Gemini Spark is Google's always-on personal AI agent: a cloud worker that runs multi-step jobs across Gmail, Docs, Calendar, and Chrome while your laptop is closed, powered by Gemini 3.5 on the Google Antigravity agent harness. Google announced it at Google I/O on May 19, 2026 and is rolling a U.S. beta to Google AI Ultra subscribers at $100/month (and the $200/month tier). Two weeks later, Microsoft Scout arrived at Build 2026 as Redmond's answer in the same product category. This guide explains what Spark actually does, who can access it, how Google gates risky actions, and how it compares to Scout and Claude Cowork.
Key takeaways
- Gemini Spark is a background agent, not a chat box: it executes recurring workflows across Workspace and the open web under your direction.
- Access today is U.S.-only beta for Google AI Ultra ($100 or $200/month); Workspace business preview and Gemini Enterprise rollout are labeled "soon."
- Governance requires explicit approval before high-stakes actions like sending email or spending money; enterprise runs use ephemeral VMs and an Agent Gateway with DLP.
- vs Microsoft Scout: Spark leads on native Google data reach; Scout leads on Entra identity, Purview pre-action policy, and M365 audit trails.
- vs Claude Cowork: Spark is cloud and unattended; Cowork is a desktop sandbox for interactive file work while you are at the keyboard.
What is Gemini Spark?
Gemini Spark is a 24/7 personal AI agent inside the Gemini app and, eventually, Gemini Enterprise and Google Workspace. Unlike the Gemini side panel that answers when you prompt it, Spark accepts a goal, plans steps, and keeps working in the background.
Google builds Spark on two layers. Gemini 3.5 supplies reasoning and tool use; Google Antigravity is the agent-first harness Google uses across Search, Cloud, and the Gemini app for long-horizon tasks. Spark also inherits context from Workspace Intelligence, the cross-app understanding Google previewed at Cloud Next '26 across Docs, Gmail, Drive, and Chat. (Sources: Google Gemini app blog, Google Cloud I/O blog)
Alphabet CEO Sundar Pichai framed the shift at a pre-briefing: "It's your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction." Josh Woodward, VP of Google Labs for the Gemini app, said Spark "transforms [Gemini] from an assistant that can answer your questions into an active partner that does real work on your behalf." (Sources: Google I/O announcements, Google Gemini app blog)
The architectural bet is cloud persistence. Spark runs on dedicated virtual machines on Google Cloud, so it does not need your phone or laptop awake. TechCrunch quoted Pichai: "It runs on dedicated virtual machines on Google Cloud seamlessly, [so] you don't need to keep your laptop open to make sure it's running." (Sources: TechCrunch, Google Gemini app blog)
Personal agents are leaving the chat window. The fight is over who owns the runtime, the identity, and the audit trail when software acts while you sleep.
What Gemini Spark can do today
Spark's launch workflows cluster around inbox triage, document synthesis, and connector-driven errands.
Background workflows you can delegate
- Recurring monitors: Parse monthly credit card statements and flag new subscription charges, or watch a shared inbox for client requests and cross-reference timeline sheets.
- Inbox-to-deliverable chains: Pull facts from Gmail, Docs, Sheets, and Slides, then draft a status email to your manager. Woodward demoed exactly that pattern at I/O. (Source: TechCrunch)
- Project kickoffs: Synthesize meeting notes scattered across email and Chat, produce a polished Google Doc, and draft the companion kickoff message.
- Web and Chrome reach: Spark can interact with the open web through Chrome, not only Workspace tabs.
- MCP connectors: Launch partners include Canva, OpenTable, and Instacart, with more integrations planned through summer 2026. (Source: Google Gemini app blog)
On the enterprise side, Google Cloud described heavier cross-system jobs: Spark can suggest code changes via Antigravity, open a Jira ticket, recalculate launch timelines from team documents, draft stakeholder email, monitor ServiceNow health, or assemble Salesforce plus Zendesk context before a client meeting. Those flows assume Gemini Enterprise connectors such as SharePoint, OneDrive, and ServiceNow, not only consumer Workspace. (Source: Google Cloud I/O blog)
Google's summer roadmap adds texting or emailing Spark directly, custom sub-agents, local browser control, and payment authorization with merchant and budget caps. (Source: Google I/O announcements)
Who gets Gemini Spark and when
Rollout is deliberately staged. Google is prioritizing safety over breadth in this first release.
