Monday May 12, 2026, 7 PM Paris time. Sameer Samat opens the show with a sentence that changes everything: "We're transitioning from an operating system to an intelligence system." In one hour of keynote, Google does not present Android 17 - it rebrands Android outright. Gemini is no longer an app or an assistant; it is now the brain of the phone, the browser, the car, and soon the laptop. All of this, two weeks before Apple's WWDC is set to unveil its own AI cards. This is no longer a product update; it is a narrative race.
Google is no longer talking about an OS - it talks about an intelligence system
The phrase was repeated three times during the show. Sameer Samat, President of the Android ecosystem, did not announce a new product on Monday - he announced a category change. Android is no longer an operating system; it is an intelligence system. The distinction is not marketing, it is strategic: an OS fades behind apps, an intelligence system steps in front.
“We're transitioning from an operating system to an intelligence system.
”
This pivot comes with timing that is anything but innocent. CNBC noted it directly: Google staged this show "just weeks before Apple is expected to show its own Gemini-powered Apple Intelligence reboot at WWDC." Translation: the narrative needed to be set before Cupertino arrived with its own. To understand the pressure on the French side given this Apple-Google axis, Arthur Mensch's hearing before parliament on the same day illuminates the other side of the month.
These numbers remind us of the base on which Google is rolling out its bet: Gemini is already everywhere, but in homeopathic doses. May 12 marks the real dose.
Gemini Intelligence: three features that change everyday routines
The strategic abstraction materializes in three concrete features, all planned for this summer on the latest Pixel and Galaxy devices.
Gboard Rambler: dictation that understands you better than you do
Gemini listens, keeps the meaning, discards the hesitations. 9to5Google summarized it well: "It goes beyond word-for-word dictation and lets you transcribe naturally." You talk to your keyboard like a human being, the AI filters out the "ums," the repetitions, the abandoned phrases. This is the first real use case where the LLM tackles the moment where smartphones betray users most: the gap between thinking speed and typing speed.
Create My Widget: from natural language to your homescreen
You say: "Suggest three high-protein meal prep recipes every week." Android generates an adaptive homescreen widget. Same for Wear OS Tiles. This is the first time a mobile OS has exposed UI generation to the general public without going through a graphical builder. The risk, which we will address later, is the Christmas-tree effect on homescreens.
Gemini in Chrome: delegated browsing
Launch announced for late June 2026. On Android Chrome, Gemini can now navigate a site and fill a form on your behalf. This is auto-browse. For the first time on mobile, your browser is no longer a tool you drive - it is an assistant that acts. The most direct entry point to mass-market AI agency.
Agency: what Gemini does for you
This is the most politically charged part of the show. Google demonstrates scenarios where Gemini executes multi-step tasks across several apps, without the user chaining manual operations.
Gemini can navigate multi-step workflows for you, such as booking a fitness class, shopping from notes, finding class materials, and planning a trip from a photo.
The demos are deliberately educational: you take a photo of a spinning class flyer, Gemini recognizes the event, opens the gym app, books the slot, and sets the reminder. You take a photo of your handwritten shopping list, Gemini orders the ingredients. On paper, this is the promise of a phone moving from a browsing screen to an execution agent.
“The human is always in the loop.
”
The guardrail is hammered home: before any transaction (payment, booking, send), explicit user confirmation is required. Whether that "always" holds in real usage remains to be seen. Under the hood, the mechanism is called AppFunctions API.
AppFunctions API: the technical contract of agency
This is the most understated announcement of the show and probably the most structurally significant over the next 18 months. The Developer's Cut, published simultaneously, lays out its contours.
“Gemini can automate tasks across selected apps on behalf of the user with built-in transparency and control.
”
Concretely, AppFunctions is the interface through which a third-party Android app exposes its actions to Gemini: "book a class," "order a product," "open a project." For now it is in private preview with 25 apps. If the API holds, agency will no longer be a Google demo - it will be a cross-app standard. If it does not hold, Gemini will remain stuck within the Google ecosystem.
Googlebook: the new category that smells like AI PC
This is the other big piece of the show. Google announces Googlebooks, a new category of laptops built around Gemini, in partnership with Acer, ASUS, Dell, HP, and Lenovo. Planned release in autumn 2026. No prices announced.
| Criterion | Chromebook (2011-2025) | Googlebook (2026) |
|---|---|---|
| Core product | Chrome (browser) | Gemini (AI) |
| OS | ChromeOS | Android |
| Phone connection | Weak | Quick Access, Cast, direct sync |
| Target | Education, light B2B | AI-native general public |
| OEM partners | Acer, Asus, HP, Lenovo, Samsung | Acer, ASUS, Dell, HP, Lenovo |
The critical angle comes quickly. Engadget, two paragraphs after listing the announcements, drops:
“[Magic Pointer is] like Clippy, but potentially 1,000 times more annoying.
