Gemini Agent: Google's answer to autonomy.
In May 2026 Google consolidated its agent story: Project Mariner was retired as a standalone product, and its technology now powers Gemini Agent — the unified agentic mode inside the Gemini app. It runs multi-step tasks across Workspace, Deep Research, and the live web. Its real edge is orchestrating several apps in one prompt — and the skill is staying in control of the steps that matter.
Task mode is now free — and it can call, pay and text.
Google says task handling in Gemini is rolling out free in the U.S., Japan, Korea, Canada, Australia and India, running on its Gemini 4 Hydrogen model, with a set of advanced features starting in the U.S. You switch the toggle on the Gemini home screen from Chat to Task. New abilities: Gemini can phone businesses for you (it says up front that it is an AI, and businesses can opt out), buy things through Google Pay with a spending cap you set and a confirmation before it clicks Buy, sign in to sites such as a utility portal through Google Password Manager with your permission, and take instructions by text message or by forwarding email to gemini@google.com (U.S. first; set up under Settings → Channels). Google says each task runs in an isolated virtual machine dedicated to your account, and that it pauses before critical steps. It also admits tasks can still fail and some sites will hand the final click back to you. The two checkpoints below matter more now that money and phone calls are in scope: keep a spending cap on, and save “Always allow” for chores you would happily repeat without looking. Source: Google’s announcement.
One prompt, many apps, several artifacts.
Other agents browse and act. Gemini Agent's edge is Workspace orchestration: a single task that searches your Drive and Gmail, then creates a Doc, a Sheet, a Calendar event, and a drafted email — together. Power like that means two checkpoints you own: confirm before anything irreversible, and verify what it shares and at what permission. Watch it run.
Run a Workspace orchestration.
One prompt: "Set up the kickoff package for Project Atlas." Hit run and watch Gemini Agent build across apps — until it reaches the one step it won't take without you.
Where Gemini Agent earns the delegation.
Multi-step web tasks
The Mariner heritage: plan, browse across sites, verify. "Find 3 restaurants with a private room, 4.3+ rating, and check next-Tuesday availability — don't book."
Workspace orchestration
Docs + Sheets + Gmail + Calendar in one go — the kickoff package you just watched.
Long-running background tasks
"Run a 30–60 min competitive analysis in the background, save to Drive, and ping me when done."
Hand Agent a real Workspace project
Pick a project that needs several Workspace files made. Use Gemini Agent to set up the whole package in one prompt — then review the share permissions and the drafted email before anything goes out. Notice where it's faster than building each file by hand, and where you'd still rather do it yourself.
What you can do now
- Use Gemini Agent for multi-step orchestration, not single questions
- Verify share permissions (view / comment / edit) match your intent
- Run long tasks in the background; don't watch them work
- Expect confirmation prompts before irreversible actions — and keep email as draft
- Combine Workspace integration with research for richer outputs