Here is the formatted article.
Google is advancing its artificial intelligence capabilities across the Gemini platform, developing and launching several new agentic features that enable the assistant to perform multi-step tasks, automate workflows, and operate proactively.
Availability and Access
Gemini Spark (agentic AI assistant) has launched in the United States for Google AI Ultra subscribers ($100-$200/month). The feature is available on web, Android, and iOS platforms, appearing as a "Spark" tab on desktop and a "Beta" label on mobile devices.
Previous access expansion included:
- United States: Google AI Pro subscribers ($20/month) gained access to earlier Spark capabilities.
- Global: Google AI Ultra subscribers, excluding the European Economic Area, Switzerland, the UK, and Nigeria.
Free users remain excluded from Spark features.
Current Features
Gemini Spark Capabilities
The Spark agent integrates with Google Workspace applications and can perform the following functions:
Calendar:
- Check calendar, RSVP, schedule events
- Update meeting times, change locations, cancel events
- Suggest open slots for scheduling
Drive, Docs, Sheets, Slides:
- Search files, read contents, check metadata, view recent documents
- Create and edit Google Docs
- Build and format Google Sheets with data, tables, and formulas
- Create Google Slides presentations from prompts
- Rename or delete files
Gmail:
- Search and summarize email threads
- Compose drafts, forward, reply
- Organize inbox with labels
Additional Use Cases:
- Declutter inbox by summarizing and archiving newsletters
- Generate meeting briefs before scheduled events
- Create custom news digests following topics over time
- Conduct deep research on topics with cited sources
Core Components
- Task: A high-level goal or project to accomplish
- Schedule: Instructions for automatic background operation at specific times or triggered by events
- Skill: Reusable instructions and context templates accessed using @ or / commands
Gemini Spark can control a remote browser to navigate websites and add items to carts, and can access a remote computer for code execution. The system stores browser session data, including login information. Users can clear stored data or disable connected services via settings.
System Limits
- Maximum 15 tasks running simultaneously
- Tasks must complete before new requests can be initiated
- Same compute-based usage limits as other Gemini features
Features in Development
Analysis of Google app beta versions (17.2 through 17.23) has identified multiple features under development. These findings are based on decompiled code and may not reflect final shipping products.
"Remy" Personal Agent
Internal documents describe "Remy" as a 24/7 personal agent for work, school, and daily life, powered by Gemini. The agent is designed to:
- Monitor for important events
- Handle complex tasks proactively
- Learn user preferences over time
- Take actions on the web and with connected apps, including communication, document sharing, and purchases
The feature remains experimental, with warnings about potential mistakes and data exposure. Users are advised to supervise tasks.
Gemini Live Upgrades
Beta versions indicate upgrades for Gemini Live including:
- Live Thinking Mode: A version that takes additional processing time for detailed responses
- Live Experimental Features: Including multimodal memory, improved noise handling, and personalized results from integrated Google apps
Proactive Assistance
Code references in version 17.18 beta describe "Proactive Assistance," a feature that would provide personalized suggestions based on user activity. Data sources include:
- On-screen content
- Notifications
- Apps (Contacts, Messages, Gmail, Calendar)
Processing occurs entirely on-device in an encrypted environment. According to Google's privacy statements, data is not used for generative AI training or human review. The feature can be toggled on or off, with users selecting which apps contribute data.
UI Automation and AppFunctions
Google has detailed two frameworks for AI automation on Android:
AppFunctions (Android 16): Allows applications to expose specific functions that AI assistants can access and execute directly on the device. Examples include:
- Task management integration
- Media playlist creation
- Cross-app workflows combining email search with shopping list population
- Calendar event creation
UI Automation: A framework enabling AI agents to execute tasks on installed applications without requiring code changes from developers. Android 17 is expected to broaden these capabilities.
Screen Automation (Codename "Bonobo")
A feature identified in version 17.4 beta aims to allow Gemini to assist with tasks such as placing orders or booking rides within specific applications. Android 16 QPR3 is reportedly laying groundwork for this integration.
Users retain the ability to interrupt automation and assume manual control. Privacy protocols indicate screenshots are subject to review by trained personnel when "Keep Activity" is enabled.
Computer Use Integration
The anticipated Gemini Agent functionality is expected to arrive on Android as part of a "Computer Use" integration. Gemini Spark can control Chrome browser and access device files, but reportedly cannot control the entire computer.
Avatar and Likeness Features
Development is underway for "Gemini Avatars," which would allow users to create 3D representations for integration into generative content. The creation process involves using a phone's camera for a head scan. Android XR previously introduced "Likenesses" — photo-realistic digital representations for video calls using XR headsets.
Desktop Interface
Information from beta version 17.5 indicates Gemini integration into an Android desktop interface. Users would access Gemini via a status bar icon or keyboard shortcut (Google Key + Spacebar). The interface is anticipated to launch as an overlay, potentially similar to phone implementation or the Chrome side panel.
Legacy Voice Removal
Code analysis indicates Google is discontinuing 10 existing Gemini voice options (named Ursa, Nova, Vega, Pegasus, Orion, Eclipse, Capella, Lyra, Dipper, and Orbit). A replacement has not been announced.
Experimental Labs Features
The Gemini web application has introduced an "Experimental Labs" section in the Tools menu, organized separately from standard features:
Experimental Features (Labs badge):
- Agent (available with AI Ultra)
- Dynamic view/Visual layout (available to all users)
- Personal Intelligence (available to paid subscribers)
A "Personalize chat when helpful" toggle allows users to manage whether Gemini uses Connected apps for the current conversation. This setting resets when a new chat begins.
Safety, Privacy, and User Controls
Users are accountable for actions performed on their behalf. Google has issued advisories regarding potential for errors in agentic features.
Recommendations include:
- Supervising tasks and not relying on agent for professional knowledge
- Avoiding input of login or payment information into Gemini chats
- Not using screen automation for emergencies or tasks involving sensitive data
The system is designed to request permission before taking sensitive actions but may occasionally share information or make purchases without explicit confirmation in certain scenarios. Users can clear stored data and disable connected services via settings.
Competitive Context
Competitors including OpenAI and Anthropic are developing agent-style systems that actively browse the web and perform tasks with minimal user input.