On July 29, 2026, Google began rolling out Gemini Spark to Google AI Pro and Ultra subscribers in India. If you have not seen it yet, you likely will in the coming weeks: Google has confirmed the feature is expanding across the country on a staggered basis. It is worth five minutes of your attention, because this is not another chatbot update. It is Google's first serious attempt to make Gemini work while you are not looking at it.
What actually launched
Spark was first announced globally at Google I/O in May 2026. The India rollout that started on July 29 makes it available to Google AI Pro subscribers, priced at roughly Rs 1,950 a month, and Google AI Ultra subscribers, priced from around Rs 6,500 a month. It is not part of the free Gemini app and not part of the cheaper Google AI Plus plan, so most casual users will not have it by default.
The core idea is simple to state and genuinely different from what came before. A normal AI chat waits for you to type something, answers, and then stops. Spark instead runs on Google's cloud infrastructure and keeps working on a task even after you close your laptop or lock your phone. You give it an outcome, optionally a schedule, and it goes and does the multi-step work in the background: reading emails, pulling data across Gmail, Drive, Docs and Sheets, and reporting back.
The shift is from an AI that answers questions to an AI that finishes tasks. That single change is why this deserves attention even if you have ignored the last five product updates from Google.
What it can actually do for a working day
Google's own examples are mostly personal, not corporate, which tells you where the product is aimed for now: pulling flight and hotel confirmations out of your inbox into a trip spreadsheet, compiling wedding RSVP emails into a table with dietary preferences, drafting a summary email to a caterer based on that table. None of that requires you to open ten emails and copy numbers by hand.
The office version of the same pattern is easy to picture. A recurring MIS email that arrives every Monday could be turned into a standing task: pull the numbers, drop them into a tracking sheet, flag anything that moved more than 10 percent. A pile of vendor invoice emails could become a running expense sheet without you touching each one. The mechanism is the same as the wedding example; only the data changes.
Notice the instruction spells out what Spark should not do. That is not optional caution, it is how the product is designed to work: Google says Spark checks in with you before high-stakes actions like spending money or sending an email on your behalf. Read-only, data-gathering tasks like the one above are the safest place to start.
The one setting to be careful with
Around July 30, Google extended Spark into desktop Chrome, letting it browse the actual web on your behalf, including sites where you are already logged in. With your permission, it can also use passwords saved in Chrome's password manager to complete a task, such as checking a flight status or starting a booking. Google says it has built in protections against prompt injection, a known attack where instructions hidden in a webpage or email trick an agent into doing something you never asked for. It has not published the technical detail of how those protections work or how they hold up against real attempts.
A quick filter before you switch this on
An unpublished safeguard is not the same as no risk, and it is not the same as a broken product either. It simply means the honest answer is "we do not fully know yet". Until independent testing catches up, a sensible rule for office use: let Spark read and summarise things, but do not connect it to accounts holding client data, banking credentials or company financial systems, and do not enable the saved-password browsing feature on any device used for work. Personal, low-stakes tasks are a reasonable place to try it. Confidential work data is not, not yet.
This is exactly the kind of judgment call that separates people who get real value from new AI features from people who either avoid them entirely or adopt them without reading past the demo. If you want an honest read on where your own AI judgment stands, the AiM AI Fluency Test scores exactly this kind of decision-making across 18 real tasks, in about 12 minutes, for free.
What to actually do this week
If you already pay for Google AI Pro or Ultra, check whether Spark has appeared for your account under the Gemini app. If it has, try one low-stakes, read-only task first, something like sorting a personal inbox or building a simple tracking sheet from emails you would not mind an AI reading anyway. Watch what it gets right and where it needs correcting before you trust it with anything recurring. If you do not have access yet, there is no urgency: the rollout is gradual, and the feature will still be there next month.
If you want a guided walkthrough of what agentic tools like this mean for everyday office work, and how to set them up without creating a data risk you did not intend, join AiM's free live webinar, Gen AI for All, on Sunday, 9 August 2026, 3 to 5 PM IST.
Google has been clear that Spark is an early release, and the gap between the flight-booking demo and a dependable office tool will close over the next few months, not days. The professionals who benefit most will be the ones who tested it early on low-stakes tasks, not the ones who either ignored it or handed it their whole inbox on day one.