Google used Samsung's Galaxy Unpacked event to roll out the first real capabilities of what they're calling Gemini Intelligence—their framework for proactive, agentic AI that runs across phones, watches, and glasses. The timing matters. We're past the "foundation model" phase and deep into the "ship it on hardware people actually buy" phase.
The announcement centered on three updates shipping with Samsung's Galaxy Z Fold8 Ultra, Fold8, and Flip8. Let's unpack what's actually interesting here.
Task automation expands from beta to 40+ apps
Back in February, Google launched a beta for Gemini task automation—basic delegation like ordering rideshares or takeout. The new version scales from "a handful of apps" to more than 40 popular apps, according to the announcement.
That's the kind of expansion that signals moving from proof-of-concept to production. The coverage now includes shopping, dinner reservations, travel booking, and event tickets. On the Galaxy Z Flip8, you can trigger this directly from the Flex Window with a long press of the power button.
What caught my attention: Gemini can now "use advanced reasoning to understand what is on your screen and parse complex images as prompts." Translation: multimodal input as a first-class interface. Show it a shopping list photo or an inspiration image, and it can reason about what you need.
The execution model is well thought out. You get step-by-step progress tracking, can multitask in other apps, or let Gemini run in the background. You can interrupt to adjust specific details, then hand control back. Gemini notifies you when it's ready for review and final confirmation.
Why this matters for production agentic systems
This is the pattern that makes agents shippable: scoped delegation with human checkpoints. Not full autonomy, not pure chat—something in between that acknowledges users want both control and offloading.
The screen understanding piece is critical. If you're building task automation, text-only prompts are a bottleneck. Real users want to screenshot, snap a photo, or share what's already on their screen. Google's betting multimodal input lowers the "intention-action distance" enough to make this useful.
Gemini Notebook ships preinstalled on foldables
Google is bringing Gemini Notebook (formerly NotebookLM) directly to Samsung's new foldables as a preinstalled app. This is the first time we're seeing Google's research-focused AI product treated as core device functionality rather than a standalone web tool.
The pitch: drag and drop photos, documents, whiteboard images, voice recordings into a side-by-side workspace on the Fold8's large display. Chat to ask questions about your sources, or generate custom resources—slide decks, videos, flash cards, quizzes, or podcasts—tailored to how you learn.
Every Galaxy Z Fold8 Ultra, Fold8, and Flip8 includes a six-month trial of Google AI Pro, which unlocks features like Audio Overviews, Video Overviews, and Slides.
The real story: Google is embedding research tools in consumer hardware
NotebookLM started as a niche tool for researchers and students. Rebranding it as Gemini Notebook and preinstalling it on flagship Android devices is a bold positioning move. Google's arguing that multimodal synthesis and custom artifact generation belong in everyone's pocket, not just power users.
The foldable form factor actually makes sense here. You need screen real estate to work with multiple sources simultaneously. The drag-and-drop interaction model is native to tablets and large-screen devices. This feels less like "AI slapped onto a phone" and more like thoughtful product-market fit.
The six-month AI Pro trial is the hook. Google's betting users will hit the feature ceiling on the free tier quickly enough to convert to paid. Whether that works depends entirely on whether the artifact generation—especially Audio and Video Overviews—delivers enough value to justify subscription pricing.
Gemini on watches and glasses: gesture control and always-on activation
The Galaxy Watch 9 gets "more convenient access" to Gemini—raise your wrist and it activates without a wake word. Small detail, big friction reducer.
The glasses story is more ambitious. Google announced intelligent eyewear at I/O 2026, created with Samsung and launching this fall. They'd already revealed designs from Gentle Monster and Warby Parker; today they added two more frames.
The interaction model: if you're wearing a compatible Wear OS watch, you can control your glasses with hand gestures. No voice commands in public, no fumbling with frames.
Why glasses are the hard problem
Smart glasses have failed repeatedly because the input modalities are terrible. Voice is socially awkward. Touch controls on frames are clumsy. Gesture control from a paired watch sidesteps both problems—assuming the latency and recognition accuracy are good enough.
Google's betting on ambient computing: Gemini running across devices with context sharing. Your watch knows your gestures, your glasses show the output, your phone handles heavy compute. This only works if the orchestration layer is seamless.
The challenge is convincing people to wear yet another device. Watches have achieved mainstream adoption. Glasses are still niche outside of specific use cases (cycling, industrial work). Google needs compelling demos beyond "here's another way to ask an AI assistant questions."
What's missing: the production readiness questions
Google's announcement is careful to say features will "roll out as they are ready" so users aren't "left waiting." That's both honest and telling. Multi-device agentic systems are hard to ship reliably.
Questions I'd want answered:
- Error handling: When task automation fails mid-flow across multiple apps, what's the recovery UX? Does the user get dumped back to manual mode, or does Gemini attempt a retry?
- Privacy boundaries: What happens to the screen content and images Gemini processes for task automation? On-device only, or does complex reasoning require cloud round-trips?
- Latency: How fast is the gesture→watch→glasses→Gemini→output loop? If it's over 500ms, the magic breaks.
- App ecosystem incentives: Which 40+ apps are participating, and what did Google offer them to integrate? If it's all Google properties plus a handful of partners, adoption will plateau.
The fact that Google's bundling a six-month AI Pro trial suggests they're still figuring out the unit economics. Free trials are a product team's way of saying "we think this is valuable but need data to prove it."
The bigger picture: Google's building the Android agentic layer
This isn't just Samsung device news. Google's using Samsung's flagship hardware as the delivery vehicle for Gemini Intelligence—their term for proactive, cross-device AI orchestration.
The pattern is clear: take advanced AI capabilities (multimodal reasoning, task automation, research synthesis) and make them native to the OS layer, not trapped inside chat apps or web tools. Preinstall the software, bundle the premium features as trials, and integrate tightly with hardware partners.
Apple's doing something similar with Apple Intelligence, but Google has the Android ecosystem advantage: they can ship across hardware partners and iterate faster. The Samsung partnership is the beachhead.
The competitive dynamic is shifting from "who has the best foundation model" to "whose agentic orchestration layer ships on the most hardware with the least friction." Google's making a strong play here.
If Gemini Intelligence delivers on the promise—truly offloading digital chores, synthesizing research seamlessly, controlling devices through natural inputs—then we'll look back at this Unpacked event as the moment agentic AI moved from demos to production at scale. If it doesn't, it'll be another ambitious Google pivot that quietly fades.
The hardware's shipping this fall. We'll find out soon enough.