Google Launches Gemini Spark: A 24/7 Personal AI Agent in Beta

30-Second Summary

Google has begun rolling out Gemini Spark, a 24/7 personal AI agent that runs on Google Cloud virtual machines and continues executing tasks when the user's device is off. Beta access opened the week of 25 May 2026 for US Google AI Ultra subscribers and select business users. Spark is built on Gemini 3.5 Flash and Google's Antigravity agent harness, structured around Tasks, Skills, and Schedules, and integrates natively with Workspace plus third-party apps through the Model Context Protocol. For operators, always-on personal agents have moved from enterprise pilot to consumer subscription pricing.

At a Glance

• Topic: Agent Systems
• Company: Google
• Date: 25 May 2026
• Announcement: Beta access to Gemini Spark opens for US Google AI Ultra subscribers and select business users
• What Changed: A 24/7 personal AI agent that runs in Google Cloud, takes action across Workspace and third-party apps, and continues working when the user's device is off
• Why It Matters: Always-on personal AI agents are now packaged as a consumer subscription, not just an enterprise platform
• Who Should Care: Founders, knowledge workers, operations leaders, and anyone running recurring research, scheduling, inbox, or document workflows

Key Facts

• Company: Google
• Launch Date: Announced at Google I/O 2026 on 20 May 2026, beta rollout began the week of 25 May 2026
• What Changed: Gemini Spark, a 24/7 personal AI agent running on Google Cloud virtual machines, launched in beta with native Workspace integration and MCP support for Canva, OpenTable, and Instacart at launch
• Who It Affects: US Google AI Ultra subscribers over 18 and select business users, with broader rollout expected
• Primary Source: Google I/O 2026 keynote and the Gemini Spark product page at gemini.google/overview/agent/spark

What Happened

Google unveiled Gemini Spark at I/O 2026 on 20 May, positioning it as a personal AI agent that runs continuously rather than ending each session when a chat tab closes. Beta access began rolling out the week of 25 May to US Google AI Ultra subscribers and a set of approved business users.

Spark is built on Gemini 3.5 Flash and Google's Antigravity agent harness. It runs on Google Cloud virtual machines, which means it can continue executing multi-step tasks even when the user's phone and laptop are off. The product is structured around three core concepts:

• Tasks are multi-step jobs assigned in natural language, such as researching contractors or tracking job listings across the web
• Skills are reusable instruction sets a user builds over time, allowing Spark to repeat complex workflows without re-specification
• Schedules trigger recurring actions, such as scanning an inbox every Monday morning and producing a prioritised to-do list with focus time blocked on the calendar

At launch, Spark integrates natively with Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube, and Google Maps, with each connection turned off by default. It also supports third-party tools through the Model Context Protocol, with Canva, OpenTable, and Instacart confirmed as launch partners. Google has stated that Spark is designed to check in with the user before taking major actions, preserving human oversight while operating autonomously on smaller steps.

Why It Matters

• Always-on personal AI agents are now a consumer subscription product, not just an enterprise pilot
• Tasks running on Google Cloud virtual machines decouple agent execution from the user's device, removing battery, network, and uptime constraints
• Skills and Schedules turn AI from a reactive chat interface into a proactive operations layer
• MCP support at launch means Spark plugs into a growing ecosystem of third-party tools without bespoke integrations
• Google's combined data graph across Workspace, Search history, and Maps gives Spark a context advantage that pure-play agent vendors cannot match
• The presence of business users in the beta cohort signals that Google will pursue both consumer and team-tier monetisation in parallel

The David and Goliath View

Until this week, always-on AI agents were enterprise infrastructure. Spark turns them into a packaged subscription. That changes the procurement question for operators of lean organisations. The barrier is no longer cost or capability. It is workflow design.

Most operators still treat AI like search. Type a question, get an answer, close the tab. Spark rewards a different posture. The teams that get value from it will be the ones who can articulate the recurring, low-judgement work worth scheduling overnight. Inbox triage. Pipeline reporting. Competitive monitoring. Calendar coordination. These are not new problems. They are the work that gets crowded out by reactive tasks every day.

The honest test is simple. List the recurring workflows your team does manually. If you can describe one in three sentences, an agent like Spark can probably run it. Start there. Do not try to automate strategy. Automate the work that drains the hours before strategy can begin.

Where This Fits in the AI Stack

AI Growth Engine: Spark can execute the recurring revenue operations work that typically consumes sales and marketing time, including prospect research, pipeline updates, and competitive scanning across Workspace and the web. Sales leaders should treat it as an always-on research analyst, not a replacement for human judgement.

Employee Amplification Systems: Spark is a direct fit. Personal Schedules and reusable Skills become a knowledge worker's automation layer for inbox triage, document drafting, calendar coordination, and recurring research. The agent runs overnight rather than during working hours, returning prepared work to operators each morning.

Questions Operators Are Asking

Is Spark enterprise-ready or a consumer product? At launch it is primarily a consumer subscription tied to individual Google AI Ultra accounts, with select business users included. Enterprise governance features such as shared accounts, audit trails, and admin controls have not been formally announced. Pilot with a small group before deploying across a team.

How does Spark compare to OpenAI's Operator or Anthropic's computer use? All three are agent platforms that can take action across applications. Spark's advantage is native Workspace integration and the data graph behind it. OpenAI's Operator focuses on browser-based execution. Anthropic's computer use is positioned as a developer primitive rather than a packaged consumer product.

What about sensitive data? Spark runs on Google Cloud virtual machines, which means data processed by Spark passes through Google's infrastructure. For regulated workflows, evaluate Google's data processing terms and consider whether agent execution on a third-party cloud is acceptable before assigning sensitive Tasks.

Where should an operator start? With a single, well-defined Schedule. A morning inbox triage that surfaces the top five emails requiring a response, with a draft reply attached, is a high-value, low-risk starting point. Validate the output for two weeks before expanding to additional workflows.

Citable Summary

What happened: On 25 May 2026, Google began rolling out beta access to Gemini Spark, a 24/7 personal AI agent that runs on Google Cloud virtual machines, for US Google AI Ultra subscribers and select business users.

Why it matters: Always-on personal AI agents are now packaged as a consumer subscription. The constraint for operators has shifted from capability and cost to workflow design and specification of work.

David and Goliath view: Most teams still treat AI as a chat interface. Spark rewards a different posture. Operators who articulate the recurring work worth scheduling overnight will pull ahead of those who only use AI reactively.

Offer relevance:

• AI Growth Engine: agentic execution of recurring revenue operations work across Workspace and the web
• Employee Amplification Systems: scheduled, overnight automation of knowledge worker workflows through Tasks, Skills, and Schedules