Key Takeaways

• Code in Google App v17.20 reveals a new task scheduler that could let Google’s Gemini-powered AI agent run tasks on a timed schedule, not just on demand
• A dedicated skill system suggests the agent will gain new capabilities through an extensible series of modules
• Google has built a multi-layer security gate to limit which devices gain access to the full agent, currently only recent Pixel and Galaxy hardware qualifies.

Google’s Gemini-powered screen automation feature arrived in March of this year for Pixel 10 series and Galaxy S26 users, letting AI perform simple on-screen tasks, like booking a ride, automatically. However, a deep analysis of Android’s Google App (version 17.20) reveals hidden code for a much more powerful agentic system, including an always-running background service, a dedicated skill system and a task scheduler.

While the tech world is focused on the newly-surfaced “Gemini Spark” UI in a later version of the Google app, this teardown of the v17.20 code reveals how the new agent operates under the hood.

In the file resources.arsc, we find the following strings:

• `assistant_robin_agent_tab_schedules`
• `assistant_robin_agent_tab_skills`
• `assistant_robin_agent_tab_tasks`

These point to new Gemini Intelligence features within the app.

Google’s Gemini Intelligence Agent: What Do We Know Already?

Back in February of this year 9to5Google found a new Gemini agent, codenamed “Bonobo,” in Google app 17.4 beta. A month later, Google officially launched the screen automation tool exclusively on Pixel 10 series and Samsung Galaxy S26, eventually branding it “Gemini Intelligence” on May 12. The feature works by taking screenshots automatically and interacting with the user interface just like a human user would, with virtual taps and swipes.

Now, a deeper look at the app code has revealed much more about where Gemini Intelligence is headed ahead of Google I/O 2026.

What’s Next For Gemini Intelligence?

The text string “agent_schedules,” found within the code, strongly suggests that Gemini Intelligence will be able to perform tasks to a planned schedule, not just on demand. This means the agent will likely be able to plan tasks for execution at a later date and time, becoming proactive rather than reactive in its actions.

This could mean the difference between asking “book me a ride,” and something like “monitor flight prices every morning and tell me if they drop.” The crucial distinction is that scheduled tasks allow an agent to work on your behalf even when you’re not using your phone.

Within the source file classes2.dex, we find:

• `ROBIN_AGENT_MODE_FG_SERVICE_STARTED`
• `ROBIN_AGENT_MODE_FG_SERVICE_END`
• `ROBIN_AGENT_MODE_FG_SERVICE_FAILURE`

The text “AGENT_MODE_FG_SERVICE” within the code reveals that the agent is designed to run continuously in the background, similar to other system services like turn-by-turn navigation or music playback.

The text string “agent_skills” reveals code infrastructure for a skills-based system, suggesting the agent will be able to tap into specific abilities similar in concept to mini apps for use just by your AI agent. We don’t know at this stage what skills Google may have in store for the agent but the architecture is in place for Google to add future app integrations as modular skills.

Two complementary systems exist within Gemini Intelligence: “Robin,” the voice agent that also works on older Pixels and “Bonobo,” which enables control of the screen and requires Android 16. Eligibility for these two services is evaluated separately, suggesting that some devices that don’t qualify for Bonobo may still receive the updated “Robin” voice agent.

Will The Pixel 9 Get Gemini Intelligence?

Almost certainly not at this stage. Only certain phones will get the full Gemini Intelligence experience, and it looks like bad news for Pixel 9 owners. The code currently checks your device against multiple criteria before the agent can be enabled. For now, this means Gemini Intelligence screen automation is limited to Google’s own Pixel 10 series and the Samsung Galaxy S26 range, running at least Android 16. Older Pixel devices appear to qualify for the voice-based agent, but not for the screen control features.

Google’s multi-layer checks validate everything from the device model to a server-side flag Google can toggle remotely. The code also checks the user’s language against a list of supported languages stored on Google’s servers, meaning the company can expand the feature with new language options without updating the app.

Which Phones Support Gemini Intelligence?

At first, compatibility is limited to Pixel 10 and Galaxy S26 series, although select flagships also meet the minimum requirements and could follow soon.

The reasons for these restrictions boil down to a stringent list of qualifying requirements detailed in a footnote on Google’s official Gemini Intelligence page:

• Nano v3 or greater on-device AI model.
• 12GB+ RAM
• Flagship SoC
• Must pass a test suite on Android 17+
• 5 OS upgrades, AVF, pKVM and 6 years of quarterly security updates.

Included in that list is a requirement for a device with the Nano v3 model installed. This includes the Pixel 10 range, Galaxy S26 series, OnePlus 15 and others, but excludes the Pixel 9 series, Galaxy Z Fold7 and OnePlus 13 which use Nano v2. The door is still open for Google to validate devices on the first list, but it looks like those stuck on Nano v2 will remain locked out of Gemini Intelligence unless Google decides to update the AI model.

Big Gemini Intelligence Updates: Why Do They Matter?

While it’s well known what Gemini Intelligence can do today, the latest app code reveals Google’s future plans to turn it into a persistent, scheduled, skills-based platform.

Google I/O starts May 19. I’ll be watching to see what enhanced Gemini Intelligence features Google announces.