Google's AI Assistant to Make Phone Calls for Users

Google’s AI assistant, Gemini, is reportedly being developed to manage phone conversations on behalf of users. An analysis of an Android Authority APK file uncovered code that outlines features where Gemini could contact personal numbers to relay messages or seek information and provide a summary. This functionality remains unofficial, as teardown analyses typically reflect work-in-progress rather than finalized products. A limited trial named Call for Me has already launched, enabling the system to reach businesses for checking operating hours, pricing, inventory, or making reservations. This service is currently restricted to Pixel 11 device users located in the United States.

Gemini Taking on Verbal Tasks

The AI would generate spoken responses instead of merely suggesting text. For instance, an assistant could draft a group message and then call members to confirm attendance for an event on Saturday. In Kenya, phone conversations remain critical in daily routines, as many tasks, from contacting plumbers who ignore messaging apps to relatives who say “nitakupigia” and remain unreachable, are better handled verbally. These scenarios contrast with Western examples like booking flights or reservations, which dominate Silicon Valley’s AI use cases.

The true potential of AI agents might stem from managing complex, interconnected tasks rather than speeding up routine activities. Expanding the system to make calls allows it to engage individuals outside digital platforms, which is significant in regions where offline interactions dominate daily life. This trend reflects a shift toward automation tools designed to handle complex exchanges across various channels, including voice.

Potential Pitfalls in AI-Mediated Calls

A failed restaurant reservation disrupts a single evening, but an AI error in a conversation with a family member could have deeper consequences. The system might misinterpret spoken words, overlook humor, turn casual inquiries into demands, or falsely confirm commitments. Google’s documentation already notes the AI’s potential to misread responses. Language diversity poses another challenge; while many Kenyans fluidly switch between English, Swahili, and Sheng during calls, the current test version supports English only. Additionally, extended interactions could incur extra costs through standard carrier plans, affecting users’ airtime or bundled minutes.

Delegating Difficult Conversations to AI

For individuals who struggle with verbal communication due to speech impediments or anxiety, allowing AI to speak on their behalf could provide meaningful assistance. However, defining acceptable use cases will determine whether this technology feels empowering or invasive. Users may benefit from features like reviewing draft messages before calls, halting conversations mid-way, or assuming control when needed. Additionally, recipients should easily decline to continue the interaction.