Prompting Is Dead
For three years, the AI interface was a box—the familiar prompt box. You typed, you waited, you read. The prompt box was the grand gesture of the LLMs era and, at the same time, the most awkward one. It asked you to articulate what you want before you even understand what it can give you.
At this year’s I/O, Google did something that looks like a coordinated attempt to bury it. Twelve features point in the same direction: an interface that watches you, predicts, and acts before you manage to type the first word.
Magic Pointer: the cursor learns to think
The most eloquent announcement was a cursor. Magic Pointer turns the mouse into a context-aware collaborator: you point to “this” or “that” on the screen, and Gemini understands what you mean without needing a full prompt. Pointing becomes language.
Google is starting with Chrome, but the logic is bigger. If AI sees what you see, typing becomes an unnecessary step between intent and execution.
Gemini Spark: the AI that works while you sleep
Gemini Spark doesn’t wait for instructions. It runs 24/7 on Google Cloud virtual machines, keeps browser sessions alive, touches connected apps and location data, completes multi-step tasks without asking again. In other words: an AI that keeps working when the laptop closes, the battery dies, or you simply disconnect.
The “assistant” paradigm ends here. This isn’t a helper. It’s a digital worker with its own shift.
For now, it’s available to trusted testers and is expected to roll out to Google AI Ultra subscribers in the US.
Search: from answers to actions
The biggest change—and the least noisy one—happened where Google was built: Search. AI Mode, with Gemini 3.5 Flash as the default, no longer returns links. You can upload files and media, continue from AI Overview into an agentic conversation, and assign Search agents to monitor the web, make bookings, and even build custom tools.
Google isn’t upgrading the search engine. It’s replacing it with something that executes. The search box—that other great empty box of our era—enters the same process of devaluation as the prompt.
Googlebook: the laptop reboots around Gemini
If the prompt is dying, the operating system itself has to be rewritten. Google introduced the Googlebook, a new category of AI-native laptop that unifies Android and ChromeOS into an operating system with Gemini as the central layer. Custom AI-built widgets, Magic Pointer built in, hardware partners ready for a fall launch.
It’s not a laptop with AI features. It’s an attempt to redefine what “personal computer” means when the interface is no longer mouse and keyboard, but intent and context.
Intelligent Eyewear: the screen leaves our hands
And around then, Google put Gemini into glasses. Intelligent Eyewear, in partnership with Warby Parker and Gentle Monster, ships in the fall as voice-first frames, with the display-equipped Project Aura to follow. The choice of partners is a statement in itself: not a geek prototype like Google Glass 2014, but glasses people would wear anyway.
Meta has already opened the field with Ray-Ban and more recently Oakley HSTN and has proven the market can support AI glasses if they’re signed by a brand people already trust. Google is entering the same game with the exact same logic.
The pattern behind the launches
If we look at all the announcements together, they share the same mechanical logic: less dictation from the user, more context from the system. In the current model, you had to explain what you want. In the model Google is building, the system knows what you’re looking at, where you are, which app you opened two minutes ago, and what’s on your calendar in an hour.
The prompt was the tax we paid because AI didn’t have enough context. Google is betting it can eliminate that tax.
And here is the real strategic story. Whoever controls the context layer controls the next interface. OpenAI has models. Anthropic has the enterprise. Google has something no one else has: your browser, your email, your calendar, your phone’s software, your search history, your geographic location, your documents in Drive.
It doesn’t need a better model to win. It only needs to turn all that context into an active layer.
The word that wasn’t said: surveillance
Of course, there’s also the word Google carefully avoided in the keynote: surveillance. A cursor that “understands” what you’re pointing at means someone is recording what you point at. A Spark that runs 24/7 in the cloud means actions of yours are being carried out by infrastructure you don’t control.
Glasses that see what you see mean a camera in other people’s faces. Gallup recently showed that 70% of Americans don’t want data centers in their neighborhood. That same suspicion will reach the interface as soon as people realize context-aware is a polite term for always on.
From explicit command to implicit consent
The prompt box had something humble about it: it forced you to choose what you would ask. Pointing, staring, ambient computing don’t offer the same luxury. Consent becomes implicit, baked into use.
At some point—probably soon—we’ll learn whether we prefer typing into an empty box or letting a system fill things in for us. Google is betting on the latter.
