Google Launches Gemini Spark, Its 24/7 Personal AI Agent

Google used the keynote at its I/O developer conference on May 19 to introduce Gemini Spark, a personal AI agent that runs in the background on a user's phone or laptop, takes action on their behalf, and keeps working even when the device is off. Sundar Pichai called it the start of "the agentic Gemini era." Whether that label sticks is up to the product, but the architecture is real and shipping.

Spark runs on Gemini 3.5 and is built on Google Antigravity, the company's new agent framework. The pitch is straightforward: an assistant that does work in the background instead of waiting for a chat prompt. To do anything useful outside Google's own apps, Spark connects to third-party services over the Model Context Protocol. The opening set of integrations spans Adobe, Asana, Box, Canva, Dropbox, HubSpot, Intuit, Monday, Pandora, Spotify, and Wix, with more coming over the next several months.

Pricing and availability

The rollout is deliberately gated. Trusted testers get it first. The beta opens to U.S. subscribers of Google AI Ultra next week, which Google has bumped to a $100 per month plan aimed at developers, knowledge workers, and creators. Google's stated reason for the cautious launch is safety. The practical reason is also that a 24/7 agent making decisions inside your accounts is a category of product with no real track record yet.

The broader I/O picture

Spark sat inside a larger announcement set. Google also released Gemini 3.5 Flash on the same day, the cheaper sibling to the upcoming 3.5 Pro, and pitched it explicitly as an agent model. Salesforce is plugging it into Agentforce; Xero is using it to run multi-week tax-form workflows. The throughline is consistent. Google is treating chatbots as the previous wave and agents as the next.

None of this is unique to Google. Anthropic's KPMG alliance and the looming OpenAI agent push are after the same market. What is different here is the consumer-facing framing. Spark is one of the first serious attempts to put an autonomous agent in regular users' hands at a fixed monthly price, rather than selling capability to enterprises by the seat. That is the part to watch.