Three-file agent13 minAn inbox agent that labels every email and drafts replies in your voice, built in plain English inside Claude Code and then put on an hourly cloud schedule. Here's the four-level method, the three files that make it work, and the safety settings you shouldn't skip.Read now →

Google Just Gave Millions of People a Free AI Agent You Can Text. Its Biggest Weakness Is Its Name
Google turned its paid Spark agent into 'Tasks' in Gemini and made it free in six countries. In the US you can text it, email it, let it shop with Google Pay and soon have it phone businesses. Here's what it does well, where it stumbles, how to set it up in five minutes, and how it stacks up against Meta's Muse.
For most of this year, "personal AI agent" meant paying for one.
Meta's Muse, Anthropic's Claude Duet, OpenAI's dots: all of them promise the same thing. Stop asking an AI for advice and start handing it chores.
On November 17, Google joined the fight in the most Google way possible. It took Spark, the agent it had been selling to Pro and Ultra subscribers since May, renamed it Tasks, put it inside the regular Gemini app and made it free for users in the US, Canada, Australia, India, Japan and Korea.
In the US, it goes further. You can text it like a friend. You can email it. It can buy things with Google Pay after you approve. Google says it will phone businesses for you.
After several days of real-world testing by early users, a clear picture has emerged: it's genuinely useful, surprisingly proactive, a little flaky and saddled with one baffling product decision. Here's the full rundown.
Who gets it, and what it costs
According to The Verge, Tasks in Gemini brings Spark's capabilities to all users in the US, Canada, Japan, Korea, Australia and India, at no cost. You switch the toggle on Gemini's home screen from "Chat" to "Tasks." It's rolling out gradually, so you may not see it yet.
Elsewhere, you'll still need a Pro or Ultra subscription, and some regions, notably the UK and the European Economic Area, are excluded for now.
The US-only extras matter a lot:
- Texting Gemini and emailing tasks to gemini@google.com.
- Calling businesses to compare prices or check inventory.
- Purchasing items with Google Pay.
- Signing into accounts that support Google Password Manager.
Under the hood, Tasks runs on Gemini 4 Hydrogen, Google's small, fast model, which we covered as an orchestrator in Gemini 4 Hydrogen. Google says each user's tasks run in a hardware-isolated virtual machine dedicated to their account, and the agent stops to ask permission before sensitive actions like purchases.
Why Google made it free
Giving away an agent it was charging for six months ago looks generous. It's also strategy.
Personal agents are becoming the new front door to the internet. If your assistant books the restaurant, compares prices and buys the gift, the company that runs the assistant sits between you and every business you deal with. Meta has Muse, OpenAI has dots, Anthropic has Duet. Google can't afford to keep its answer behind a paywall while rivals hand theirs out.
Google also has something competitors don't: the apps. Billions of people already use Gmail, Calendar, Docs, Slides, Sheets and Google Pay. An agent that works natively inside all of them, with no setup, is a powerful default. Making it free turns that installed base into an advantage overnight.
There's a hardware angle too. The Verge reported this week that Google plans to shut off Google Assistant on smart speakers next year. The direction is clear: the old command-style assistant is giving way to Gemini, and Tasks is the part of Gemini that actually does things.
How Google says it keeps your data separate
Any agent that reads your email and spends your money has to answer one question first: where does the work happen?
Google's answer is that Tasks runs in a hardware-isolated virtual machine dedicated to your account and never shared with other users. In plain English: each person's agent works in its own sealed-off computer in Google's cloud, so one user's session shouldn't be able to see another's.
On top of that, Google says the agent pauses for permission before sensitive steps such as purchases, and you can give it new instructions while it's working. That human checkpoint is the most important safety feature in any agent, so don't click through it on autopilot.
None of this makes an agent risk-free. Prompt injection, where a malicious web page or email tries to give the agent hidden instructions, remains an industry-wide problem. Isolation limits the blast radius; your judgment at the approval step does the rest.
An agent you give work to
The difference between a chatbot and an agent is the verb.
Instead of asking how should I manage my inbox?, you tell it: triage my inbox and draft replies. Instead of what should I cook this week?, you say: plan dinners and add the ingredients to my Instacart.
That part isn't new. Spark could already do it. What's new is how easy Google has made it to hand off work.
Texting is the killer feature
The standout is that Tasks works through your phone's regular messaging app on both iPhone and Android.
That sounds small. In practice it changes how often you use it:
- You can tell Siri or Google Assistant to "text Gemini" and dictate a chore, including hands-free while driving with CarPlay.
