Google tests a five-template AI agent builder inside Gemini for Android

Why it matters

Android could turn agent creation into a mass-market Gemini feature, giving Google distribution and service integrations that standalone consumer-agent startups cannot easily match.

Google is testing a "New agent" tool inside Gemini for Android that could let users create task-specific AI agents from a phone or tablet, according to an August 3rd APK teardown by Android Authority.

The interface appeared in version 17.45.14 of the Google app, which provides access to Gemini on Android. The code offers evidence of a product under development rather than a public release: Google has not announced the mobile builder, and features found in unfinished app code can change or disappear before launch.

The test extends the consumer agent strategy overseen by Josh Woodward, the Google vice president who leads Google Labs, the Gemini app and AI Studio. Woodward previously helped start Google's Next Billion Users effort and worked on Chromebooks, according to his official Google biography. Putting agent creation in the Android app would follow the same distribution playbook: reduce a technical product to a familiar mobile interface and place it in front of Google's existing users.

Five agents, with limited customization

The unfinished Android flow presents five templates:

• Family coordinator: Plans chores, events and shared household tasks.
• Tutor: Explains study topics, creates quizzes and tracks learning progress.
• Group organizer: Coordinates outings and gatherings involving several people.
• Marketing agent: Produces material for social posts, newsletters and campaigns.
• Finance agent: Tracks spending and provides bill-payment reminders.

Android Authority found the "New agent" entry beneath recent conversations in Gemini's sidebar. In the version examined, users could select a template but could not describe an arbitrary job for the agent, as customers can through Google's web-based enterprise tools.

That constraint makes the current mobile design closer to a template gallery than a general agent builder. It may also make the feature easier to introduce to consumers who have never configured tools, instructions, permissions or data sources for an AI agent. Google could use the templates to teach the concept before exposing the more complicated controls available to organizations.

Google's existing Gemini Enterprise agent platform allows employees to create no-code agents with Agent Designer, deploy agents built with Google's development tools and manage agents supplied by Google or outside partners. The product emphasizes centralized administration and governance, requirements that carry less weight for a household coordinator or personal tutor.

Google has the distribution advantage

A consumer release would give Google a direct route from agent creation into Android, Gmail, Calendar and other Google services. The useful version of a family, finance or scheduling agent depends on access to that context. The permissions and data boundaries Google chooses would determine whether the templates can complete work or merely produce advice in a chat window.

Google has already started moving persistent agents toward consumers. At Google I/O on May 19th, CEO Sundar Pichai introduced Gemini Spark, an always-running personal agent hosted on Google Cloud and designed to act across connected tools. Google initially placed Spark with trusted testers and paid subscribers. In July, Pichai said Spark had expanded in the US and internationally.

Alphabet also said in its July 22nd earnings remarks that the Gemini app had reached 950 million monthly active users and that its daily active user count had tripled over the preceding year. Those are company-reported figures, but they show the scale behind Google's agent push. Google does not need to persuade users to install a separate agent product if creation becomes another option inside Gemini.

The approach also puts pressure on independent startups selling consumer agent wrappers. A startup can still compete through deeper workflows, cross-platform support or specialized integrations. Google can bundle a basic agent builder into a mobile product with hundreds of millions of reported users and connect it to services many of those users already rely on.

Google is entering a market where customization is already established. OpenAI's GPTs let users configure no-code versions of ChatGPT with their own instructions, knowledge and capabilities. Microsoft's Agent Library offers customizable templates for Copilot Studio and Microsoft 365 Copilot, while Anthropic's Claude Skills package instructions, scripts and resources for repeatable tasks.

The Android test suggests Google is preparing a simpler entry point built around recognizable personal jobs. Its significance depends on whether Google eventually lets users edit the templates, connect outside tools, share agents or build their own from a blank prompt. For now, the code establishes only that Google is experimenting with mobile agent creation, not that free Gemini users will receive it.