Gemini Spark Joins Chrome: The AI Agent That Will Do Your Online Shopping for You

On July 30, 2026, Google announced the integration of Gemini Spark into Chrome — and it's more than just an AI button slapped onto a browser. Spark can now navigate, fill out forms, log into sites, and initiate transactions on your behalf, directly using your Chrome credentials and saved passwords.

In the same move, Google is expanding access to Spark in 160 additional countries, currently reserved for Google AI Pro and Ultra subscribers. France is included in this expansion wave.

What exactly is Gemini Spark?

Announced at Google I/O 2026 in May, Spark is an agent, not a chatbot. The distinction is important: where classic Gemini answers a question, Spark executes a multi-step task for you, possibly over several days, even if your device is off.

It operates on Google's secure cloud infrastructure — not on your machine — and natively connects to Gmail, Docs, Sheets, Calendar without prior setup if you're already logged into your Google account. Zero plugins, zero manual OAuth configuration.

An agent running in the background even when your phone is locked, with native access to your email and browser. The line between "assistant" and "actor" has officially blurred.

What the Chrome integration does

The July 30 addition leverages Chrome's new "auto-browse" capability, allowing an AI agent to control the browser like a human: open tabs, click links, fill fields, navigate between multiple sites. Spark uses this API to handle what Google calls "web chores".

Specifically, with your permission, Spark can:

• Book multiple viewing slots for apartments you've saved
• Compare and pre-fill flight options, then hand control back to you just before payment
• Respond to an email asking for your availability by checking your Google Calendar
• Fill out an administrative form using information from your Docs / Sheets
• Log into a third-party service using your saved Chrome passwords

The security rule: you validate what matters

Google insists: Spark hands control back for any sensitive action. A payment, a signature, a final email send-off — you're brought back into the loop and manually confirm. The agent prepares, you confirm.

Protection against prompt injection (malicious sites slipping hidden instructions into their pages to manipulate the agent) is built-in from the start. Google describes it as "defense in depth": entry filtering, restricted scope of authorized actions, forced human review at critical points.

How to use it?

1. Prerequisites

• A Google AI Pro subscription (€20/month) or Ultra (€250/month)
• Up-to-date Chrome (feature rolling out gradually from July 30)
• A Google account logged in on Chrome
• Residence in an eligible country (initially US, international expansion ongoing)

2. Activate

Open Gemini (via gemini.google.com or the mobile app), select the Spark mode in the model selector. On eligible browsers, the option "Use Chrome for this task" appears in the conversation settings.

3. Formulate the request

Describe your task in natural language, as you would to an intern. Examples:

• "Find me three round-trip flights Paris → Lisbon the week of September 15, under €350 direct, and prepare the booking on the cheapest one — I'll confirm the payment myself."
• "Schedule a visit for the 5 apartments I've favorited on SeLoger this week, suggest Thursday or Friday evening."

4. Track execution

Spark works in the background. You receive notifications at key stages: confirmation request, obstacle encountered, task completed. You can close Chrome — the task continues on the cloud side.

The relationship with Chrome: deeper than it seems

This integration marks a strategic turning point for Google. Chrome is no longer just a rendering engine — it has become the preferred execution surface for Google's AI agents. Specifically:

1. Cookies and passwords become action tokens. Where OpenAI Operator and Anthropic Computer Use run in blind sandboxes, Spark naturally inherits your session — your already logged-in accounts are immediately accessible.
2. The auto-browse API is integrated into the browser, not an external driver (like Playwright or Puppeteer). This changes the game in terms of performance and anti-bot detection.
3. Google locks its ecosystem. A user who experiences Spark + Chrome will have little reason to switch to Edge or Firefox — the agent won't be there.

In the longer term, it's likely that Google will converge Android, Chrome, and Gemini Spark into a single agentic layer integrated into the OS and browser. The thesis is: the next operating system is neither Windows nor iOS — it's the agent that has access to all your apps.

Gray areas

Three questions remain open and deserve attention:

• Privacy: what exactly does Spark see when it scans your Gmail to respond to an email about your availability? Do emails pass through the model in clear text? Google mentions "guarantees," but the technical details remain sparsely documented publicly.
• Responsibility: if Spark books the wrong flight, fills out an administrative form with an error, or sends an awkward message to an employer — who is responsible?
• Impact on websites: tens of millions of agents filling out forms simultaneously sounds a lot like a bot flood. How will CAPTCHAs, paywalls, and rate limiters react?

One thing is certain: the battle of consumer agents is officially on. OpenAI has Operator, Anthropic has Computer Use, and Google is now deploying Spark in the only browser used by 3.5 billion people. The winner won't be the one with the smartest model — it will be the one with the most comprehensive execution surface.