Gemini will bring a true AI assistant. It will learn from your habits and monitor your emails and meetings for you

Large tech companies are slowly but surely moving from passive chatbots to active assistants that can actually do things for users. While Anthropic offers the Cowork tool and OpenAI is developing its Agent platform, Google has an ace up its sleeve. Shortly before the start of the I/O 2026 developer conference, details leaked about what its answer within the Gemini ecosystem will look like.

Gemini Spark as an assistant always at your fingertips

Details about Gemini Spark emerged from multiple sources simultaneously. Screenshots appeared on the X platform from various users, illustrating what the AI agent will be capable of, and further details came from sources like 9to5Google (which states it encountered the new feature in the beta version of the Google app, number 17.23).

Spark is expected to reside in the app as a separate section in the navigation menu, and upon opening it, two tabs will be displayed – one for classic chat and the other for the agent itself. The leaked screenshots indicate that you will be able to set one-time tasks or schedule recurring ones that will launch at specifically designated times. Google itself describes the new feature as an assistant available “24 hours a day, 7 days a week.”

What will Gemini Spark be able to do?

In the leaked welcome guide, Google provides specific examples of activities Spark is intended to help users with. These include, for instance, email inbox cleanup – the agent should be able to summarize or archive newsletters or unsubscribe from annoying promotional mailings itself. Before important meetings, it should be able to generate a concise overview of relevant information to help the user quickly get oriented. The third scenario mentioned is a personalized news summary from around the world, focusing on topics of interest to the individual and tracking their development over time.

Gemini Spark 👀 pic.twitter.com/2kuFoHXEol

— Fandu (@mrfanduuuuu) May 14, 2026

For Spark to be capable of this, it must know practically everything about the user. Therefore, it will draw from connected applications, created skills, chat history, scheduled tasks, websites where the user is logged in, Personal Intelligence features, location, and other data sources – thus reaching into areas where Gemini could not yet look.

Skills will play an important role

One of the most interesting things that has emerged so far in connection with Spark is the system of so-called Skills. They could function on a similar principle to Projects in Claude – the user can define a set of instructions for recurring tasks, into which they then insert only changing variables via a prompt. The TestingCatalog server points out that skills can currently only be created by copying text; Google is not yet preparing the import of files in Markdown format, which is popular for these purposes.

In addition to skills, Spark should be able to handle multi-step tasks across applications and combine information from various sources. So far, the leaked images primarily show Google Workspace services, but support for third-party applications is not ruled out. TestingCatalog also notes that Spark could also control the Chrome browser in agent mode and work with files stored on the user’s device. However, full control of the entire computer, as offered by competitors Claude Cowork from Anthropic or OpenClaw, will not be available at launch.

What about privacy?

In the accompanying texts, Google openly admits that Gemini Spark is experimental and should not be approached without caution. While the assistant is supposed to ask for permission before sensitive actions, the creators also cover their backs by stating that it may, in certain situations, share information or even make purchases without explicit user confirmation. At the same time, users are warned against turning to Spark for financial, health, or legal advice.

In its work, the agent is supposed to store data from the remote browser, including login credentials, so that it doesn’t have to re-establish sessions every time. According to Google, everything can be deleted or turned off in the settings, where management of connected applications and Gemini activity will also be available.

Officially (perhaps) as early as next week

Spark is currently gradually appearing to a select handful of testers who have shared their findings online. The public beta version is expected to arrive around the main presentation at Google I/O 2026, which will take place on May 19. Its predecessor, Gemini Agent, was available exclusively to subscribers of the premium Google AI Ultra plan, but according to speculation, Google wants to make the new feature accessible to a wider group of users through the allegedly upcoming AI Ultra Lite plan. It will be priced between AI Ultra and AI Pro, between which there is currently a gap of several thousand. However, official confirmation is missing, so take everything with a grain of salt.

The question remains regarding the agent’s availability in the Czech Republic. Previous advanced Gemini features have several times arrived on our market with a delay of several months.

Would you entrust an AI agent with access to your emails and private data?

Sources: 9to5Google, Android Authority, TestingCatalog