Gemini Spark: Google’s 24/7 Cloud AI Agent Now Executes Tasks in Third-Party Apps

Google debuted Gemini Spark at its I/O 2026 developer conference on May 19, marking its most consequential AI product shift in years: a persistent personal agent that runs on dedicated cloud servers around the clock and can now execute tasks inside third-party apps, not just Google's own services. Beta access for US Google AI Ultra subscribers begins this week, at the newly reduced $100-per-month tier.

Spark runs on Gemini 3.5 Flash and Google's Antigravity 2.0 agent harness. Unlike the standard Gemini chatbot — which ends when a user closes the tab — Spark persists on dedicated Google Cloud virtual machines, continuing to work when a phone is locked or a laptop shut. CEO Sundar Pichai described it at the I/O keynote as "your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction."

That last clause matters. Spark is designed to act under user direction, with explicit opt-in required for every connected app and a confirmation step before high-stakes actions — including spending money or sending external emails. Google's official Spark product page labels it "Coming soon to AI Ultra" and instructs users to "check responses, supervise closely, interrupt when needed" — language that signals both the product's potential and its current beta status.

Gemini Spark vs. Regular Gemini: What Changed

The practical difference between Spark and the existing Gemini assistant is the difference between asking a colleague a question and delegating a project. Standard Gemini responds when prompted; Spark operates on tasks, skills, and schedules that run on a clock or fire when a condition is met.

Google's official documentation describes three interaction modes. Tasks let users assign multi-step work to Spark — for example, "find and track interior design internships in New Orleans for this summer." Skills are reusable instruction sets a user builds over time: someone can ask Spark to read their last 50 outgoing emails, derive a personal writing style guide from them, and invoke that guide automatically whenever Spark drafts a new message. Schedules trigger recurring actions — such as scanning an inbox every Monday morning, generating a prioritized to-do list, and blocking focus time on the calendar.

These capabilities require persistent context and scheduled execution, making Spark architecturally closer to a background service than to a chatbot. Spark can monitor Gmail for school deadline emails, parse credit card statements for hidden subscription fees, synthesize meeting notes into polished Google Docs, and extract client leads from inbound inquiries into a Google Sheet — all without the user present.

Which Apps Does Gemini Spark Connect Through MCP?

The third-party integration layer is where Spark's architecture becomes most significant. Google has adopted the Model Context Protocol — an open standard introduced by Anthropic in November 2024 that defines a standardized, secure way for AI agents to communicate with external data sources and services. At launch, Spark ships with MCP connections to Canva, OpenTable, and Instacart. Adobe, Samsung, Spotify, CapCut, GitHub, Notion, and Slack are confirmed for summer 2026.

The practical implication is significant: a user can ask Spark to prepare a Canva presentation from notes in Google Docs, make an OpenTable reservation based on context in their inbox, or assemble an Instacart grocery order — tasks that previously required manually switching between apps and entering the same information multiple times. Because MCP is an open, cross-platform standard now adopted by Anthropic, Google, and other major AI platforms, any developer that builds an MCP-compliant server can make their product accessible to Spark without writing custom integration code for Google specifically.

Cornell University's Karan Girotra, a professor of operations, technology and innovation, told CBS News that Google's native data access is a competitive differentiator. "It knows more about you than many others because it connects to Gmail and other apps," Girotra said, "so personal intelligence will come through in the agent." The Antigravity 2.0 harness handles tool invocation and long-running planning loops underneath, while Spark functions as the user-facing layer on top.

Gemini Spark Privacy: What the Leaked Warning Said

Before the I/O keynote, a pre-release APK leak of the Gemini app surfaced an onboarding screen disclosing that Spark "may do things like share your info or make purchases without asking." Google softened that language before launch — the shipped version says Spark is "designed to check with you before taking major actions" — but the company has not published a Spark-specific privacy policy as of the keynote date.

That gap arrives alongside an existing proposed class-action lawsuit, Thele v. Google LLC, filed in federal court in San Jose in November 2025. The suit alleges that in October 2025 Google secretly enabled Gemini across all Gmail, Chat, and Meet accounts without user consent, giving the AI access to the full contents of private communications. The complaint names violations of the California Invasion of Privacy Act, the California Computer Data Access and Fraud Act, and the Stored Communications Act. Google has not commented publicly on the case.

