Google I/O 2026's keynote ran for two hours. The first hour was about features. The second hour was about philosophy.
DeepMind CEO Demis Hassabis took the stage and said this might be the "foothills of the singularity." The developers in the audience applauded. The people reading the comments section — they didn't.
The Verge's Dominic Preston put it more bluntly: Google's I/O demo demands both your trust and your personal data. And the comments suggest that more and more people no longer want to make that trade.
The Cost of an "Always-On Agent"
Google's core narrative this year is "agentic Gemini" — transforming Gemini from a chatbot you actively query into a 24/7 agent that proactively manages your schedule, searches for information, and completes tasks. It can read your emails, check your calendar, use your camera, access your documents, and then "proactively" offer suggestions and get things done.
Technically, it's impressive. Google's ecosystem integration is genuinely the strongest in the industry — there's only one company that can simultaneously access Gmail, Maps, Photos, and Calendar at scale.
But "can do" and "should do" are two different questions.
Emma Roth's commentary captured the mood in the comments: a growing number of people don't want to make this trade anymore.
Think about what an always-on agent actually means. It needs continuous access to your data streams. Not for a specific query, but for "proactive" assistance. It needs to know what you're thinking and what you need — before you even ask.
That's not an assistant. That's surveillance with a friendly UI.
"Foothills of the Singularity" or a Convenience Trap?
Hassabis used "foothills of the singularity" to describe what Google I/O showcased. The metaphor itself is the problem.
"Singularity" implies an irreversible technological leap. But what Google showed today — agentic search, AI Studio vibe coding, Volvo's external cameras reading parking signs — none of these individually qualify as "singularity"-level breakthroughs. Combined, they form a very Google strategy: AI-ify every touchpoint and use the data flywheel to reinforce the moat.
This is what Google does best. Not the smartest AI, not the prettiest UI, but the deepest integration and the widest data collection surface.
Google did the same thing in the 2010s with search and advertising. Now it wants to do it again with AI.
Gemini 3.5 Flash and AI Studio Goes Android
I/O also announced Gemini 3.5 Flash and an Android version of AI Studio, letting users build other apps using AI and prompts on their phones.
The expansion of vibe coding from desktop to mobile is noteworthy. But the real signal is that Google is pushing the barrier for AI programming as low as possible — so low that you don't need to know code, or even sit at a computer.
Good for Google. More people using AI Studio means more data flowing back into Google's ecosystem.
For users, it depends on your perspective. If you're a startup founder or indie developer, this could be a good thing — more tools in the chain. If you're just a regular user, you might want to think about where your data goes when you're "just playing around" with AI Studio on your phone.
My Take
Google I/O 2026's technology roadmap is clear: use AI agents to turn every Google ecosystem entry point into an intelligent touchpoint.
It's not a bad strategy. From Google's position, it's almost the only correct one — OpenAI is moving up (Agents, Operator, ChatGPT Pro), Apple is moving inward (Apple Intelligence deeply integrated into iOS), and Google's choice is to AI-ify the entire ecosystem.
But I'm skeptical about the "always-on agent" narrative. Not because it can't be built, but because of user psychology. The comments already gave us the answer — more and more people are asking: "Is this trade worth it?"
Google's answer has always been "yes, because our services are free." But that answer is getting harder to accept in 2026. Users aren't naive. They know what "free" costs.
The next metric to watch isn't what features Google announced at I/O. It's how many people actually turn on those "proactive agent" permissions. There's usually a canyon between applause at a keynote and real user behavior.
Main sources: