{"slug": "google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free", "title": "Google's Gemini Tasks Makes Calls, Shops, and Runs 60 Apps for Free", "summary": "Google launched Tasks in Gemini free in the U.S., Japan, Korea, Canada, Australia, and India, a task-running agent built on the new Gemini 4 Hydrogen model that can place outbound phone calls, complete purchases via Google Pay, and work across Gmail, Calendar, Drive, and more than 60 third-party apps. Outbound calling is U.S.-only, excluding Indiana, Louisiana, Minnesota, Montana, and Nebraska, and Google says the agent identifies itself as an AI at the start of each call and lets businesses opt out. Purchases run through the Agent Payments Protocol (AP2) with American Express, Mastercard, and Visa digital tokens, within user-set spending limits, and each user's tasks run in a hardware-isolated VM on Google Cloud Agent Compute Engine.", "body_md": "# Google's Gemini Tasks Makes Calls, Shops, and Runs 60 Apps for Free\n\nGoogle's new agent mode runs on Gemini 4 Hydrogen, makes phone calls, checks out with Google Pay, and goes free across six countries.\n\n- Google launched [Tasks in Gemini](https://blog.google/innovation-and-ai/products/gemini-app/tasks-in-gemini-free-new-features/?utm_source=alphasignal) free in US, Japan, Korea, Canada, Australia, and India.\n- Powered by new Gemini 4 Hydrogen model, optimized for instruction following and tool use.\n- Agent can place phone calls on your behalf, dialing multiple businesses in parallel.\n- Autonomous checkout via Google Pay using Agent Payments Protocol (AP2) with Visa, Mastercard, Amex.\n- Tasks run in hardware-isolated VMs on [Google Cloud Agent Compute Engine](https://cloud.google.com/blog/topics/ai-infrastructure/agent-compute-engine-deep-dive?utm_source=alphasignal) , one per user.\n- New entry points: forward email to gemini@google.com or text via RCS/iMessage.\n\nGoogle is rolling out [Tasks in Gemini](https://blog.google/innovation-and-ai/products/gemini-app/tasks-in-gemini-free-new-features/?utm_source=alphasignal), a task-running agent that can place calls, complete approved purchases, and work across Gmail, Calendar, Drive, and more than 60 third-party apps. Google says the free release will reach hundreds of millions of users and run on its new Gemini 4 Hydrogen model.\n\nThe Gemini home screen now separates quick conversations from longer jobs. Chat handles back-and-forth questions, while Task can continue a project in the background under the user’s direction.\n\n| Initial availability announced by Google |  | \n|---|---|\n| Feature | Availability | \n|---|---|\n| Task mode | Free in the U.S., Japan, Korea, Canada, Australia, and India | \n| Selected advanced tools | Launching first in the U.S. | \n| Outbound calling | U.S. only, excluding Indiana, Louisiana, Minnesota, Montana, and Nebraska | \n\n## Gemini can make the call\n\nOutbound calling allows Gemini to contact several businesses in parallel and act on the result. Josh Woodward, Google’s vice president of Labs, said he asked the agent to find a Friday morning service appointment for a broken minivan. Using Personal Intelligence, which draws on connected Google account data, Gemini identified nearby dealerships, called two of them, and added the appointment to Calendar.\n\nGoogle pairs calling with several safeguards. Gemini identifies itself as an AI at the start of a call, businesses can opt out, and the agent withholds payment or private account information unless the user explicitly authorizes sharing it.\n\n## Checkout runs within user-set limits\n\nTasks can initiate and complete purchases after users set a maximum spend and grant store-level permission. Google says Gemini uses the Agent Payments Protocol, or AP2, while American Express, Mastercard, and Visa provide digital tokens for supported cards. Eligible rewards, cash back, and purchase protections continue to apply.\n\nAP2 provides a framework for authorizing agent-led payments without exposing raw card numbers. Google also plans to integrate the Universal Commerce Protocol, or UCP, which gives merchants a standardized way to support product discovery and checkout by software agents.\n\nReceipts and tracking information appear in Google Wallet, which is expanding to iOS. Before Gemini clicks Buy, it asks for confirmation or relies on preapproval already granted for that action.\n\n## A task can arrive by chat, email, or text\n\nUsers can start work through three channels, allowing the agent to process new instructions without requiring the Gemini app each time:\n\n- **Gemini:** Switch from Chat to Task, describe the job, and let it continue in the background.\n- **Email:** Forward a message to`gemini@google.com` . In Google’s example, Gemini extracts dates from school emails, adds them to a family calendar, and flags a carpool conflict.\n- **Text:** Send instructions and follow-up photos through RCS on Android or iMessage on iOS. One example adds pictured items to an Instacart order.\n\nOnce a task is running, Gemini can create and edit Docs and Sheets, draft and send Gmail messages, and schedule several Calendar events without moving the user into each app.\n\n## Each user gets an isolated sandbox\n\nBecause Tasks can read messages, use payment credentials, and interact with external websites, Google isolates execution through its Agent Compute Engine. Each account receives a hardware-isolated virtual machine that is sandboxed from other users and tested by Google’s security teams.\n\nPrompt injection remains a central risk for browsing agents. A hostile webpage or email can hide instructions that the model may mistake for commands from the user. Google scans retrieved content for those instructions and applies Safe Browsing controls to the links Gemini can open. Hardware isolation limits the potential damage, while hostile content that evades filtering can still influence the model.\n\nGemini pauses before consequential actions, asks clarifying questions when instructions are ambiguous, and supports an `Always allow` setting for selected recurring actions. Google is also inviting researchers to test the system through its [AI reward program](https://bughunters.google.com/about/rules/google-friends/ai-vulnerability-reward-program-rules?utm_source=alphasignal).\n\n## Google’s account graph cuts setup\n\nChatGPT offers Operator, Anthropic supports computer use, and startups continue to release browser agents. Google’s advantage comes from the services already attached to a user’s account. With permission, Gemini can draw on email, calendars, contacts, maps data, saved payment methods, and other account context without requiring users to reconnect each service manually.\n\nFree distribution across six countries lowers the adoption barrier for that integrated system. The same integration increases the consequences of mistaken actions or overly broad permissions, which makes approval flows, isolation, and reliable handoffs central to the product’s performance.\n\n## Three tests for the rollout\n\n1. **Completion on real websites.** Google acknowledges that Gemini may reach the final stage of a workflow and return control to the user for the last click. The company is working with site owners and security providers on ways to distinguish authorized personal agents from malicious bots.\n2. **Merchant adoption.** AP2 and UCP depend on support from payment networks and retailers. Broad adoption would let agents use structured checkout flows instead of navigating merchant pages designed for people.\n3. **Tool-use accuracy.** Google presents Gemini 4 Hydrogen as a model that follows instructions more closely and selects tools more reliably. Hydrogen also powers a[Search upgrade](https://blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade/?utm_source=alphasignal) , suggesting Google plans to use it across products that combine generation, retrieval, and external actions.\n\nThe rollout gives developers a concrete agent architecture to evaluate: per-user isolated execution, tokenized payment credentials, approval gates before irreversible actions, and explicit handoffs when automation fails. Its production value will depend on completion rates, permission clarity, and recovery from mistakes during routine use.", "url": "https://wpnews.pro/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free", "canonical_source": "https://alphasignal.ai/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free", "published_at": "2026-11-17 17:04:16+00:00", "updated_at": "2026-11-17 19:40:38.056139+00:00", "lang": "en", "topics": ["ai-agents", "ai-products", "generative-ai", "large-language-models", "agent-protocols"], "entities": ["Google", "Gemini", "Tasks in Gemini", "Gemini 4 Hydrogen", "Josh Woodward", "Agent Payments Protocol", "Google Cloud Agent Compute Engine", "Google Pay"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free", "markdown": "https://wpnews.pro/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free.md", "text": "https://wpnews.pro/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free.txt", "jsonld": "https://wpnews.pro/news/google-s-gemini-tasks-makes-calls-shops-and-runs-60-apps-for-free.jsonld"}}