Gemini Spark: What Google’s 24/7 AI Agent Actually Is (And What It’s Not)

Gemini Spark is Google’s always-on personal AI agent, announced at Google I/O 2026. It runs on Google Cloud servers continuously, taking action across Gmail, Calendar, and Docs even when your devices are off. Google introduced two new Ultra pricing tiers at the same event: a new $100/month entry tier and a reduced $200/month top tier (down from $250). Spark is currently US-only beta. This article covers how it works, how it compares to Claude and ChatGPT agents, what the pricing restructure actually signals, and what to watch before you pay for it.

Google has spent two years calling Gemini an “assistant.” At I/O 2026, they stopped using that word.

The keynote replaced it with “agent.” Sundar Pichai stood on stage and described Gemini Spark as something that “takes action on your behalf” and “works in the background on your phone or laptop even while they’re turned off.” That is a carefully chosen sentence. An assistant waits to be asked. An agent decides when to act.

That shift in language is not a branding update. It describes a fundamentally different product category, one that every major AI company is now racing to define. OpenAI shipped ChatGPT Agent. Anthropic shipped Claude Cowork. Microsoft Copilot works inside Office 365. Apple is preparing a Siri revamp for WWDC 2026. Google arrived last to this specific race, but they arrived with the largest installed base: 900 million monthly Gemini users and deep integration across the most widely used productivity suite in the world.

Whether that advantage translates into a better product is a different question. Here is what is actually confirmed, what it costs, and where the real risks are.

What Gemini Spark Actually Is

Gemini Spark is a 24/7 autonomous AI agent that runs on dedicated Google Cloud virtual machines, not inside a chat window.

That architectural distinction matters more than the feature list. Every other AI assistant (ChatGPT, Claude, earlier Gemini) operates within a session. You open the app, it runs, you close it, it stops. Spark runs as a persistent cloud process. It keeps working when you lock your phone. It keeps working when you close your laptop. It keeps working at 3am when you are asleep.

The architectural shift from “session” to “process” is the same leap email made over fax. That analogy is slightly breathless but structurally accurate. A session-bound chatbot requires your presence. A persistent agent does not.

Spark is built on Gemini 3.5 Flash and is powered by Antigravity, Google’s agent-first development platform. Antigravity is the orchestration layer that handles subagents, hooks, and asynchronous task management, it is what makes multi-step, long-running tasks possible. The cloud persistence (keeping Spark running when your devices are off) comes from running on Google’s infrastructure. Antigravity is what coordinates the work. It operates autonomously under your direction. You choose to turn it on, and for certain tasks, it will ask for your approval before proceeding rather than acting unilaterally.

That last point is the most important one in the entire announcement and the one most coverage glosses over. The approval model is task-specific, not category-based. From the keynote demo, Spark autonomously colour-coded calendar events, drafted a neighbour invitation, and built a school-year planning doc without asking permission for any of them. It paused for approval on tasks with external consequences. Google has not published a precise definition of where that line sits.

What Spark Can Do Right Now

Based on Google’s I/O 2026 presentation, Spark’s capabilities include email management (triaging your inbox, drafting responses, flagging important messages, and following up on threads you’ve neglected), calendar management, document creation, and multi-step workflow automation.

Three concrete examples from the keynote demo illustrate the range:

Voice-initiated task batching: The demo showed a user dictating three separate tasks at speed, find all upcoming meetings with a specific person and colour-code them, write a note to a new neighbour with a block party invitation, and create a school-year checklist doc categorised by deadline and priority. Spark captured all three, broke them into individual threads, and worked on them in the background while the phone was put away.

Proactive recurring tasks: Spark can look ahead on your calendar, see that you’re on snack duty for your kid’s game on Saturday, and proactively set up your Instacart order on its own, including remembering dietary restrictions like nut allergies.

Complete multi-step documents: The demo showed Spark creating a Google Doc with full formatting, colour coding, and categorisation from a single voice prompt, not a plain text output but a properly structured document ready to use.