- Conversations stay in one long thread, like texting a human assistant.
- Results come back in the same thread, or as links into the Gemini app.
In one test, a user texted Tasks to find out how people were reacting to OpenAI's newest model, got a summary back, then texted "turn that into a presentation." Minutes later a finished deck appeared in Google Slides. No integrations to configure. It simply uses Google's own apps: Slides, Sheets, Calendar, Gmail.
It follows up on its own
The most striking behavior is proactivity.
The same tester asked Tasks to phone a local climbing gym about holiday hours. Calling wasn't enabled on the account yet, so it said no. About 45 minutes later, unprompted, it texted back. It had checked the gym's website, found no holiday calendar, and offered to draft an email to the gym asking whether it would be open on Thanksgiving. It asked permission before sending.
That's the line between a feature and an assistant: it remembered the goal and found another way.
Email it like a coworker
Email is the second channel, and it's clever.
- Forward any email to gemini@google.com with an instruction, for example "find a time on my calendar and email me back with options" or "they need that report; find it and send it."
- Because it comes from your address, Gemini knows it's you.
- Or CC or BCC gemini@google.com on a reply: "Sounds great. I'll have my assistant put the report together and send it over." The agent reads the context and does the work.
After you send, it replies to confirm. In one example: "I'm working on this for you. I'll track the updates." Completed tasks show up in the Gemini app.
Phone calls and payments
Like Muse, Tasks is getting phone calls: booking appointments by voice and then adding the date to your Google Calendar. Many early users still see "phone calls aren't enabled on your account yet," so treat this as rolling out.
Payments through Google Pay let the agent buy things for you, with an approval step for each purchase. It's the same direction OpenAI took this month with ChatGPT Wallet, which we explained in our week-in-AI roundup.
The problem: it's called "Task"
Here's where Google made a choice reviewers keep questioning.
Inside the Gemini app, the agent isn't a character. It's a mode. There's no name you choose, no personality, no avatar and no dedicated home. Your tasks sit intermingled with ordinary chats, and each one starts fresh.
Compare that with Muse, where you name your assistant, give it a face and talk to it in one long running conversation. That framing matters. People naturally say "I'll have Muse do it." Nobody is going to say "I'll have Task do it."
Ironically, the old name, Spark, at least sounded like something. Merging it into Gemini as a toggle makes it feel like a feature rather than a helper, and it may cost Google the emotional pull competitors are building.
Google's bet may be deliberate. A mode inside an app used by hundreds of millions of people needs no new habit. But it does mean the best way to use Tasks isn't the default one.
The fix: use it from your messages app
Reviewers noticed that Tasks feels far more like a real assistant when you use it by text. You get one long continuous thread, and it appears to be more proactive there than inside the app.
So the recommendation is simple: set up texting, save the number as a contact, and treat that thread as your assistant.
How to set it up (about five minutes)
Setup currently has to happen in the Gemini mobile app, not on a computer.
- Open the Gemini app, tap the side menu, then the gear icon for settings.
- Scroll to the new Channels option.
- Tap Text with Gemini, enter your phone number and the verification code.
- Tap Send text to start the conversation in your messaging app.
- Check your spam folder. Early users found the first message landed there. Mark it as known or save the number as a contact.
- In the same Channels screen, turn on Email with Gemini so you can use gemini@google.com.
- Back in settings, open Personal intelligence and turn on memory if you want it to learn your preferences over time.
- Open Connected apps and link what you use. Current options include Instacart, OpenTable, Peloton, SeatGeek and Shopify, with more being added.
More things it can do
- Images in your texts. Ask for an image and it arrives in the thread. Video isn't supported yet.
- Living dashboards. One user asked for a dashboard tracking AirPods prices, updated every morning. The link opened in the Gemini app.
- It delegates. That dashboard revealed something important: Tasks acts as an orchestrator. Simple jobs it does itself; bigger ones it spins off into separate tasks.
- Cross-channel memory. Start a task in the app ("write me a scary story"), then text "send me the first paragraph of that story" and it finds it. This works for items started as Tasks, not ordinary chats.
Where it stumbles
The most common complaint is silence. Users send a longer job, Tasks reacts to acknowledge it, then never reports back. Ask for an update and it says it finished long ago.
It seems to happen most with tasks that take a while. Until Google fixes it, check in on anything time-sensitive.
Other limits:
- Fewer third-party integrations than Muse.
