Google rolls out Gemini Spark, a 24/7 AI agent that handles tasks while your devices are off

Google AI Ultra Subscribers Gain Access to Gemini Spark

Google has officially rolled out Gemini Spark to Google AI Ultra subscribers in the US, days after unveiling the feature at Google I/O 20262

2

. The AI agent marks a significant shift in Google's approach to artificial intelligence, moving beyond conversational chatbots toward autonomous systems designed to automate complex tasks across users' digital ecosystem5

5

. Available as a dedicated tab within the Gemini web experience and positioned between Search chats and Daily brief on mobile platforms, Gemini Spark functions as a 24/7 personal agent that operates continuously in the background3

3

.

The service is accessible to subscribers of Google's AI Ultra plan, which now starts at $99.99 per month, positioning it as one of the company's most exclusive AI offerings3

3

. Users can access Spark across Android, iOS, and web platforms, with the agent appearing in the side panel alongside the standard Chat tab on desktop4

4

.

Deep Google Workspace Integration Powers Automation

Gemini Spark distinguishes itself through extensive Google Workspace integration, connecting directly to Gmail, Calendar, Drive, Docs, Sheets, and Slides to handle digital errands that typically require switching between multiple applications2

2

. The agent can schedule meetings, manage invitations, search emails, summarize conversations, create documents, build spreadsheets, generate presentations, and organize files2

2

.

For Gmail specifically, Spark can search across email threads, summarize conversations, compose drafts, reply to existing threads, and organize inboxes using labels4

4

. Calendar functionality includes checking schedules, responding to invitations, scheduling new events, rescheduling meetings, and suggesting open slots that accommodate everyone's availability4

4

.

The system operates on cloud-based virtual machines running Gemini 3.5, enabling tasks to continue processing even when users close their laptops or lock their phones2

2

. This persistent digital assistant model represents a fundamental departure from traditional chatbots that require active user engagement2

2

.

Tasks, Skills, and Schedules Define the Framework

Gemini Spark operates through three core components that work together to execute user requests. Tasks represent the high-level goals or projects users want accomplished4

4

. Schedules determine when the agent should automatically run, either at specific times or in response to triggering events3

3

. Skills provide reusable instructions that teach Spark how to perform specific actions and which tools to use4

4

.

A practical example demonstrates how these elements combine: A user might assign Spark the task of planning a business trip to London, schedule it to notify them when flights are delayed, and deploy both a "Travel Booking" skill and "Gmail Writing" skill simultaneously to rebook accommodations and send confirmations4

4

.

Users can access Skills using @ or / commands, and Spark can automatically apply relevant skills when appropriate4

4

. The beta rollout includes the ability to set up time-based or conditional triggers, such as "Every Monday at 9:00 AM, scan my inbox and review emails from the past week"3

3

.

Real-World Performance Reveals Both Promise and Limitations

Early hands-on testing revealed Gemini Spark's impressive capabilities alongside occasional technical hiccups. When asked to plan a birthday party with minimal detail, the agent scanned emails and calendar entries to locate an actual karaoke bar reservation, then generated a five-page itinerary complete with guest lists, venue rules, nearby dining options, after-party locations, email invites, and theme suggestions1

1

. The agent correctly identified the maximum capacity of 15 people for the venue and pulled exact reservation details including the last four digits of the credit card used for the deposit1

1

.

However, attempts to book dinner reservations exposed current limitations. While Spark operated a remote browser and triggered a six-digit verification code, it ultimately failed to complete the booking despite multiple attempts1

1

. The remote browser capability allows Spark to navigate websites, fill in information, and perform actions without users clicking through each step themselves2

2

.

User Trust Emerges as the Critical Challenge

While technical capability continues advancing, user trust represents the more significant hurdle for widespread adoption. Most people feel comfortable asking AI to summarize emails or answer questions, but granting permission for independent action requires a different level of confidence5

5

. Google addresses this concern by designing Spark to check with users before taking major actions, maintaining user control while operating autonomously3

3

.

The system currently allows up to 15 tasks running simultaneously, requiring users to wait for completion before making additional requests4

4

. Schedules will not execute if 15 tasks are already running, and the service shares the same compute-based usage limits as the broader Gemini platform4

4

.

Google plans to expand Spark's capabilities this summer, including the ability to spend money on users' behalf and availability on the Gemini desktop app3

3

. The company's vision positions Gemini as the coordination layer between users and their applications, fundamentally changing how people interact with their digital tools5

5

.