Gemini Agent is Google's name for Gemini doing multi-step jobs for you, and it now lives inside tasks in Gemini. You describe a goal and when you want it done, and Gemini works through it using your connected apps, websites and a browser. You need Google AI Pro or Ultra, a personal Google Account and to be 18 or over, and for now Google doesn't offer tasks in the UK, the European Economic Area, Switzerland or Nigeria.
If you go looking for Google's Gemini Agent help article, it now opens as Manage your tasks and workflows in Gemini Apps. Google's help page also answers the question of where Spark went: it tested Spark in beta with subscribers from May, people kept calling what they wanted "tasks", so Google renamed it Task. Your existing tasks and schedules are still there, in the sidebar. The idea's the same: you hand Gemini a job, not a question. Here's how it works, what it can reach and the rules Google itself gives for using it safely. If you'd rather build your own agents than use a ready-made one, LangChain and CrewAI compared looks at two popular frameworks.
What Gemini can do as an agent
Google's examples fall into three groups:
- Saving time: decluttering your inbox by summarising or archiving newsletters, writing an urgent to-do list after a holiday, or unsubscribing from email lists.
- Watching prices: monitoring flights and hotels and alerting you when prices drop within your budget.
- Research and files: pulling research with cited sources and packaging it into downloadable PDFs, documents, spreadsheets and presentations.
To do this, Gemini uses your Connected Apps, skills, past chats, websites you're signed into, Personal Intelligence and location.
Inside Google's own apps, the Gemini Apps help page on tasks spells out what it can touch. Workspace has to be connected to Gemini Apps through Connected Apps first.
| App | What Gemini can do in a task |
|---|---|
| Calendar | Check your calendar, RSVP to invitations, schedule new events, update times or locations, cancel events and suggest open slots that fit everyone |
| Gmail | Search and summarise threads, draft emails, reply, forward and organise your inbox with labels |
| Docs, Sheets, Slides and Drive | Search, read, rename or delete Drive files; write and edit documents; build spreadsheets with tables and formulas; create and edit presentations; add images and Sheets charts |
| Keep | Find notes and lists, and create private ones |
| Google Tasks | Find, add, update and delete reminders and tasks |
Two of those rows behave differently. Edits to shared documents, spreadsheets and presentations wait for you to review and confirm Gemini's planned changes. Google Tasks doesn't: Google says Gemini can make bulk changes to your private tasks there without asking, so read your request twice before you send it.
Who can use it
| Requirement | Detail |
|---|---|
| Plan | Google AI Pro or Ultra |
| Age | 18 or over |
| Account | Personal Google Account; not yet available on work or school accounts |
| Where | Gemini mobile app, Gemini app on Mac, and gemini.google.com |
For what each plan includes, see Google AI Pro explained.
Google's help page lists one more limit that matters if you're reading this in Britain: for now, tasks in Gemini are available wherever Gemini Apps are, except the European Economic Area, Nigeria, Switzerland and the United Kingdom.
If you cancel your Google AI subscription or drop to a plan without Tasks in Gemini, you lose access to tasks and schedules, but Google keeps your data. Tasks already running will finish, and schedules are paused rather than deleted, so they come back if you upgrade or resubscribe. Skills don't need a subscription at all and still work in ordinary chats.
How to start a task
- Open Gemini on the web, the Mac app or your phone.
- Click Task. Google's steps put it below the text box, and its note on the Spark rename describes it as the Chat and Task toggle next to the prompt box, so look for the switch between Chat and Task.
- Describe what you want done and when, such as around a time or after an event.
- Type @ to pick an existing skill if you want Gemini to use one.
- Open the task thread later to check progress, permissions and schedules.
If you're not sure where to begin, Google suggests typing "set up", "get started" or "interview me" in a task, and Gemini helps you build your first skills and tasks conversationally.
How it uses a browser
This is where Gemini Agent differs from a chat. To work on websites, Gemini either connects to the Chrome browser on your desktop or uses a separate remote browser in the cloud.
| Your local Chrome | Remote cloud browser | |
|---|---|---|
| What it can reach | Every site you can, including ones you're signed into | A separate browser Gemini runs for you |
| What you need to do | Keep your device and Chrome awake and running | Nothing; it keeps going if you close your device |
| On mobile | Not used | The only option |
If you close your laptop mid-task, Gemini may switch to the remote browser to finish. For steps like signing in to a website, the task pauses until you take over. With your permission it can also use logins saved in Password Manager. The first time it browses on a device it asks permission, and after that it asks for confirmation on every task that involves browsing.
