actions require confirmation.
- How Google explains the Agent Compute Engine’s isolation and data handling in practice.
Discussion spark: For an assistant that can act across your apps, which matters more: the number of useful integrations or clear controls over what it can access and do?
Sources and evidence
- Put Gemini to work for you (17 November 2026, 17:00 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind