Google Debuts Gemini Spark AI Agent For Pro And Ultra Subscribers In Singapore

From large learning models to agentic experiences, the artificial intelligence (AI) era is all about streamlining workflows and eliminating repetition for improved efficiency. Google has certainly embraced the push, having rolled out its biggest search engine upgrade in over 25 years, and its efforts will continue with Gemini Spark, described as the user’s “24/7 personal AI agent.”

Google Gemini Spark
Photo credit: Kirill Kudryavtsev (AFP via Getty Images)

First announced at the company’s annual Google I/O event in May and powered by the Gemini 3.6 Flash model, it can perform tasks, monitor conditions, and take actions on a user’s behalf — even during periods of inactivity, such as when they are sleeping or when the screen is locked. While that may raise concerns about rogue agents or the lack of control, the tech giant assures that users can decide whether to enable Spark and limit its access to applications, with the software seeking permission before executing “high-stakes” actions, including sending emails or spending money.

Advertisement ▼

The agent was previously launched in the U.S., but is now available to Google AI Ultra subscribers in Singapore and will expand to AI Pro users in the coming weeks. It appears as a dedicated app within the Gemini app on the web and runs on the company’s cloud infrastructure, requiring zero setup by connecting natively with Google Workspace tools like Gmail, Docs, and Sheets, which offers an advantage over open-source alternatives.

Spark is built around three key elements: Tasks, Schedules, and Skills. The first is similar to setting an objective that can range in complexity, from organising emails to planning an itinerary for an overseas trip across different platforms, while Schedules determines when the agent should carry out a task, proving handy for automating routine or recurring activities. As for Skills, it works by teaching Spark how to perform a particular task via a reusable set of instructions (think a template of sorts), so users can apply it to various Tasks. There are some caveats, including the inability to integrate multiple inboxes and limited capability to detect images and PDF attachments in emails, though Google says the latter will be fixed in three weeks.