Google、パーソナルAIエージェント「Gemini Spark」を日本でリリース

On July 16, Google began offering Gemini Spark, its personal AI agent, to users in Japan. The service is available as a beta exclusively to Ultra subscribers (¥14,500/month and up). Gemini Spark is powered by the Gemini 3.5 AI model and runs on Google's Antigravity agent development platform.

The core appeal of Gemini Spark is its ability to work autonomously around the clock. Unlike traditional assistants that require active user input, Spark executes tasks continuously in the cloud, processing workflows even when a user's device is powered off. It integrates deeply with Google Workspace services—Gmail, Google Calendar, Google Docs—and third-party applications via the Model Context Protocol (MCP), allowing users to chain actions across multiple tools via a single text instruction. For example, a user can ask Spark to gather RSVP responses from a circle's group email, consolidate attendance data into a spreadsheet, send follow-up confirmations based on responses, add meeting times to the calendar, and send notifications—all without manually switching between apps.

Spark operates on user-defined schedules or automatic triggers and executes tasks on Google Cloud virtual machines. Before taking high-impact actions like sending emails, it requests user confirmation to maintain control. The system can memorize recurring task workflows as "skills" and reuse them, enabling automation of repetitive processes over time.

Google's Gemini vice president, Josh Woodward, signaled on X that Pro plan members (¥2,900/month) should expect access updates, indicating the company plans to expand availability beyond Ultra subscribers in the future. Gemini Spark was first announced in May at Google I/O 2026 and rolled out to U.S. Ultra plan subscribers as a beta. The Japan launch represents the first geographic expansion outside the United States and adds support for Japanese and other languages.

Gemini Spark represents Google's entry into the autonomous AI agent market, built on the company's Gemini 3.5 model and Google Antigravity development platform. The service addresses a specific pain point: users often repeat routine administrative tasks across multiple cloud applications. By running 24/7 in the cloud, Spark can execute these workflows asynchronously without requiring the user's device to be active, a technical distinction from earlier assistant models that depend on user-initiated requests.

The Japan launch is strategically significant as the first international expansion beyond the U.S., following the May announcement at Google I/O 2026. The company has carefully segmented access: Ultra tier (highest-paid) gets the beta first, while the public hint that Pro tier members will eventually receive access suggests a deliberate pricing ladder to drive conversions. This tiering aligns with Google's broader Gemini monetization strategy, where each capability roll-out is tied to a specific subscription level.

The integration depth—spanning Gmail, Calendar, Docs, and via custom MCPs, virtually any third-party app—gives Spark a broad utility surface compared to point assistants. Users can define repeatable workflows as "skills" and have the agent re-execute them, reducing friction for knowledge workers who perform similar multi-app orchestrations daily.

No comments yet. Be the first to share your thoughts!