(Credit: Jeffrey Hazelwood/PCMag; Google)
Gemini Spark is Google's latest attempt to turn AI agents into something genuinely useful. Billed as a 24/7 personal agent, it can browse the web, access your Google data, and tackle tasks in the background with minimal input. After spending time with Spark, I found it delivers on that vision more effectively than any AI agent I've tested so far. But while Spark is impressively capable, it still runs into the same reliability issues that plague competing agents—and many of its headline features overlap with capabilities already found in Gemini itself. Here's everything you need to know before deciding whether Spark is worth your attention.
What Exactly Is Gemini Spark?
Gemini Spark is an AI agent you can access via the Gemini app or site, but it currently requires an AI Ultra plan ($99 per month). However, I expect Spark to eventually become available to lower-tier plans, much like the Veo 3 AI video generation model did after its initial launch.
Spark is essentially a next-generation version of Project Mariner, Google’s first major attempt at an AI agent that it shut down earlier this year. Like Project Mariner, Spark can pilot a web browser to perform tasks for you, but it can also access all your Google data, complete recurring tasks, integrate with third-party services, work in the background, and more. Whereas Project Mariner was a proof-of-concept for a Gemini-powered AI agent, Spark is a fully realized product (though it's still in beta).
You interact with Spark by prompting it, just as you would with the standard Gemini chatbot. To get the most out of Spark, you need to allow it to browse the web and connect it to Google’s other services (such as Docs and Gmail), which it asks you to do. If that sounds like an invasion of privacy, it sure is. But if you use Gemini, you've already surrendered your data. Spark at least doesn't make things any worse.
Spark's Secret Weapon Is Google's Data Advantage
Gemini’s Personal Intelligence feature pulls relevant information from other Google services, such as an email in Gmail or your watch history on YouTube, to better respond to prompts. For example, if you ask Gemini what video you should watch next, it automatically leverages Personal Intelligence to analyze what you watch on YouTube. Then, it uses that information to inform its recommendations. Privacy concerns aside, Personal Intelligence generally delivers on its premise.
Personal Intelligence is the bedrock of Spark and gives it an advantage over other AI agents. When I asked Spark simply to “find me jobs that are a good fit for my background,” it automatically searched my email, found my resume, and reviewed previous applications to determine which positions would be relevant. Then, it searched the internet and presented me with a list. For comparison, when I ask ChatGPT Agent the same thing, I need to specify what I'm looking for, upload my resume, and provide any other relevant details.
As you might expect, Spark has significant potential, depending on how much information you store on Google’s servers. If whatever data Spark needs to complete a task exists somewhere within Google’s data ecosystem, Spark can likely find it and put it to good use.
Spark Is Surprisingly Competent and Fast
Beyond its Personal Intelligence capabilities, Spark performs well as a standard AI agent. For example, when I asked it to create a spreadsheet of the last 10 characters released in the video game Warframe, the components necessary to craft them, and where to acquire each part, Spark did so with ease, presenting me with a clean Google Sheets document. I’ve sent this same task to many different AI agents, and they almost always struggle with it.
Gemini’s blazing-fast 3.5 Flash model powers Spark, effectively addressing another one of my historic problems with AI agents: speed. Spark still doesn’t complete most tasks faster than you can yourself, but it works quickly enough that it feels reasonable to let it run in the background while you do other things. For instance, Spark took about five minutes to complete my job search.
The Same AI Agent Problems Haven't Disappeared
I like to task AI agents with finding a recipe and adding the corresponding ingredients to an Instacart shopping cart. With Spark, when it prompted me to sign in to my Instacart account, it said signing in was blocked on the page, so it couldn’t add my items. Spark also routinely had trouble browsing Amazon. However, Spark’s problem with Instacart is also evidence of its bugginess, since it should be able to access Instacart directly as a connected app.
Spark also can’t fix the underlying limitations of AI models. For example, while it was able to find my resume and locate job listings based on my bare-bones prompt, multiple listings Spark recommended were no longer live. You will run into this same problem if you ask ChatGPT or Gemini to do something similar, so Spark clearly isn't bringing massive intelligence upgrades. You should plan to do some prompt engineering and send follow-up messages to get the best results from Spark.
Spark's Biggest Problem Is Gemini Itself
What bothers me much more than the actual issues I ran into with Spark is how much of what it does is already possible with Gemini.
Above, I asked Spark to add items to a digital grocery cart, find jobs for me, and make a spreadsheet. However, I can just ask Gemini to add those same items to my Instacart virtual cart (which it does by leveraging its Instacart integration). Alternatively, I can simply attach my resume to a Gemini prompt and ask it to find me jobs. And for spreadsheets, I could use Gemini’s top-notch deep research feature to compile any information I need.
Even the capabilities Google highlights on its Spark overview page aren’t necessarily things that require it specifically. For example, Google says Spark can scan your inbox and give you a recap on a set schedule. However, Gemini’s AI Inbox in Gmail already does something similar on demand. Google demonstrates how to use Spark to create a style guide for responding to emails in your voice. Once again, this feature (called ‘help me write’) already exists in Gmail right now. Google also shows off using Spark to automatically track internships that pop up, but AI agents that let you “stay updated on whatever matters to you” are coming to Google Search soon.
Furthermore, Personal Intelligence was a Gemini feature first. Spark’s ability to automatically pull in relevant information about you from different corners of the Google universe is something that Gemini proper already does. In fact, even though I mentioned earlier that I could attach my resume to a Gemini prompt before asking it to find jobs for me, that's not strictly necessary. When I give Gemini the same prompt I would in Spark, it automatically finds my resume via Personal Intelligence and looks for relevant jobs.
Spark's interface (which is separate from the traditional Gemini one) reinforces its awkwardness. I would much prefer to just use the standard Gemini window and let it automatically defer tasks to Spark when necessary. With this change, Spark would feel far more seamless than redundant.
The Best AI Agent Yet—But Not an Essential One
I didn't encounter any show-stopping issues with Spark. In fact, I came away largely impressed by how well it performs relative to other AI agents I've tried. Nonetheless, you can already do a lot of what Spark does in Gemini without splurging for an AI Ultra subscription. The good news is that you probably won’t have to wait long for Spark to come to more affordable (or even free) Gemini tiers. I plan to revisit Spark once it reaches a full release, and I'm hoping it manages to better distinguish itself from Gemini in the interim.
About Our Expert
I’ve been writing about consumer technology and video games for over a decade at a variety of publications, including Destructoid, GamesRadar+, Lifewire, PCGamesN, Trusted Reviews, and What Hi-Fi?, among many others. At PCMag, I review AI and productivity software—everything from chatbots to to-do list apps. In my free time, I’m likely cooking something, playing a game, or tinkering with my computer.
- I Can't Code, But I've Made Five Game Mods With AI. Here's How
- AI Built a Photoshop Clone in 24 Hours. Is This Piracy, Plagiarism, or Progress?
- AI Has Ushered In the Era of Personalized Software, and Honestly? I’m Excited
- Airtable Review: Powerful, Flexible Work Management Powered by Deep AI Integration
- I Let 3 AI Agents Take Over My Life. Only 1 Didn't Fail
- More from Ruben Circelli
