Google Gemini Spark: the AI agent that keeps working after you close the laptop
Google at its I/O conference on May 19 introduced Spark, an AI assistant that runs in the cloud and performs tasks in Gmail, Docs and Slides without interruption.
Do you actually need AI that keeps working after you close the laptop? If you spend hours in a packed Gmail inbox and repeat the same tasks every Monday, the answer is probably yes. Google on May 19 at its I/O conference introduced Gemini Spark, an assistant that does not wait for you to ask but runs in the background, even when you are offline.
Spark is an upgrade to the Gemini app, not a separate application. It is powered by Gemini 3.5, the new model Google released the same day, and it uses the so-called Antigravity harness, a system that lets the model control multiple tools simultaneously. In practice, that means Spark can open Gmail on its own, find a specific email, rewrite it in Google Docs and send it as a Slides presentation, without further input from you.
1. What it does and what it does not
Spark can currently connect to Workspace tools (Gmail, Docs, Slides, Sheets) and to three external services: Canva, OpenTable and Instacart. You can tell it to compile a list of deadlines from unread emails every Monday morning. Or to go through a credit card statement each month and flag hidden charges. The kinds of things that eat fifteen minutes a week without real value, it does on its own.
What it does not do: it does not change existing documents without your explicit per-step approval, does not pay anything without confirmation, and does not send emails on your behalf unless you directly authorise it.
2. How is this different from regular Gemini
Old Gemini is reactive. You ask a question, you get an answer, session ends. Spark is proactive and persistent. It is launched against your account, stays active on Google servers, and continues a task until it has a result. That is the difference between an assistant you have to constantly tell what to do and one that knows your schedule.
For a typical user, this is the first time an AI tool feels truly like an assistant in the office sense of the word, rather than a chat window where you type questions.
3. Who can use it and what does it cost
Spark is currently available to test users and to subscribers of the Google AI Ultra plan. Ultra costs 250 dollars a month (about 29,300 dinars), a significant amount for an individual but reasonable for a small firm with a handful of employees.
The beta will roll out to all Ultra subscribers in the coming weeks, and Google says it will come to cheaper plans later in the year. The exact date has not been announced.
4. Privacy, because that is the main concern
Spark sees everything that your connected Google accounts see. That is a security risk you must take seriously. Google says the data stays in your Workspace environment and is not used to train models, but if you use Spark with a private Gmail account used for work, we recommend you set up a separate account just for experiments until you see how it behaves.
The second item is connected services such as Instacart. If you give it access to that account, it can order goods on your behalf. Convenient for the kitchen, but it requires vigilance.
5. Direct competitor to Anthropic’s Claude
Spark arrives exactly three months after Anthropic introduced similar agentic capabilities to Claude Sonnet 4.6. The main difference is integration: Spark is immediately connected to the Google ecosystem, while Claude works with files you provide. For users already living in Gmail, Spark will be the more natural choice. For those working with a diverse set of tools, Claude remains the more reliable bet.
OpenAI has not yet shipped an equivalent, although similar features are flagged for GPT-5.5 Pro later this year.
Our recommendation
If you already pay for Google Workspace at work and your company has the Ultra plan, it is worth trying Spark on one concrete repetitive task. Not all your work, one task. After a week, see whether it freed up at least half an hour. If yes, it is worth it.
For an individual already paying for ChatGPT Plus or Claude Pro, Spark is not yet enough of a reason to switch. It is worth waiting for cheaper plans or for concrete use cases that cannot be solved with other tools.
Google’s next move is expected in Q3, when Gemini 3.5 should also become available in the API for development teams outside the Ultra plan.
