Google announced Gemini Spark today at I/O 24.
It’s loud. It’s ambitious. It might just be the heaviest thing the company has dragged onto stage in recent memory.
Called your personal AI agent. It runs continuously in the cloud, no Mac Mini required. No dedicated server rack gathering dust in the corner of your house. Just code spinning up tasks while you do whatever you do when you’re ignoring your email.
Powered by Gemini 3.5 and baked into Google’s Antigravity IDE.
It hooks into the whole Google ecosystem, then reaches out. More than thirty third-party apps connect via MCP (Model Context Protocol). Adobe, Asana, Dropbox. Lyft. OpenTable. Uber, Zillow, Zocdoc. The list keeps growing.
Need to update your boss? Spark digs through Inbox and Docs, pulls the threads together.
Running a side hustle? Tell it to watch customer messages, flag the angry ones.
“We think of it as giving a teenager their first debit card.”
— Josh Woodward, VP of Google Labs
It’s rolling into Gmail and Chat. Soon. AI Ultra subscribers in the U.S get first dibs next week.
But there is the question. The big one. The elephant in the room.
What stops it from buying things? Sending emails you didn’t write? Deleting files?
Google’s answer: Agent Payments Protocol (AP2).
It’s a framework. Hard limits on spending. Strict lists of approved merchants. User approval for every transaction, for now. A permanent digital paper trail. Returns. Disputes.
It sounds safe.
It probably feels safer.
But the goal is autonomy. Gradual, slow-motion freedom. Spark starts with constraints. It earns its stripes.
Google Shopping integration is coming later this year. The leash gets shorter. Or longer, depending on who you ask.
Is this helpful?
Maybe.
It’s certainly a lot of access, all at once, wrapped in a nice little agent-shaped bow.
We’ll see what it buys first.