Spark also supports MCP (Model Context Protocol) connections to third-party services. Integrations at launch include Canva, OpenTable, and Instacart, with more partners coming through summer. The roadmap includes the ability to text or email Spark directly, create custom sub-agents, and authorize payments within a specified budget and merchant list through Google’s new AP2 (Agent Payments Protocol).

Two upcoming capabilities confirmed at the keynote that most coverage missed:

Chrome integration: Gemini Spark will operate directly within Chrome, acting as an agentic browser that can take action on your tasks across the web, not just within Google’s own apps.

Android Halo: Google is building a dedicated home base for your agents on Android, arriving later this year. This is the mobile interface purpose-built for managing Spark and future agents.

Workspace and Enterprise: A version of Spark is also coming to Google Workspace and Gemini Enterprise, making it relevant beyond personal use.

The Pricing Restructure: What Actually Happened

The pricing story from Google I/O 2026 is more interesting than most coverage reported and more nuanced than a single price cut.

Google did not simply slash the price of Ultra. They restructured the entire tier. Here is what Sundar Pichai actually announced at the keynote:

Google introduced a new $100/month Ultra entry tier to give Spark a mass-market price point. Simultaneously, they reduced the existing top-tier Ultra plan from $250/month to $200/month for users who need maximum limits.

This is a two-move pricing play, not a one-move cut. The $100 tier is new, designed to get Spark into as many hands as possible at a price competitive with Claude Max and below ChatGPT Pro. The $200 tier is a retention move for heavy users who were already paying $250 and might have felt exposed by a cheaper tier appearing below them.

The strategic signal is real regardless: Google looked at the competitive landscape and decided their agent needed a consumer-accessible entry point on the same day it launched. That is not a coincidence. It is an acknowledgment that an agent nobody can afford to try will not build the usage data Google needs to improve it.

Ultra also bundles 20 TB of cloud storage, YouTube Premium, and 5x the Gemini usage limits of the $20 AI Pro tier. At $100 with that bundle, Google is not just selling an AI agent. They are selling the argument that you should consolidate several subscriptions into one.

Whether that argument holds depends entirely on how well Spark works in practice. The Beta starts with US subscribers in June. Real data will follow.

How Spark Compares to the Competition

OpenAI’s ChatGPT agent operates primarily through a browser. Anthropic’s Claude Cowork works directly on a user’s desktop. Microsoft’s Copilot is grounded in Office 365 data. Each one reflects the company’s core infrastructure advantage.

Spark’s advantage is Gmail and Google Workspace combined with Chrome. The Chrome integration is significant: it means Spark can act across the entire web, not just within Google’s own apps. That is a broader surface area than any desktop-only agent. If your work lives in Google’s ecosystem and you spend most of your time in Chrome, Spark has native access to nearly everything you touch without integrations, permissions headaches, or data export steps.

The limitation is the same thing: if you do not live in Google’s ecosystem, Spark’s value depends on how well the MCP partner integrations expand. The Canva and OpenTable connections are useful but not essential. The integrations that would matter most to developers and knowledge workers (Notion, Slack, GitHub, Linear) are not on the confirmed list yet.

One more honest framing: Google is committing between $180 billion and $190 billion to AI infrastructure in 2026 alone. That figure, confirmed in Sundar Pichai’s keynote, reframes every announcement from I/O. This is not a product roadmap. It is a structural reorganisation of compute around agentic workloads. Spark is the consumer face of that investment. The underlying compute scale is genuinely different from what any other company is currently deploying.

What to Watch Before You Pay for It

Spark is very early. Google said so directly at the keynote: “We’re prioritizing safety in this first release, that’s why we’re rolling it out to trusted testers.” That is an honest framing, and it should set expectations accordingly.

Three specific things worth watching before committing to an Ultra subscription for Spark:

The consent boundary in practice. The keynote demo showed Spark acting autonomously on several tasks (calendar changes, document creation, draft emails) and pausing for approval on others. Google has not published a clear definition of what triggers an approval request versus autonomous action. Until that boundary is documented and tested in practice, you are relying on Google’s internal judgment. For routine personal tasks, that is probably fine. For anything involving external communications or financial actions, it is worth understanding before you enable it.

