Gemini’s new AI agent is as good as Google’s demo

Google’s new 24/7 AI agent, Gemini Spark, can be shockingly good at doing things for you. But I’m not sure it’s worth the financial cost and potential privacy trade-offs.

The company gave me access to Spark last week. Google advertises Spark as an AI agent that can take over tasks and work on them in the background — even tasks with multiple steps — allowing you to put down your phone or step away from your computer. As you declare At the top From Spark website That it is “always under your direction,” that you “choose to run it,” and that it is “designed to check you in before you take major actions.” Given the growing skepticism towards AI, “my ‘I’m not involved in rogue AI’ t-shirt makes people ask questions that have already been answered by my t-shirt.”

I didn’t know where to start, so I took a page from it My colleague Antonio’s book: I decided to use Spark to handle tasks like what Google demonstrated on stage at I/O. Will it be as good in my home office as it was on the big stage?

At I/O, Google Vice President Josh Woodward showed off a few different examples. The first was to ask Spark to draft an email to a team at Google, compile everything about the Gemini Live launches and “wins from the past week,” and use a special AI skill to make the email look like it. Google asks Google to do things for Google He should It’s the easiest elevator in the world, so I tried to push it further.

I asked Gemini to draft an email to my wife that collected our average monthly grocery spending in 2026. I thought this test would tell me a few things: could Spark know who my wife was (without me giving Spark her name), could it locate our budget spreadsheet in Drive (which doesn’t have “budget” in the file name), and could it actually draft an email in Gmail?

When I got the result from Spark shortly after, I really said, “Wow, this is really crazy.” Spark found my wife’s email address, extracted the correct information from our 2026 budget spreadsheet, got the monthly grocery totals including incomplete data from May (which wasn’t finished yet when I ran the test), averaged the totals, and put it all in a draft email in my Gmail. The text of the email is addressed to my wife by her first name, even though her email address does not contain her first name. It even included a sign-out that we used only for each other.

In the following example, Woodward asked for some help planning a block party. I I’m not planning a block party, but I asked Spark for help using the same questions he asked. Things didn’t go well. I created a schedule for friends and family as a “very realistic reference of who brings what,” drafted an email in my Gmail that referenced a non-existent shared sign-up sheet, and created an ugly deck with slides detailing information about city permits.

To push Spark, I asked her to create the missing registration sheet and add a link to the already drafted email. While it took Spark a few minutes to figure this out, it worked; I created a spreadsheet, went back to the draft of the email body, and dropped the link.

Woodward’s final demo was arguably the most impressive. He talked into Spark to ask her to do a bunch of things: mark his meetings with CEO Sundar Pichai bright pink on his calendar, write a note to a new neighbor inviting him to his block party, and create a document to help with assignments for his kids at the end of the school year. For my own version, I had him create a calendar event every month before my wife’s birthday and make it hot pink, and draft an email to my family about sending them the first episode of the final season of the series. Task managerand create a document that includes the most important things my wife and I need to know about preparing our toddler for preschool.

I started this order at 3:35pm PST on Friday. During I/O, Woodward made a little show about putting down his phone and promising to check the results later in the keynote, which he did. But after addressing one issue—Spark wanted to access my contacts, which I denied—my task was accomplished after about four minutes.

Once again, I was amazed by the results, although they were not perfect:

Spark can be a powerful tool. But there are some caveats I should mention. Like all AI tools, you still have to verify its output for accuracy, which can have greater risks when it pulls from personal information to set up things you share with people you know. Although Google promotes Spark as something that can work on its own, I found myself constantly watching it or checking the notifications it sends to my phone. What good is an assistant if you have to micromanage his every move instead of trusting him? And why do I feel uncertain about draining power from a resource-hungry data center for relatively unimportant tasks?

Currently, Spark is only available to subscribers to the Google AI Ultra plan, which starts at $99.99 per month, and to users in the US only and in English only. Google gave me free access to test Spark, and I don’t think that’s good enough to be the only reason these plans are so expensive. Especially when I can do all the tasks I asked Spark to do on my own, it will take more time.

Spark also works best if you’re already deep into the Google ecosystem and have personal intelligence running. I’ve had a Google account for nearly two decades, so Spark has a lot of data it can use to provide its answers. But while Google He promises Gemini “Doesn’t Train Live” Into Your Gmail Inbox With Personal Intelligence turned on, you still have to put your faith in Google that it will serve as A good proxy for your data. Right now, I’m not sure if this is worth the cost or risk.