- Calling isn't live for everyone.
- US-only channels. Texting and email don't work in the other five countries yet.
- Mobile-only setup.
A realistic day with Tasks
Here's how a busy freelancer might actually use it, start to finish.
7:30 AM. On the walk to the train, she texts: "Triage my inbox. Draft replies to clients, flag invoices, summarize newsletters in three bullets." By the time she sits down, drafts are waiting in Gmail for her to review and send.
10:00 AM. A client emails asking for a call next week. She replies, CCs gemini@google.com and writes: "My assistant will send some times." The agent checks her calendar and emails options.
1:00 PM. She texts: "Research what small businesses are saying about the new Gemini 4 Neon pricing and make five slides." A Slides deck lands in her Drive.
4:00 PM. "Plan four weeknight dinners under 30 minutes and add the ingredients to Instacart." She reviews the cart and approves.
9:00 PM. "Track prices on noise-cancelling headphones and send me a dashboard every morning."
Not one of those jobs is impossible without an agent. Together they could save an hour a day, and that's the real pitch.
Who should try it, and who should wait
Try it now if you're in the US, live in Google's apps and want help with email, scheduling, shopping and quick documents. The texting channel alone is worth setting up.
Try it cautiously if you're in Canada, Australia, India, Japan or Korea. You get the agent free, but not texting, email or payments yet, so it's closer to a powerful mode inside the app.
Wait if you depend on tools Google doesn't connect to yet, or if you need rock-solid follow-through on time-sensitive jobs. The silent-completion bug is real.
Gemini Tasks vs Meta Muse
| Gemini Tasks | Meta Muse | |
|---|---|---|
| Price | Free in six countries | Free |
| Feel | A mode inside Gemini | A named persona you talk to |
| Channels | App, SMS, email (US) | App, messaging, voice |
| Ecosystem | Gmail, Calendar, Slides, Sheets, Google Pay | Meta apps plus many third-party tools |
| Integrations | Growing, still limited | Broader today |
| Underlying model | Gemini 4 Hydrogen | Meta's model |
Reviewers' verdict: not a Muse killer yet, but a strong start. Some testers judge Google's underlying model the more capable one. Muse wins on integrations and on feeling like a personal assistant. If you live in Google's apps, or trust Google more than Meta with your data, Tasks is the easy pick. For Meta's side, see Meta Muse Spaces.
Smart ways to use it this week
- Inbox triage every morning. Text: "Triage my inbox, draft replies to anything urgent, summarize the rest."
- Meeting scheduling by CC. Copy gemini@google.com and let it find slots.
- Weekly meal plan plus grocery cart via Instacart, approve before checkout.
- Price watches with a daily dashboard for anything you're thinking about buying.
- Instant decks. Research a topic by text, then say "make this a Slides presentation."
Risks and common sense
- Permissions add up. Email, calendar, payments and password-manager sign-ins are a lot of access. Connect only what you need.
- Approve purchases carefully. Approval steps only protect you if you read them.
- Don't assume silence means failure, or success. Ask for status.
- Mind what you forward. Sensitive emails sent to the agent become part of its context.
FAQ
Is Gemini Tasks free? Yes, for users in the US, Canada, Australia, India, Japan and Korea. Elsewhere it requires Pro or Ultra, and the UK and EEA are excluded for now.
How do I text Gemini? In the Gemini mobile app go to Settings, then Channels, then Text with Gemini, verify your number and tap Send text. Check spam for the first message. This is US-only.
What's the Gemini email address for tasks? gemini@google.com. Forward emails to it or CC it, after enabling Email with Gemini (US only).
Can it make phone calls? Google says it can call businesses for US users, but many accounts don't have calling enabled yet.
What model powers it? Gemini 4 Hydrogen, according to The Verge.
The bottom line
Google just put a capable, proactive personal agent in front of millions of people for free, and made it reachable by plain text and email. It can build Slides decks, track prices, schedule meetings, shop with approval and chase down answers on its own.
It's held back by spotty follow-up, thinner integrations than Muse and a name that turns an assistant into a toggle. Set it up by text, give it real chores and check in on long jobs. For a free agent, that's a remarkably good deal.
Sources
- The Verge: Google's Spark AI agent becomes "Tasks," and now it's free
- Android Police: Google is giving millions of Gemini users a free personal AI agent
- TechRepublic: Gemini Tasks now free in 6 countries
- Android Central: Google just made its best Gemini agent free
- Google: Gemini productivity overview





