Google has launched Gemini Spark, a new AI agent built to take entire assignments off users' hands — conducting research, writing code, preparing reports, and returning with the finished product. The service went live on Thursday through Google Cloud, offering enterprises an AI system designed to complete work end to end, according to a Google Cloud blog post.

As described on Google's website, users can assign work today and have it finished long after they have shut down their laptops. The company frames this as a break from the familiar pattern of delegating a single task, such as drafting an email, while handling everything else personally.

The launch places Alphabet (GOOGL) alongside rivals that have made similar moves. In September, OpenAI introduced its dots agents, which work for users on a continuing basis, and Meta (META) followed with Muse, a system that can handle purchases, travel reservations, emails, and payments. Alphabet's shares gained 2.2% following Thursday's announcement. With Alphabet, OpenAI, and Meta now all offering systems intended to work on users' behalf beyond a single request, the industry's emphasis is shifting from one-off assistance toward software that carries an assignment through to completion.

Working across apps without constant supervision

The notable aspect of Gemini Spark is how much responsibility Google wants businesses to hand over. In a scenario described by the company, an employee who needs information pulled from several departments, organized into a spreadsheet, and turned into a presentation-ready report can assign the entire job to Gemini instead of requesting each step separately.

The agent determines what needs to be done, identifies the appropriate software, collects the necessary information, and prepares the requested output. For business users, that means fewer intermediate prompts: instead of steering an assistant through each step, an employee describes the outcome and the agent handles the sequence in between.

Gemini can operate inside Gmail, Docs, Sheets, and Calendar, but Google is not keeping it locked inside Workspace: companies using Microsoft 365 or Slack can also access the service. That flexibility extends to devices as well. Employees can use the agent through browsers, iPhones, Android phones, Windows computers, Macs, and even developer command lines, and software companies can connect it to their own products without requiring users to open a separate Gemini window.

Google has also addressed one of the more irritating aspects of working with AI: having to explain yourself repeatedly. Because Gemini runs on Google Cloud, it maintains the same working memory and context wherever an employee accesses it from. A user could start a task on an office PC, track its progress from a smartphone, and continue the next day without restarting.

Most importantly, long-running assignments do not depend on a device staying powered on. Employees can schedule tasks to be performed later, or have the agent initiate them when certain events occur within a connected system. That independence from any single machine is what makes the assign-now, receive-later model Google describes workable outside an employee's own working hours.

Google has additionally built in a way for Gemini to divide difficult assignments among several smaller AI workers. These temporary sub-agents can tackle different parts of a project simultaneously or complete them in a particular order, with Gemini managing their activities and combining the results.

AI coworkers with their own corporate identities

Google is extending the concept with what it callser agents, through which companies can create digital workers with ongoing responsibilities, individual storage, and even company email addresses ending in @agents.company.com. An AI coworker could handle different assignments over several days, communicate with other agents, and maintain its own working history.

There are boundaries. Google says each coworker receives access only to the company information employees choose to share with it. Businesses can also establish dedicated project environments, allowing Gemini to build familiarity with particular tools, instructions, and workflows. The system uses past interactions to better understand how someone works, rather than treating every request as an entirely new conversation.

Perhaps the more unusual decision is that Google is not making Gemini depend exclusively on its own AI models. The platform currently uses both Gemini and Anthropic's Claude, selecting whichever is better suited to a particular assignment. A complicated coding problem might require a more capable model, while something straightforward could run on a cheaper one.

Google argues that this flexibility could help companies control expenses without sacrificing performance on demanding work. It also intends to introduce other commercial and openly available models as the technology develops, meaning businesses would be able to switch between AI engines while keeping their existing information, configurations, and work history. For enterprises weighing agent platforms, that stated plan — swapping underlying models without losing stored context and setups — is a concrete development to watch as the lineup expands.

Source: Cryptopolitan