Users can disable Gemini's access to Workspace apps entirely through the Data and Privacy section of their Google Account settings. Independent reviewers have recommended starting with a conservative permissions whitelist — connecting only one or two low-stakes services — and expanding access after observing how the agent behaves in practice. Professionals in healthcare, legal, or financial fields who handle privileged or sensitive data should review applicable data-handling policies before opting in.

The EU AI Act's Article 50 transparency obligations for AI systems that interact with users take effect August 2, 2026. The absence of a Spark-specific privacy policy ahead of that deadline represents a compliance gap Google will need to address before it can expand the product into the European Union.

How Gemini Spark Compares to Claude Cowork and ChatGPT Agent

Gemini Spark enters a crowded but still-young category. Anthropic's Claude Cowork — which reached general availability on April 9, 2026 — runs as a desktop agent that drives applications locally on a user's machine. OpenAI's ChatGPT Agent operates primarily through a browser. Microsoft's Copilot is grounded in Office 365 data. Apple is preparing a revised Siri powered in part by Google Gemini models, expected at WWDC 2026.

What distinguishes Spark from all of them is persistent cloud execution tied to a native data advantage. Because Spark runs on Google's own infrastructure and connects natively to Gmail, Calendar, Drive, and the full Workspace suite, it does not need to simulate a user's actions to access that data — it reads from the source. No competing agent has that level of direct access to the most widely used productivity suite in the world, which Google reports now serves 900 million Gemini monthly active users across 230 countries.

The trade-off is trust concentration: a cloud agent with standing access to a user's inbox, calendar, and third-party purchasing accounts is a meaningfully different consent model than a chatbot that operates session-by-session. Clarence Lee, a tech entrepreneur and visiting lecturer at Cornell's SC Johnson College of Business, advised users to move gradually into the world of AI agents by initially entrusting them with low-level tasks. A tool that books a restaurant reservation based on inferred preferences carries a different risk profile than one that processes a medical record or a legal communication.

Is Gemini Spark Worth $100 a Month?

Access to Spark requires the Google AI Ultra subscription, now priced at $100 per month — down from $250 — available in the US only at launch. The plan also includes five times the usage limits of the $20 AI Pro tier, 20 terabytes of Google One cloud storage, and YouTube Premium. A $200 tier exists for higher usage limits. For EU and UK subscribers, availability is contingent on AI Act compliance review; analysts expect a Q3 2026 timeline at the earliest.

The $100 price point matters because Google, Anthropic, and OpenAI now all offer a premium AI tier at roughly the same price. Google AI Ultra at $100 is the only one in that range to include a persistent cloud agent. Anthropic's Claude Pro plan starts at $20 with Cowork included; its Max plan at $100 to $200 matches the price but runs a desktop model rather than a cloud-resident one. For users already invested in the Google Workspace ecosystem — where Spark's data advantage is greatest — the cost argument is reasonably strong, provided the privacy and autonomy trade-offs are acceptable.

Frequently Asked Questions

What is Gemini Spark and how is it different from regular Gemini?

Gemini Spark is Google's 24/7 personal AI agent, announced at Google I/O 2026, that runs continuously on Google Cloud virtual machines and takes actions across your apps — even when your device is off. Unlike the standard Gemini chatbot, which ends when you close the tab, Spark operates on tasks, skills, and scheduled workflows, handling multi-step jobs in Gmail, Google Docs, and third-party apps without requiring you to be present for each step.

Which apps can Gemini Spark work with?

Spark connects natively to Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube, and Google Maps. At launch it also supports Canva, OpenTable, and Instacart through the Model Context Protocol (MCP), the open integration standard developed by Anthropic. Adobe, Samsung, Spotify, GitHub, Notion, and Slack are confirmed as additional MCP partners arriving over summer 2026.

Is Gemini Spark reading my emails all the time?

Google says no — Spark does not read emails indiscriminately. It works under your direction, on tasks and schedules you set, and all Workspace connections are turned off by default. Users must enable each integration individually in their Google Account settings. For recurring tasks involving Gmail, however, Spark maintains standing access to execute those workflows on the schedule the user defines.

How much does Gemini Spark cost, and who can access it now?

Gemini Spark is available to US Google AI Ultra subscribers, whose plan was reduced from $250 to $100 per month at I/O 2026. Beta access began rolling out to trusted testers the week of May 19, 2026, with broader US access starting the week of May 26. EU and UK availability is pending AI Act compliance review, with a Q3 2026 timeline expected by analysts. No standalone Spark subscription exists; the AI Ultra plan is the only access path.