The governance gap for business use. An agent with access to email, calendar, and documents is powerful for personal productivity. For business use, it creates questions about data handling, audit trails, and who is accountable when the agent takes an action that causes a problem. Google’s AP2 payment protocol includes tamper-proof digital mandates and a permanent paper trail, which is a thoughtful approach. The broader governance question for enterprise adoption will take longer to settle.

The US-only launch timeline. Spark is a US-only beta at launch. Google has not announced international availability dates. If you are outside the US, this is a watch-and-wait situation regardless of how compelling the feature set looks.

Should You Pay for Gemini Spark?

The honest answer is: not yet, unless you are already on Google AI Ultra or you have a specific recurring workflow that Spark would automate.

If you are a heavy Gmail, Google Docs, and Chrome user, the value proposition is real. Spark has direct access to the apps you already use, without friction. The new $100 entry tier makes Ultra more defensible as a bundle than the previous $250 price point did.

If you are primarily on Microsoft 365, Apple, or a mixed-tool setup, the ecosystem moat works against you until the MCP integrations expand significantly. The Chrome browser integration helps here; it extends Spark’s reach beyond Google’s own apps, but the core value is still deepest inside the Google ecosystem.

The honest comparison: Spark in June 2026 is roughly where ChatGPT Agent was six months after launch. Technically impressive, occasionally unreliable, and improving fast. The people who benefit most from being early adopters are those with clear use cases and a tolerance for rough edges.

Frequently Asked Questions

What is Gemini Spark and how is it different from regular Gemini? Regular Gemini is a conversational AI that responds when you ask it something. Gemini Spark is a 24/7 autonomous agent that runs on Google’s cloud servers continuously, taking action across your Gmail, Calendar, Docs, and Chrome even when your devices are off. It is proactive rather than reactive.

How much does Gemini Spark cost? Spark requires a Google AI Ultra subscription. At I/O 2026, Google introduced two Ultra tiers: a new $100/month entry tier (new, not a price cut) and a reduced $200/month top tier (down from $250). Spark is not available on the $20 AI Pro plan.

Is Gemini Spark available outside the United States? No. At launch, Spark is a US-only beta for Google AI Ultra subscribers. Google has not announced a timeline for international availability.

Can Gemini Spark act without my permission? For routine tasks like creating documents, colour-coding calendars, and drafting notes, Spark acts autonomously. For tasks with external consequences, the keynote demo showed it pausing for approval. Google has not published a precise definition of what triggers an approval request versus autonomous action.

What apps does Gemini Spark work with? Deep integration with Google Workspace (Gmail, Docs, Slides, Calendar, Drive) and Chrome as an agentic browser. Third-party integrations via MCP at launch include Canva, OpenTable, and Instacart. The roadmap includes custom sub-agents, direct text and email access to Spark, Chrome web automation, and payment authorization through Google’s AP2 protocol.

How does Gemini Spark compare to ChatGPT Agent and Claude Cowork? All three are persistent AI agents that work in the background. ChatGPT Agent primarily operates through a browser. Claude Cowork works on your desktop. Spark combines Google Workspace integration with Chrome browser automation, giving it the broadest surface area of the three at launch. Its limitation is that the deepest integration is with Google’s own ecosystem.

Is there a version for businesses? Yes. Google confirmed at the keynote that a version of Spark is coming to Google Workspace and Gemini Enterprise. Timing was not specified beyond “coming.”

The clearest thing Google proved at I/O 2026 is that the session-bound chatbot era has a defined endpoint. The agents replacing it are more capable, more autonomous, and require more deliberate decisions about what you trust them to do without asking.

Gemini Spark is the most ambitious consumer agent launch so far. The Chrome integration and Workspace version expand its reach significantly beyond what was initially reported. Whether it earns that ambition will be visible in the Beta reviews over the next few months. The new $100 entry tier and same-day pricing restructure suggest Google knows they need adoption to happen fast.