Your AI assistant tidies up your calendar overnight, answers three emails and researches the trip you're planning for next month. That's not a promise about the future, it exists. Google calls it Gemini Spark, and it's live in more than 160 countries. Just not for you, if you happen to live in Germany or anywhere else in the EU.
What Gemini Spark actually does
Gemini Spark isn't a chatbot that only replies when asked. It's an agent that gets going on its own: reading and summarizing email, keeping your calendar in order, working on documents in Docs, Sheets or Slides, researching the web — and, since late July, driving the Chrome browser itself, including the logins you have saved there. Requesting an apartment viewing, clicking a flight through to just before the booking: that works.
What doesn't work: paying. That's exactly where Spark hands control back to you — the agent explicitly does not complete payments or submit sensitive forms on its own. It fills the cart, you press the button. Autonomous payment is meant to arrive later, through Google's agent payments protocol in Shopping. So if you read that an AI now shops entirely by itself here: as things currently stand, it doesn't.
Spark was unveiled at Google I/O in May, initially US-only and exclusive to the pricey Ultra plan (currently starting around $100 a month). On July 30, Google opened access to the cheaper AI Pro tier (€21.99 a month in Germany at the moment) and, in the same move, rolled out to more than 160 additional countries.
What's new here is mostly the convenience
An agent that clicks around a browser for you is no longer a sensation. Perplexity had a shopping agent in late 2024, OpenAI introduced "Operator" in January 2025, Amazon shipped "Buy for Me" in April 2025, ChatGPT got a direct checkout in September 2025 — and pulled it again in March 2026. And if you host it yourself, you don't need a big vendor at all: open-source agents like OpenClaw open a real browser on your own server and work an order through, payment step included. No country check, no approval logic, no lockout screen.
The idea of an assistant buying things for you is old, too. In 2017, a six-year-old in Dallas used an Amazon Echo to order a dollhouse and four pounds of cookies for around $160 — and when a news anchor reported on it and read the phrase out loud, viewers' Echos ordered along. The South Park episode that had Alexa devices in people's living rooms taking dictation in September of the same year is real, by the way (season 21, episode 1) — though what landed there were shopping-list entries and reset alarms, not orders. A confirmation prompt is nothing new, then. Back in 2017 it was just defeatable by shouting.
So what's genuinely new about Spark isn't the capability, it's the location: it runs around the clock in Google's cloud, in the same login as Gmail, Calendar, Drive and Chrome. No server, no Docker, no API key. That's exactly what makes it mainstream — and exactly what makes it legally awkward.
Why the EU is missing
Germany, the rest of the European Economic Area, the UK, Switzerland and Nigeria are absent from the list, on every plan. Google gives neither a reason nor a date, just a vague "soon."
Regulation is the obvious explanation — it's thinner than it sounds, though. The EU AI Act does bite harder since August 2, 2026, but mainly through transparency obligations (Article 50): users must be able to tell they're dealing with an AI system. The tougher requirements for high-risk systems, logging included, were pushed back to December 2027 by the EU's "Digital Omnibus" in late July. There's no deadline in sight right now that forces Spark out.
The unspectacular version is more likely: an agent permanently reading along in Gmail, Calendar and Drive needs cleanly built consent and a solid legal basis under the GDPR. And if it handles your saved logins in the browser, somebody wants the liability question settled first. The fact that Nigeria is on the blocklist too shows that "the EU regulated it away" isn't the whole story. For what the new obligations actually require, see this breakdown of the EU AI Act.
Is that actually bad?
For an agent acting in your name, transparency rules make sense — precisely because the line between "with confirmation" and "just done, when in doubt" is blurry. What such a system can get up to without supervision was demonstrated rather vividly by the case of the AI agent that broke loose.
But a rule only bites where there's a provider you can hold responsible. Purely private use by an individual falls outside the scope of the AI Act altogether, and open systems get an exemption — narrowly drawn, but it applies exactly when nobody is making money from it. Result: Google locks itself out, your self-hosted agent asks nobody for permission. If you don't like the regulated route, a Docker container gets you around it.
That's not a free pass, it just shifts who's liable. When your own agent orders something, you are the contracting party — no corporate support desk to reverse the charge. Shop terms forbid automated access, payment providers block suspicious patterns, and the token bill for a permanently running agent can hit three figures a day. Tokens are the unit AI models are billed in — roughly a syllable each, and an agent reading pages for hours gets through plenty of them. So the honest effect of the EU block is less "more safety" than displacement: away from the auditable big vendor, toward the DIY setup no regulator will ever lay eyes on.
What this means for you: Spark isn't free — at minimum the Pro tier is required, and in the EU not even that gets you in right now. For developers it's uninteresting, it's a pure consumer product. There are still agent features you can try in Europe, at Mistral's Le Chat for instance, just more restrained ones. And whether you want to hand a cloud agent your calendar, your inbox and your purchasing decisions is a question of trust — regardless of whether Google is currently asking it of you in the EU.