Google’s AI Agent Browses, Shops & Books for You

Google’s Gemini Spark runs as a background agent on Google’s Antigravity infrastructure using Gemini 3.5 Flash. It performs multi-step tasks — browsing the web, reading emails, updating calendars — while your devices are off, which separates it from Chrome’s earlier right-click summarizer that required you to be actively browsing. The agent connects to Gmail, Calendar, Drive, Docs, Sheets, Maps, and YouTube natively, giving it access to your entire Google account as a workspace.

How is Gemini Spark different from Chrome’s AI feature?

Chrome’s AI browsing, which we covered back in May, asked you to highlight text and right-click for a summary. Useful enough, but passive — you had to be sitting in front of the screen, selecting text, making requests. Spark runs 24/7 server-side without needing your browser open. Your laptop can be closed. Your phone can be dead. It keeps working.

That distinction matters because it shifts the AI from a tool you actively operate to one that operates on your behalf. Think of the Chrome feature as a smart assistant at your desk; Spark is an employee who never clocks out.

What can Gemini Spark actually do?

Spark operates on three building blocks: Tasks, Skills, and Schedules.

Tasks are one-off multi-step jobs. Tell it to find interior design internships in New Orleans for this summer, and it browses job boards, filters results, compiles a spreadsheet, and emails you a summary — no keyboard touches from you.

Skills are reusable behaviors you train once. Feed Spark 50 of your past emails, and it extracts your writing style into a “ghostwriter” skill that drafts future emails in your voice. That isn’t a prompt you rewrite every time — it’s a persistent capability the agent calls whenever you need it.

Schedules are time-based triggers. “Every Monday at 9 AM, scan my inbox, summarize the important stuff, and block out deep-work time on my calendar.” You set it once, and your Monday mornings begin organizing themselves before your coffee finishes brewing.

The web browsing piece is what makes Spark practical for everyday errands. Google’s own pitch: “Gemini can do all the time-consuming research and live web browsing to gather information across multiple sites, compare options, and even help you complete bookings.” So you say “find me a hotel in Austin for under $200 a night next weekend with good reviews and free cancellation,” and Spark browses booking sites, filters by your criteria, presents options, then handles the reservation with your approval. That’s the jump between asking an AI to recommend hotels — which any chatbot can manage — and asking one to actually book one for you.

Can an AI agent really shop and book things for you?

Yes, with an asterisk. Spark browses real websites, compares real options, and — once you approve — completes real transactions. For solo builders, the practical payoff is specific: the gap between “I should research this” and “I’ve researched this, here’s the spreadsheet” collapses from hours to minutes. Competitor analysis, pricing research, vendor comparison, tool evaluation — Spark handles the browsing while you make the decisions.

I tested this mental model with a few scenarios. Say you’re comparing four email marketing platforms, each with different pricing tiers, feature sets, and integration limits. Manually, that’s 90 minutes of tab-hopping and note-taking. Spark can pull that into a comparison sheet during a single coffee break.

But the agent has boundaries. It’s designed to check with you before major actions — reservations, purchases, sending emails on your behalf. That means you’re still in the loop for judgment calls. The agent does research and organization; you still approve the work. If you’re hoping it’ll run your business without you, temper that expectation now.

What does Gemini Spark cost, and who is it for?

Gemini Spark is available to Google AI Ultra subscribers at $249.99/month, plus select business users. Google positions Ultra as the tier for people who want AI to actually do work rather than answer questions.

$250 a month puts it in the same range as a part-time virtual assistant, but the comparison favors Spark on paper: a VA works business hours; Spark works all hours. A VA handles one task at a time; Spark can run multiple tasks across your email, calendar, documents, and the open web in parallel.

Google says broader access is coming “over the coming weeks,” which probably means a lower-tier version for AI Pro ($19.99/month) or free users with limitations — fewer parallel tasks, slower processing, no scheduling, possibly no web browsing.

What should solo builders consider before subscribing?

For anyone running a one-person business, the question isn’t whether the technology is impressive. It’s whether it changes what’s realistic as a team of one.

Research becomes parallel. Instead of spending an afternoon comparing tools, pricing, or competitors, you hand Spark the criteria and return to a spreadsheet. The AI agent space has promised this for a year; Google’s version works because it has real web access and native integrations rather than hypothetical API connections.

Scheduling becomes automated. Inbox triage, calendar management, follow-up reminders — these eat 30 to 60 minutes of a solo builder’s day. Spark’s scheduling turns them into background processes. Not “set a reminder” — actually process the information and act on it.

Your Google account becomes an operating system. Gmail, Calendar, Drive, Docs, Sheets, Maps, YouTube — Spark connects all of them natively. It reads a client email, creates a project folder, drafts a response, and blocks time on your calendar without touching a third-party tool. That’s the no-code automation idea built directly into the platform most of us already use.

Privacy is the real trade-off, though. Spark reads your emails, scans your Drive, and browses the web on your behalf. Google says connections are off by default and the agent asks before major actions. You’re still giving an AI system access to your entire digital life. If that makes you uncomfortable, the monthly price isn’t your primary barrier.

The AI tools world keeps splitting into two tiers: chatbots that answer questions and agents that do work. Gemini Spark lands firmly in the second category. Whether you need it today depends on how much of your day is browsing, organizing, and researching versus actually building.

How much does Gemini Spark cost? Gemini Spark is included with Google AI Ultra at $249.99/month, with broader access expected on lower tiers in the coming weeks.

Can Gemini Spark browse the web and make purchases? Yes. It browses real sites, compares options across multiple sources, and can complete bookings or purchases after you approve the action.

Does Gemini Spark work when my computer is off? Yes. It runs server-side on Google’s Antigravity infrastructure, so it continues processing tasks even when your devices are closed or powered down.

Is my data safe with Gemini Spark? Google states that agent connections are off by default and the agent requests permission before major actions. However, the system does have access to your email, calendar, and files, which is the core privacy trade-off.

Who is Gemini Spark best suited for? Google targets power users and professionals who want AI to handle multi-step work — research, scheduling, email drafting — rather than simple question-and-answer interactions, particularly those already embedded in Google’s ecosystem.