| Audience | Availability | Source tier |
|---|---|---|
| Trusted testers | From May 19, 2026 | Primary (Google I/O list) |
| U.S. Google AI Ultra ($100 or $200/month) | Beta in the week after I/O | Primary (Google AI subscriptions) |
| Workspace business customers | Preview via Gemini app, "soon" | Primary (Google Cloud blog) |
| Gemini Enterprise customers | Rolling out "soon" | Primary (Google Cloud blog) |
| Gemini macOS desktop app | Summer 2026 (local file workflows) | Primary (Google Gemini app blog) |
Google AI Ultra now ships at two price points Google confirmed at I/O: $100/month (5X Pro usage limits, 20TB storage, YouTube Premium, Antigravity priority) and $200/month (20X Pro limits, down from a prior $250/month list). Gemini Spark is listed on both Ultra tiers, U.S. only. Daily Brief and Gemini Omni reach lower tiers globally, but Spark does not. (Source: Google AI subscriptions blog)
Operator note (first-hand): On June 11, 2026, a fetch of https://gemini.google.com/subscriptions returned a public pricing page without a Spark enrollment toggle, consistent with a gated U.S. Ultra beta rather than a self-serve global switch. Treat Spark as invite-or-rollout-dependent until Google exposes an in-app enablement path in your account region.
Governance and safety gates
Always-on agents fail in production when autonomy outruns policy. Google is leading with confirmation gates in the consumer beta and stronger isolation in enterprise.
Consumer and prosumer controls
Spark "is designed to ask you first before performing high-stakes actions like spending money or sending emails," Google says. You choose whether to turn Spark on, which apps it connects to, and how much leash it gets. Kimbley's rollout guidance matches Google's: start with draft-only email workflows before widening permissions. (Sources: Google Gemini app blog, Kimbley IT)
Enterprise controls
For Gemini Enterprise, Google Cloud states each Spark task runs in a fresh, strictly isolated, ephemeral VM so sessions do not share data. Traffic routes through a secure Agent Gateway that enforces Data Loss Prevention (DLP) policies, and user credentials stay encrypted rather than exposed directly to the agent. High-risk actions like sending email still require explicit approval. (Source: Google Cloud I/O blog)
That is a different shape than "trust the model prompt." It is closer to a managed agent runtime with network and data policy in the path, similar in spirit to how enterprises already expect API gateways in front of services.
Gemini Spark vs Scout vs Claude Cowork
Three vendors are productizing the same idea with different runtimes and governance postures.
| Dimension | Gemini Spark | Microsoft Scout | Claude Cowork |
|---|---|---|---|
| Runtime | Cloud VM on Google Cloud; device can be off | Cloud + desktop + web across M365; OpenClaw-based | Desktop sandboxed VM on your machine |
| Native data | Gmail, Docs, Drive, Calendar, Chrome | Teams, Outlook, SharePoint, OneDrive, Graph | Local files + MCP connectors (Slack, HubSpot, etc.) |
| Identity / audit | Google account; enterprise Agent Gateway + DLP | Governed Entra identity per instance; Purview pre-action; Purview Audit | Interactive sessions; enterprise audit gaps noted by analysts for Cowork |
| Approval model | Ask before email send, calendar writes, purchases | Human approval on sensitive actions; DLP before execution | Conservative ask-before-acting bias |
| Availability (June 2026) | U.S. Ultra beta; Workspace preview soon | Frontier experimental release; Intune + Copilot license | Generally available on Claude Pro/Max tiers |
| Price anchor | From $100/month (Google AI Ultra) | Frontier + GitHub Copilot license (enterprise) | From $20/month (Claude Pro add-on path) |
Microsoft Scout, announced June 2, 2026 at Build, is Microsoft's first Autopilot: an always-on agent with its own Entra directory identity, Purview sensitivity and DLP checks before actions run, and logging to Purview Audit. Scout is powered by OpenClaw and grounded in Work IQ context across Teams, Outlook, and SharePoint. Access today requires Copilot Frontier enrollment, Intune policy, opt-in attestation, and a GitHub Copilot license. (Source: Microsoft 365 Blog)
Claude Cowork sits in a different lane. It is an interactive desktop agent in a sandboxed Linux VM for file creation, spreadsheet work, and MCP-connected apps while you collaborate in real time. It is not marketed as a 24/7 cloud worker that runs while your machine sleeps. Analysts also flag that Cowork activity may not yet surface in Anthropic enterprise audit exports, which matters for regulated buyers. (Sources: community comparisons, General Analysis Cowork security guide)
Inference: Spark and Scout will compete for "who owns workplace delegation" in Google vs Microsoft shops. Cowork competes for "who produces the deliverable while I watch," not unattended overnight ops.
