Google Wires Gemini Live Into Spark to Run Background Tasks by Voice

Gemini Live just got a productivity transplant. Google is wiring its real-time voice assistant into the same agent stack that powers Spark, its background task runner, plus new hooks into Gmail, Calendar, and a long-term memory layer called Personal Intelligence. Google's pitch is that Gemini Live can now handle complex tasks on your behalf, so you can talk through your day and delegate to-dos as you go.

Voice is no longer just a way to prompt a chatbot. It is the front door to an orchestrator that decides whether your request needs a quick answer, an inbox lookup, a morning digest, or a multi-day agent job running in the background.

Four features stitched into one voice loop

The rollout bundles together four capabilities that used to live in separate corners of the Gemini app:

• Spark integration. Natural voice commands can set up complex, multi-step tasks that run on their own across Google Docs, Sheets, Drive, and the web while remembering your goals. Spark also handles scheduled jobs that span days or weeks, even when you are not actively using the app.
• Daily Brief. Ask "What is my daily brief?" and Gemini combines your most important updates from Gmail and Calendar into one concise digest.
• Hands-free Gmail. Prompts like "Any new emails?" or "Are there any urgent emails from the kids' school today?" let you search, summarize, star, archive, or delete messages by voice.
• Personal Intelligence. Gemini Live remembers what you have chatted about and connects the dots across past conversations and Google apps like Gmail, Photos, Search, and YouTube.

How the orchestration actually works

Spark does the heavy lifting under the hood. Google's persistent agent framework taps Connected Apps like Workspace and Search, plus Personal Intelligence, a remote web browser and computer with code execution, chats, Canvas, and location data. Its workflows are structured around Skills (reusable instructions), Schedules (automated triggers), and Tasks (high-level goals). Users can run up to 15 tasks at a time, drawing on the same compute-based usage limits as the rest of Gemini.

Spark itself runs on Google's newer workhorse model, Gemini 3.7 Flash, and is available to Google AI Pro and Ultra subscribers in over 160 countries.

You never explicitly invoke any of these subsystems, which is the interesting design decision. In Google's example, you can ask what your day looks like, get a spoken briefing, then follow up by asking Gemini to grab event invites from email, create calendar events, and include driving times. Gemini decides that this is a background job and hands it to Spark while the voice conversation continues.

Availability and pricing

Gemini Live itself is free and global, but the new agentic pieces sit behind Google's paid tiers:

• Daily Brief requires a Google AI Plus subscription or higher.
• Spark requires a Google AI Pro subscription or higher.
• Personal Intelligence and Gmail voice actions are tied to your connected apps settings inside the Gemini app.

To turn everything on, connect your apps in Gemini's Personal Intelligence settings and tap the Live icon. Spark's international expansion currently excludes the European Economic Area, the UK, Switzerland, and Nigeria.

Where it shines and where it stumbles

The strong suit here is the boring stuff: converting a rambling voice note into a structured Google Doc, keeping a rolling meal plan and grocery list, or triaging inbox noise on a commute. Google's own examples lean heavily on family and household logistics rather than technical work, which is a fair signal of where the reliability envelope currently sits.

The weaker areas involve the usual agent caveats. For actions such as sending emails or making purchases, Spark is designed to seek user approval before proceeding, which is good for safety but adds friction to fully autonomous workflows. And because Personal Intelligence pulls from Gmail, Photos, Search, and YouTube history, the quality of results depends heavily on how much of your Google footprint you are willing to expose.

Why it matters

Voice assistants have been stuck in a loop of timers, weather, and one-shot questions for a decade. Bolting a real agent runtime onto a low-latency voice interface is the first serious attempt to make spoken commands worth using for anything longer than a sentence.

The competitive read is also worth noting. Google's $100 tier is the only one among Google, Anthropic, and OpenAI with a persistent cloud agent, and pushing Spark down to voice on the $20 Pro tier is how Google turns that differentiation into something regular users actually touch. If you already live inside Gmail and Docs, this is the first version of Gemini Live that starts to feel like an assistant rather than a demo.