Google is building agentic Gemini tasks into Gmail — letting the AI draft replies, act on action items, and surface work that needs human review. The feature isn’t live yet, but strings in Gmail v2026.09.28 confirm it’s in active development. And it’s designed to pause, not plough ahead: Gemini won’t auto-send replies or close tasks without explicit user approval. Instead, it flags “Needs your input” when uncertain, shows “Ready for Review” after completing work, and offers a “Give input” button to guide it forward. A “Start with Gemini” button will appear beside suggested action items, like “Book meeting with Alex” or “Follow up on invoice”. Turning them into AI-managed workflows. The rollout timing isn’t public. No launch date, no beta sign-up, no regional availability window. Just code, and intent. This is Gemini moving beyond summarising and suggesting, into doing. Not autonomously. Not silently. But as a co-pilot that waits for the nod before hitting send or marking ‘done’. Next step? Google has to ship it. Then users in the US, India, and elsewhere will get their first real test of what ‘agentic’ means in an inbox. And whether they trust it to act, not just advise.