Which personal agent should you pick?
Match the agent to where your data and identity already live.
Pick Gemini Spark if your team runs on Google Workspace, you want unattended background workflows, and you can access the U.S. Ultra beta (or wait for Workspace preview). The $100 Ultra tier is Google's prosumer on-ramp; enterprise buyers should plan around Gemini Enterprise connectors and Agent Gateway policy reviews.
Pick Microsoft Scout if you are a Frontier M365 customer that needs Entra-attributed agents, Purview enforcement before action, and audit search across agent behavior. Scout is the stronger fit when security teams ask "who sent that email" and need a directory answer.
Pick Claude Cowork if you need hands-on document and file automation with a cautious approval bias, not an always-on cloud delegate. Many early adopters run Cowork for interactive deliverables and a cloud agent like Spark for monitoring, accepting two subscriptions for two shapes of leverage. (Inference: based on early 2026 practitioner patterns)
Start every rollout with one low-risk job, draft-only outputs, and a written policy for which connectors Spark or Scout may touch.
Frequently asked questions
What is Gemini Spark?
Gemini Spark is Google's 24/7 personal AI agent inside the Gemini ecosystem. It uses Gemini 3.5 and the Antigravity harness to run multi-step workflows across Gmail, Docs, Calendar, Chrome, and MCP-connected apps while your devices are off. Google announced it at I/O on May 19, 2026.
How much does Gemini Spark cost?
Spark is bundled with Google AI Ultra, not sold standalone. Google launched a $100/month Ultra tier and cut the top tier to $200/month (from $250) at I/O 2026. Both include Spark in the U.S. beta. Lower tiers get Daily Brief and Omni, not Spark.
Is Gemini Spark available outside the U.S.?
Not yet for Spark. Google labels the agent U.S. only during the initial beta. Workspace business and Gemini Enterprise previews are marked "soon" without regional dates. Daily Brief and other Gemini features roll out more broadly on lower tiers.
Does Gemini Spark send emails without permission?
Google says Spark asks before high-stakes actions, including sending email, adding calendar events, and completing purchases. Enterprise deployments add Agent Gateway DLP and ephemeral VMs. Operators should still start with draft-only email tasks until outputs are trusted.
How is Gemini Spark different from Microsoft Scout?
Spark is Google's cloud-native personal agent for Workspace and Chrome, gated behind Ultra in the U.S. beta. Scout is Microsoft's OpenClaw-based Autopilot for M365 with governed Entra identity, Purview pre-action policy, and Purview Audit logging. Scout targets Frontier enterprises; Spark targets Google ecosystem users first.
How does Gemini Spark compare to Claude Cowork?
Spark runs unattended in Google Cloud across Workspace and the web. Cowork is a desktop sandbox for interactive file and document work with MCP connectors. Spark fits background monitoring and cross-app synthesis; Cowork fits producing deliverables while you collaborate live.
Related coverage
- Workspace Intelligence in Google Workspace: What Actually Shipped
- Scout & OpenClaw: Windows becomes an agent OS
- Google Search agents: your 24/7 AI info agents
- Anthropic's Claude Managed Agents turns agent runtime into an API surface
References
- Google AI subscriptions I/O 2026 - https://blog.google/products-and-platforms/products/google-one/google-ai-subscriptions/
- Google Cloud I/O 2026 innovations - https://cloud.google.com/blog/products/ai-machine-learning/innovations-from-google-io-26-on-google-cloud
- Google Gemini app evolution (Spark launch) - https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/
- Google I/O 2026 announcements - https://blog.google/innovation-and-ai/technology/ai/google-io-2026-all-our-announcements
- Microsoft Scout announcement - https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/02/introducing-microsoft-scout-your-always-on-personal-agent/
- TechCrunch Spark coverage - https://techcrunch.com/2026/05/19/google-introduces-gemini-spark-a-24-7-agentic-assistant-with-gmail-integration
