I was half-asleep last Tuesday night when someone in my WhatsApp group dropped a link. “Google IO 2026 bhai, watch the keynote.” I ignored it. Next morning I finally watched it and honestly I had to rewind one part three times because I thought I misheard. Google called Gemini Spark a “24/7 personal AI agent.” Not an assistant. Not a chatbot. An agent. Something that stays running in the cloud even when your phone is locked, doing work for you while you sleep.
That is a very different thing.
I’ve been using Gemini on and off since the 2.5 Pro days, and I think most people — including me — were still treating it like a smarter Google search. Ask a question, get an answer, close the tab. Spark is trying to be something different. Whether it actually works that way in practice, I don’t fully know yet. But the announcement alone tells you a lot about where Google thinks this whole AI thing is going.
What Gemini Spark Actually Does
So basically Spark is an AI that runs in the background on Google Cloud. You don’t need your laptop open. You don’t need the app running. It just… goes. The demo they showed at I/O was kind of wild — someone asked Spark to help plan a block party and it went and created an RSVP tracker in Google Sheets, populated it with details from Gmail, and sent follow-up reminders to people who hadn’t responded yet. By itself. While the person was presumably doing something else.
That’s the pitch. Connect your Gmail, Docs, Calendar, and let Spark keep things moving without you having to babysit it.
Google did say it asks for your approval before it does something with real consequences — like sending an email or making a purchase. Which is good. Because the last thing I want is an AI that books me tickets to something because it decided I “probably” wanted to go. I’ve seen enough autocomplete disasters to be cautious about that.
For now, Spark connects to Gmail and Google Workspace apps, with third-party support — Canva, OpenTable, Instacart — coming through MCP later this summer. No date confirmed for the MCP rollout, which is a bit annoying. “Later this summer” could mean July. Could mean September. Nobody knows.
The Price Problem
Here’s the thing nobody is saying loudly enough. Gemini Spark is locked behind the new Google AI Ultra plan, which starts at $100 per month.
Let me just sit with that for a second. A hundred dollars. Per month. In India that’s roughly ₹8,300 just to use one AI feature that Google announced two days ago.
The old top plan was $250 a month, and they’ve now dropped that to $200 with a new $100 entry point. Google is framing this as a good deal — you’re getting more for less than half the original price — which is technically true but also the kind of math that sounds better in a press release than in your bank account.
Trusted testers got access to Spark starting this week — the week of May 19. US Google AI Ultra subscribers get it in beta starting next week, around May 26. Everyone else, including Google Workspace business users, gets it “soon.” International users: no date yet. So if you’re reading this from Hyderabad or Bengaluru, you’re probably just watching from the sidelines for now.
I don’t know how to feel about that. The technology sounds genuinely interesting. The pricing feels like it was designed for a very specific kind of person — a developer or startup founder with a corporate card who needs this to run actual workflows. Not really for me and most people I know.
The Search Bar Is Changing Too
This part I think got less attention than it deserved. Google also announced that Search itself is getting overhauled, and it’s the biggest update to the search box in close to 30 years. Gemini 3.5 Flash is now powering autocomplete — so instead of just predicting your next word, it’s trying to help you build your actual question as you type. There’s a new chatbot-style box underneath your results for follow-up questions, and the whole thing is meant to feel less like a search engine and more like a conversation.
AI Mode has already crossed one billion monthly users globally, according to Google. That number is big enough that they’re not exactly being cautious about this transition.
The old search bar isn’t going away tomorrow. But the direction is clear.
Gemini 3.5 Flash — The Model Under the Hood
Spark runs on Gemini 3.5, and there’s a separate model specifically called Gemini 3.5 Flash that Google launched at I/O. Flash is apparently 4x faster than other frontier models in terms of output tokens per second. It surpasses 3.1 Pro in coding and multimodal benchmarks, Google says, while being cheaper to run.
I haven’t tested it myself yet so I’m going off what Google and a few early reviewers said. The speed claim is the one I’m most curious about. Gemini has sometimes felt a bit slow compared to other models when doing longer tasks. If Flash actually fixes that, it would make a real difference for agentic tasks where you’re waiting for something to complete in the background.
3.5 Pro is still in testing and supposedly coming next month. So there’s a faster, cheaper model available now and a more capable one still not ready. Feels a bit like they pushed the keynote date and the models had to keep up, but I’m probably oversimplifying.
A Few Things That Were Quietly Interesting
Google also announced something called the Daily Brief — it goes through your Gmail, Calendar, and Tasks and gives you a morning digest of what’s coming and what you should probably do first. This one is actually rolling out to Plus, Pro, and Ultra users today, not just Ultra. So more people can use it. I think this might end up being more useful day-to-day than Spark for most people, because it doesn’t require you to change how you work — it just sits there and summarizes things.
Also: the Gemini app is moving away from daily prompt limits. Instead of “you used up your 50 messages,” it’s now tracking compute. More complex questions use more compute budget. Simpler ones use less. This sounds cleaner in theory but I have a feeling people are going to be confused about why a long conversation suddenly stopped working mid-thread. The docs don’t make this super clear yet.
And the app got a full redesign — something Google calls Neural Expressive, which has new colors, fluid animations, a pill-shaped prompt box. Looks nicer from what I saw. I spent about 45 minutes with the old Gemini interface just last week, and the navigation was honestly a bit clunky. So good.
My Honest Take
Gemini Spark is an interesting idea that I can’t try yet because it’s US-only and costs more than what most people in India spend on streaming subscriptions combined. That’s not a knock on the technology — the demo looked solid — but it does mean this is going to feel very theoretical for a large chunk of the world for a while.
The thing that stuck with me most from the I/O keynote wasn’t Spark itself. It was Sundar Pichai opening by saying it’s been ten years since Google committed to making AI the center of its product strategy. Ten years. And they’re still in the middle of figuring out what that actually means for everyday users. Spark might be a real step toward something. Or it might be one of those things that sounds big in May and quietly fades by October. I’ve seen both happen too many times to be sure.
What I do know is that the search bar — the thing that has basically defined how billions of people use the internet — is changing. Maybe not this month. Maybe not this year. But Google is done pretending the box with the blinking cursor is enough.
Whether Spark is the replacement, or just a step toward one, I guess we’ll find out.