The keynote at Google I/O 2026 pushed the company further into an AI-first direction, with multiple new models and platform features that aim to change how people search, build, and interact with content. At the center of the announcements are refreshed model families and agentic tools that run in the background to automate tasks across apps. The presentation emphasized speed, multimodal inputs, and tighter guardrails, while also showing off new hardware and developer tooling designed to make AI more directly useful inside workflows.
These changes affect general users, developers, and enterprise teams differently, but the throughline is the same: Google is turning Search and its productivity suite into an environment where answers, actions, and mini-applications can be produced without a click. Expect more synthesized responses, interactive visuals, and always-on helpers that can read emails, draft content, and monitor data sources. Below, the main headlines are grouped by model updates, agentic capabilities and developer tools, and platform-level changes such as search, shopping, image provenance, and XR.
New Gemini models and a refreshed app experience
Google introduced a next wave of models beginning with Gemini 3.5 Flash, which will become the default model in the Gemini app and in AI Mode in Search. A higher-capability sibling, Gemini 3.5 Pro, is scheduled to arrive next month. Google claims the Flash model improves responsiveness, handles more agentic tasks, offers stronger coding agents, and can produce richer web interfaces and graphics. Alongside the models, the Gemini app receives a visual overhaul Google calls “neural expressive,” featuring new animations, color accents, a new font, and haptic feedback; this redesign begins rolling out on May 19th for web, Android, and iOS. The company also unveiled a new family, Gemini Omni, with Omni Flash available across the Gemini app, Google Flow, and YouTube Shorts starting today; Omni aims to accept mixed inputs — text, images, audio, and video — to create short clips and, eventually, more complex outputs from any combination of inputs.
Agentic assistants and developer-focused tooling
Two major developer and automation directions stood out: background agents that act on your behalf and tools that let creators build native Android experiences quickly. The always-on assistant, Gemini Spark, runs on virtual machines in google cloud to monitor tasks like drafting emails, compiling study guides, and spotting hidden fees across subscriptions. Spark connects to Google Workspace apps (Docs, Gmail, Sheets, Slides) and third-party services such as Canva and Instacart, with planned expansion to access local files via the Gemini app on macOS. For builders, AI Studio now supports “vibe-code” for generating full native Android apps, including an embedded emulator, phone-side installation for testing, and export to Android Studio or GitHub, plus ZIP output. Google said developers will soon be able to publish vibe-coded apps privately to friends and family, with Firebase integrations arriving later.
Gemini Spark and Workspace integration
The background assistant model is built to be practical: Spark uses Gemini 3.5 Flash as its engine and offers hooks into productivity flows so users can automate monitoring, summarization, and policy checks. Because Spark can access documents and app data, Google emphasized privacy controls and guardrails intended to reduce harmful outputs. For teams, this means automations that pull from Drive and Gmail can save time but also require governance. Google plans incremental expansion so Spark can operate with local files on macOS and integrate more deeply with third-party services, making the assistant a cross-application companion rather than a single-app feature.
Search, shopping, image provenance, and XR developments
Search is shifting toward multimodal queries and interactive results. Google expanded the search field to support longer, contextual prompts and inputs that include text, images, files, video, and even Chrome tabs. New information agents provide continuous summaries and updates on topics, with early access for AI Pro and Ultra subscribers this summer. Search will also generate interactive visuals and “mini apps” for frequently repeated tasks. On the shopping side, the new Universal Cart aggregates items across merchants such as Nike, Target, Walmart, Ulta Beauty, Sephora, Wayfair, and Shopify, surfacing compatibility warnings and loyalty perks; it arrives in Search and Gemini this summer, with YouTube and Gmail support later. For images, Google is expanding provenance tools using SynthID watermarking and C2PA credentials so users can inspect creation and edit histories in Search and Chrome.
XR hardware, Workspace images, pricing, and availability
Hardware updates included a refreshed look at Project Aura, developed with Xreal, where the external compute puck was redesigned and now includes a fingerprint sensor and lanyard support. New Android XR audio-only glasses from Warby Parker and Gentle Monster will ship this fall, offering features like live translation, Gemini-powered navigation, and notification summaries. Google showed the continued evolution of Beam (formerly Project Starline) with a demonstration of Sophie, a personable video agent that can read documents shown to the camera and participate in group calls; Google compared the experience to high-end systems like the $25,000 HP Dimension setup. Finally, pricing for Google’s premium tier shifted: when introduced at I/O 2026 the AI Ultra plan cost $249.99 per month; Google now offers a lower entry at $100 per month and a $200 per month tier that includes Project Genie, aligning pricing more closely with market alternatives.