”
Magic Pointer is the Gemini cursor that surfaces while you work to suggest actions. The Clippy comparison is not trivial: it invokes the entire baggage of intrusive AI assistants and, more importantly, the commercial failure of Microsoft's Copilot+ PCs in 2024, whose AI features the press ended up describing as "generative AI junk" before Microsoft partly pulled them. Engadget concludes:
Whether folks will be more accepting of Gemini Intelligence and Googlebooks is something to keep an eye on over the next couple of years.
Deployment: who gets the magic first
Here is the part that gets glossed over between two live slides but deserves to be laid out plainly. Gemini Intelligence will not be available on May 12. Deployment happens in waves, by device, by feature, by country.
- May 12, 2026The Android Show: I/O Edition
Official announcement, YouTube replay, Developer's Cut published in parallel.
- Late June 2026Gemini in Chrome (Android)
Auto-browse and form filling on mobile browser.
- Summer 2026Wave 1: Pixel + Samsung Galaxy
Gboard Rambler, Create My Widget, in-app agency. Reserved for "latest" models.
- Autumn 2026Googlebook launch
Acer, ASUS, Dell, HP, Lenovo. No price communicated.
- Late 2026Extension to Wear OS, Auto, XR
Watches, cars, Android XR glasses. The intelligence system unfolds.
The word that stings is "latest." Gemini Intelligence starts on the latest Pixel and latest Samsung Galaxy devices. For the 3 billion Android devices still in circulation, the promise remains to be seen. Apple, by contrast, will deploy Apple Intelligence on every iPhone 15 Pro and 16. Uniform coverage versus a de facto hierarchy: that is the major strategic difference between the two camps.
Security and privacy: Google heard the criticism (or did it?)
Dave Kleidermacher, VP of Platforms Security and Privacy at Google, got his own dedicated post in the May 12 announcement salvo. The message is clear: Gemini Intelligence rests on three technical mechanisms.
On-device compute enclave for sensitive AI processing. Inherited from recent Pixel devices.
Encrypted tunnel between the device and Google servers when cloud processing is needed.
Android hypervisor certified SESIP Level 5, hardware isolation.
“Users have granular authority to opt-in and out of entire features or disable specific components at any time.
”
The discourse is solid on the mechanics, opaque on the practice. No public source specifies which features run on-device and which require a cloud call. For cases like Rambler or Create My Widget, that is not a minor detail: it determines whether your voice and prompts leave your phone. CNBC noted this sharply:
“Gemini's power is its deep Workspace reach, and that's also its biggest risk. Because Gemini runs inside Gmail, Docs, Sheets, and Drive, it carries search-level access into everyday work.
”
That is where the real issue lies. If Gemini Intelligence delivers on its promises, it will have access to a data perimeter equivalent to Google Search - but this time on your daily professional data. Trust will not be a slogan; it will be a purchasing decision.
What we did not see on stage
Every keynote has a flip side. Here is what the public sources from May 12 do not say and what you should keep in mind.
- Googlebook prices
No pricing. No range. Just "autumn 2026" and five OEM partners. Given the Chromebook price dispersion (from 1,500), the mystery is total.
- On-device vs cloud breakdown
Google names the three security layers (Private Compute Core, Private AI Compute, pKVM) but does not publish a feature-by-feature matrix. Independent evaluation of the guarantees is impossible at this date.
- Real adoption of existing Gemini features
No figures on actual usage of Gemini app, Circle to Search, or other already-deployed Android AI features. The baseline comparison is missing.
- Performance benchmarks
No inference speed, no latency for agentic actions, no battery impact. All of this will come from summer press tests.
- Apple's response
Radio silence at the time of writing. WWDC 2026 will either lift or maintain the taboo in a few weeks.
Going further
The full show is online. To form a genuine opinion before listening to the recaps, it is worth hearing Samat say these phrases out loud.
The sources behind this breakdown:
Where this is heading
May 12, 2026 will be looked back on in two years as the moment Google stopped defending Android as an OS and started defending it as an agent layer. The promise is real, and so is the risk: if Apple Intelligence comes back clean at WWDC and if Engadget is right about the Clippy reflex, Google will have opened a category it must defend alone for two quarters. If the deployment holds and if AppFunctions API makes its way into third-party developers, then this is the first time since 2007 that a major mobile platform has changed its nature - not just its interface.
For builders and brands that rely on Android, the practical question is immediate: what does this change for your stack and your product? If you want to talk through it, open a conversation with Blokby.
Discuss your Android and Gemini stack with Blokby