Your AI Agent Now Works While Your Laptop Is Off. Which Means It Was Never on Your Laptop.

At I/O on May 19, Google introduced Gemini Spark — a "24/7 personal AI agent" — and buried in the pitch was a line that's easy to skim past and shouldn't be: it works in the background "even while your phone or laptop are turned off." It runs recurring tasks, learns new skills you teach it, and reaches across Workspace, ServiceNow, Salesforce, SharePoint, and the open web through Gemini Enterprise connectors. It asks permission for the high-stakes stuff — sending email, anything risky — and just does the rest.

The detail that matters isn't the autonomy. We've been talking about autonomous agents for a year. It's the location. An assistant that keeps working while your devices are off was never running on your devices. It's a process in Google's cloud, holding your credentials, acting as you, around the clock, whether or not a single screen you own is lit.

Two Shifts at Once: Device to Cloud, On-Demand to Always-On

For decades "my computer" meant the thing in front of me, and "my software" did things when I told it to and stopped when I closed the lid. Spark quietly retires both halves of that. The agent isn't on the laptop — closing the laptop changes nothing. And it isn't waiting for instructions — it's running a standing list of recurring tasks, on its own schedule, while you're asleep or in a meeting or on a plane with the device powered down.

That's two shifts at once, and the "off" detail exposes both. The agent moved from your device to the cloud, and it moved from on-demand to always-on. Put those together and what you've actually got is a persistent, credentialed identity that acts in your name 24/7 with broad reach into your work systems — and the only thing standing between "helpful" and "incident" is a permission prompt for the handful of actions Google decided were high-stakes, plus the connectors you wired up and probably forgot the scope of.

I'm not waving this away as dystopian. The capability is genuinely useful, and "ask before sending email, act freely otherwise" is a reasonable default. But notice what just became normal for hundreds of millions of non-engineers: an autonomous process with your access, running while you're not present, deciding for itself which tasks to advance. We spent the last year arguing about that exact thing in the context of engineering teams and production systems. Spark ships it to everyone with a Workspace login and a phone, as a consumer feature, framed as convenience.

If You Run a Platform Org, This Lands on Your Desk Anyway

The consumerization of always-on agents lands on your desk whether you greenlit it or not.

Every Spark is a non-human identity with standing access. It's not a session that ends when the user logs off — it's a persistent actor with credentials and connector scopes into Workspace, ServiceNow, Salesforce, SharePoint. That belongs in the same governance bucket as every other agent and service account you're trying to track: who owns it, what's it scoped to, what's the blast radius, how do you revoke it. The difference is this one got provisioned by an end user clicking "set up Spark," not by a platform team filing a request. The identity sprawl you've been fighting just got a self-serve front door.

"Works while your devices are off" breaks a monitoring assumption you probably didn't know you were making. A lot of security intuition quietly relies on presence — the user is at the machine, the session is active, someone would notice. An agent acting at 3am with no human and no lit screen has no presence signal at all. The detection story has to be behavioral — what is this identity doing, is it normal — because "is the user here" is no longer a meaningful question.

And the connector scopes are the whole ballgame. A 24/7 agent reaching into four enterprise systems is exactly as dangerous as the broadest scope you granted it and exactly as observable as the logging on the other side of each connector. Treat connector authorization as the real security boundary, because for an always-on agent, it is.

The Frontier Shipped Wearing the Costume of an Assistant

Here's what I think actually happened on that stage, and it's bigger than a feature. The frontier of "agents act on their own, persistently, with access to real systems" has been a thing enterprises debate carefully — governance reviews, identity strategies, the works. Google just took that frontier and shipped it as a consumer convenience, and the genius (and the risk) is that it doesn't feel like the frontier. It feels like a better assistant. The "works while your laptop is off" line is delivered as a perk. It's actually the most consequential sentence in the announcement, because it's the moment the agent stopped being a tool you operate and became a worker you employ — one that clocks in whether or not you do.

The pattern I keep watching: every genuinely new capability arrives wearing the costume of the old familiar thing it's replacing, so we underreact. Spark wears the costume of "assistant" — Siri, but useful. Underneath, it's a persistent autonomous identity with cross-system reach, and we're going to onboard hundreds of millions of them in the next year while calling it a productivity upgrade. The governance conversation that took enterprises eighteen painful months to start having about their agents is about to be necessary for everyone's agents, except nobody's going to frame the consumer version as needing governance at all.

So the question I'd put to anyone — IT leader or just a person setting this up — is the one the convenience framing is engineered to make you skip: what is this thing allowed to do while I'm not watching, and who finds out if it does something I didn't intend? An agent that works while your laptop is off is a wonderful thing right up until the day it does exactly what it was told, by someone who wasn't you, and the first you hear of it is the cleanup.

Sources: Google introduces Gemini Spark, a 24/7 agentic assistant with Gmail integration, at I/O 2026 | TechCrunch · Gemini Spark – Your 24/7 personal AI agent for productivity | Google · Inside Google I/O 2026: Gemini Spark and the Rise of Autonomous AI Agents | CTO Magazine · Innovations from Google I/O 26 on Google Cloud | Google Cloud Blog