Gmail's Gemini agent for automatic email replies discovered

Summary

Decompiling the APK for Gmail version 2026.09.28 revealed code strings indicating that Google is developing a Gemini agent capable of drafting email replies, flagging tasks that require human intervention, and requesting manual approval before sending specific messages.

Why it matters

Email remains the dominant workflow surface for knowledge workers. An agentic layer that triages the inbox and drafts replies could significantly reduce daily cognitive load. The design keeping humans in the loop (via "input required" prompts and pre-send approvals) suggests Google is attempting to manage the trust and liability risks associated with autonomous email actions. However, this is still a code leak, not a released product, and the timeline and scope remain uncertain.

Highlights

• An AI inbox is expected to feature a "Use Gemini" button next to action items. This button will trigger inbox scanning and "Draft Reply" prompts.
• The agent transitions to an "Input Required" state when uncertain, accompanied by an "Provide Input" button.
• A human approval gate for specific outgoing emails, reducing the risk of fully autonomous sending.

Critical take

The entire report relies on hidden code comments and string constants within a single APK build. Google routinely distributes dead or experimental strings that never reach users. Therefore, the feature's practical existence and launch window have not yet been verified. The 35-point AI score in the source text and speculative language (e.g., "maybe," "likely") emphasize that this is early-stage telemetry, not a product announcement.