I'd also point you to Googlebook.
Google's own safety rules
- Don't type sign-in details, payment details or anything sensitive into the task thread. Take control of the browser and enter them on the webpage instead.
- Avoid scheduling tasks you'd consider sensitive. Google warns that if a schedule runs while you're offline, you may not be able to stop an unintended action.
- Some tasks ask for your confirmation before Gemini completes them. Google names sending messages, changing or deleting your data, making purchases, financial transactions, submitting web forms and signing in with Google as actions that need your review.
- To stop browsing, click Stop in the Gemini in Chrome side panel. Gemini may then try another route or abandon the task. To cancel completely, click Stop in the Gemini text box.
Google also says schedules in tasks are different from schedules in chats. If you just want a prompt answered at the same time every day, Gemini scheduled actions are the simpler tool.
The prompt injection warning
Google is unusually direct about the main risk. A website, email, document or file can hide instructions meant to mislead an AI agent. If Gemini reads them mid-task, Google says it could be tricked into posting your private information on a public website, sending your Gmail messages to an outside service or exposing insights from your Connected Apps. Gemini also shares the information a task needs with the websites it uses, which can include your name, contact details and files.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
In practice: give agent tasks a narrow job on sites you trust, and check what it did in the task thread afterwards. To undo something Gemini did in a Workspace app, Google says to click the link in Gemini's confirmation and change it in that app.
Tasks aren't available on work or school accounts yet, so if you're using it at work, read Gemini Enterprise for the business version instead.
Managing tasks and files
Each task has a thread and a work panel. From the thread's menu you can pause or resume its schedule, and on the web there's also a Scheduled page for this. When Gemini creates Markdown or plain text files, it shows them in Canvas so you can edit them directly; for other files it gives you links you can open and edit conversationally.
It helps to know the three parts Google uses to describe every job:
| Part | What it is | Google's example |
|---|---|---|
| Task | What you want Gemini to do | Give me an update on AI news |
| Schedule | When it should happen, either a time or a condition | Every day at 8AM, or when my flight is delayed |
| Skill | Reusable instructions and context that teach Gemini how to do a job and which tools to use | Using a Travel Booking skill and a Gmail Writing skill together to rebook a room and send a confirmation |
Put together, Google's flight example reads: "When my flight is delayed, notify me and propose an update to my itinerary." You can name the skill you want with @, or let Gemini pick one when it's relevant.
A browser version of the idea is explained in Chrome Auto Browse.
My verdict
Gemini Agent is worth trying for jobs that repeat on a clock or a trigger: Google's own "every day at 8AM, give me an update on AI news" is a good first task, because nothing can go wrong if it misfires. Newsletter clean-ups and price watching come next. Keep payments, logins and anything client-sensitive out of it until you've watched a few tasks run. Supervise first, schedule later, because Google warns that a schedule running while you're offline may complete an unintended action before you can stop it.
If you'd like help choosing which AI tools your team should use and setting them up around the way you work, here's how we can work together.
Frequently asked questions
What happened to Gemini Spark?
Google tested Spark in beta with subscribers from May, then renamed it Task because that was the word people kept using. Your existing tasks and schedules are in the Gemini sidebar.
What happens to my tasks if I cancel Google AI Pro?
You lose access to tasks and schedules, but Google keeps your data: in-progress tasks finish and schedules are paused, not deleted, until you resubscribe. Skills keep working in chats.
Does Gemini Agent work on my phone?
Yes. Tasks are available in the Gemini mobile app, the Gemini app on Mac and gemini.google.com. On mobile, Gemini uses a remote browser in the cloud.
Can Gemini Agent use websites I'm signed into?
If it uses your local Chrome browser, yes, it can reach the same sites you can. With your permission it can also use logins saved in Password Manager.
What can Gemini do in Gmail and Calendar for a task?
In Gmail it can search, summarise, draft, reply, forward and use labels. In Calendar it can check your schedule, RSVP, create, reschedule or cancel events and suggest open slots.