Google just dropped what might be the most comprehensive single-month AI product rollout I've ever seen. May 2026 was packed with announcements spanning models, hardware, applications, and infrastructure—all organized around one central thesis: the "agentic era" is here, and Google is shipping the full stack for it.
This wasn't a research preview or a waitlist launch. This was Google putting agentic capabilities into Search, Gmail, Shopping, Android, cars, glasses, and a brand-new laptop form factor called Googlebook. Let's break down what actually shipped and why it matters.
The Model Layer: Gemini 3.5 and Omni
The foundation of everything announced is Gemini 3.5, which Google positions as combining "frontier intelligence with action." The key phrase here is "action-taking capabilities"—this isn't just a smarter model, it's explicitly built to "reliably execute complex, multi-step agentic workflows across your apps."
That's the promise, anyway. Multi-step reliability is still the hard part of agentic systems. Google is clearly betting that 3.5's architecture can handle the planning, error recovery, and context management that breaks most agent demos in production.
Alongside 3.5, they announced Gemini Omni, described as a model that "can create anything from any input—starting with video." You can combine images, audio, video, and text as input and generate high-quality videos "grounded in Gemini's real-world knowledge."
This is Google's play in the multimodal generation space where models like Sora and others have been pushing boundaries. The "real-world knowledge" grounding is interesting—it suggests Omni isn't just trained on video datasets but can pull in factual knowledge from Gemini's broader training to keep generated content coherent and accurate.
Search Gets Agentic (and Builds Custom UIs)
The Search announcements are where things get wild. Google is launching "information agents in Search" that run 24/7 in the background, monitoring information on your behalf and sending detailed updates.
Think of it as persistent queries with push notifications—you tell Search what to track, and it continuously scans for relevant updates, then packages them with links and actionable context. This is proactive search rather than reactive query-response.
But the most technically ambitious piece is "Antigravity" and agentic coding capabilities that let Search "build generative UI and interactive visuals tailored to your questions, plus custom experiences like dashboards or mini apps for your ongoing tasks."
Google gives the example of asking Search to code a custom fitness tracker using real-time data like reviews, live maps, and weather. That's Search generating and running application code on the fly, not just returning links.
This is a huge architectural shift. Search is no longer just retrieval and presentation—it's computation and synthesis. The "intelligent Search box" they mention is its "biggest upgrade in over 25 years," which tracks given that the original Google Search box was literally just a text input and two buttons.
The Application Layer: Universal Cart, Gemini App, Android Halo
On the application side, Google shipped three interesting pieces of connective tissue:
Universal Cart is exactly what it sounds like: a persistent shopping cart that works "across merchants and across services." You can add items while browsing Search, chatting with Gemini, watching YouTube, or reading Gmail.
This is interesting because it requires deep integration across Google's product stack and presumably some standardization layer for handling checkout flows across different merchants. It's Google trying to own the shopping intent-to-purchase pipeline end-to-end.
The updated Gemini app is becoming what Google calls a "proactive 24/7 partner." Instead of just answering questions when you open it, it now features personalized daily briefs, manages your inbox, schedules appointments, and anticipates needs in the background.
They also introduced "Gemini Spark" (no details on what that actually is—probably a sub-feature or interaction mode) and a new UI. The pitch is agent-as-assistant rather than chatbot-as-oracle.
Android Halo is a new OS-level space for managing your agents. It lets you "see your agents' progress and receive contextual assistance without interrupting your flow." This is smart—if you have multiple agents running background tasks, you need a unified surface to monitor them without constant notification spam.
Think of it as a task manager for AI agents rather than processes. The OS is adapting to the reality that you'll have multiple autonomous systems running on your behalf simultaneously.
Hardware Purpose-Built for Agents
Google also shipped or announced hardware explicitly designed for this agentic world:
Googlebook is a new laptop form factor built "from the ground up for Gemini Intelligence." Key features include the "Magic Pointer" for contextual suggestions, custom widgets for task organization, and cross-device features with Android phones.
This is Google's answer to the question: what does a laptop look like when the OS assumes you have an AI agent available at all times? The Magic Pointer name suggests cursor-level contextual awareness—probably similar to the idea of the OS surfacing relevant actions based on what you're pointing at.
Fitbit Air is their smallest tracker yet—described as a "tiny, discreet pebble" with high-fidelity sensors for 24/7 heart rate, heart rhythm monitoring with AFib alerts, SpO2, resting heart rate, HRV, and sleep tracking. They're positioning it as a "proactive wellness partner," which fits the theme of anticipatory agents rather than reactive dashboards.
They also teased intelligent eyewear for Android XR that lets you get directions, send texts, and snap photos without pulling out your phone. Classic AR glasses pitch, now with agent capabilities baked in.
And Android cars are getting "highly conversational voice controls, proactive routing and richer entertainment options." Proactive routing is interesting—that's not just responding to "navigate to X," it's the car suggesting routes or stops based on your context and history.
The Rest: Music, 3D Simulation, Quantum, and Climate
A few other announcements that didn't fit the core agentic narrative but are worth noting:
- Google Flow Music partnered with Believe to give artists an "AI creative collaborator" throughout the music creation process
- Project Genie expanded to combine with Street View for simulating realistic, interactive 3D environments of real places
- REPLIQA is a $10 million initiative applying quantum computing and AI to life sciences across five universities
- AlphaEvolve is tackling real-world optimization problems in supply chains, chip design, molecular simulation, and power grids
- A DeepMind Accelerator program launched in Asia Pacific to support startups using AI for climate and environmental challenges
The quantum and life sciences piece is a longer-term bet, but the AlphaEvolve framing is interesting—Google positioning their optimization models as production-ready for industrial applications, not just research demos.
Content Transparency and Verification
Given all this generative capability, Google also announced expansions to content transparency tools in Search, Gemini, Chrome, Pixel, and Cloud to help users understand whether content is AI-generated.
This is table stakes now—if you're shipping multimodal generation at scale, you need provenance and labeling systems. The fact that it spans Search, browser, OS, and cloud suggests they're building this as a platform capability rather than per-product feature.
What This Actually Means
Google just shipped an entire vertical stack for agentic computing in a single month:
- Models with explicit action-taking and long-running task capabilities
- OS primitives (Android Halo) for managing multiple agents
- Application integration (Universal Cart, proactive Gemini app, Search agents)
- Hardware designed around persistent agent availability (Googlebook, Fitbit Air, eyewear)
- Developer surfaces (agentic coding in Search, Gemini for Science tooling)
This is Google's version of what OpenAI was trying to do with Operator, Anthropic with Computer Use, and others with various agent frameworks—except Google has the advantage of owning the full stack from silicon to search box.
The big question is execution. Agentic reliability at scale is still unsolved, and Google has a history of launching ambitious integrated experiences that don't quite gel (Google+, anyone?). But the architectural vision here is coherent: agents that run continuously, work across apps, generate custom interfaces on demand, and adapt to your context.
If even half of this works as advertised, May 2026 will be remembered as the month Google went all-in on agents. We'll see if the demos survive contact with production.