Google used its I/O 2026 keynote to announce a major overhaul of Search, introducing AI agents, a redesigned search box, and agentic coding capabilities that can generate custom apps and dashboards on the fly. This marks one of the most significant updates to the world's most popular search engine in over two decades, reflecting the company's aggressive push to integrate generative AI into every corner of its ecosystem.
A new search box
The most visible change is a revamped search box that Google is calling the biggest update to the service in over 25 years. The new box dynamically expands as you type, offers AI-powered suggestions that go beyond autocomplete, and accepts text, images, files, videos, and Chrome tabs as inputs. This multimodal input capability allows users to start a search with a screenshot, a video clip, or even a live webcam feed, and the system will interpret the content to provide relevant results. Google has long experimented with visual search through Google Lens, but this integration brings all input types into a single, unified interface.
Under the hood, the revamped search box is powered by Gemini 3.5 Flash, Google's newest Gemini model optimized for agentic tasks and coding. The model allows the search box to anticipate user intent and suggest follow-up questions, similar to how a human assistant might guide a conversation. For example, if you type "best hiking trails near Seattle," the search box might ask if you prefer easy, moderate, or difficult trails, or if you want trails with water features. This dynamic interaction is designed to reduce the number of searches needed to find the right information.
The new search box is rolling out today in all countries and languages where AI Mode is already available. AI Mode, introduced earlier this year, provides conversational AI responses directly in Search results. With this update, AI Mode becomes more proactive and context-aware. Google is also upgrading AI Mode's default model to Gemini 3.5 Flash, which brings faster response times and enhanced reasoning capabilities. Users who subscribe to Google AI Pro or Ultra will get additional features, including priority access to the latest models and longer context windows.
Search agents and mini apps
Google is also introducing what it calls "information agents" to Search. These background agents continuously scan the web, financial data, sports scores, and social posts, and notify you when something relevant to your query changes. The idea is that you describe what you're looking for once, and the agent handles the monitoring from there. For instance, a user could set up an agent to monitor stock prices for a specific company, and the agent would send a notification if the price drops below a certain threshold. Similarly, an agent could track social media mentions of a brand or product, alerting the user to emerging trends or crises.
Information agents are built on Google's existing infrastructure but leverage Gemini's ability to understand natural language instructions and execute multi-step tasks. The agents can be configured to check for updates at regular intervals—every hour, daily, or in near real-time—depending on the user's subscription tier. Google has emphasized that users maintain full control over what data the agents access and how results are delivered. Notifications can be sent via email, Google Chat, or the Google app itself.
Information agents will launch first for Google AI Pro and Ultra subscribers this summer. Pro subscribers will be able to create up to five agents, while Ultra users can create unlimited agents. Google has not yet announced pricing for these tiers, but AI Pro currently costs $19.99 per month, and Ultra is $29.99 per month. The company expects to roll out information agents to free users later, though no timeline has been given.
Separately, Google is bringing agentic coding to Search. Using its new Antigravity platform, Search can now generate custom UI, interactive visuals, and mini apps tailored to specific queries. With this, building a wedding tracker or fitness dashboard becomes a one-prompt task. For example, a user searching for "track my marathon training" could receive a fully functional dashboard that displays running logs, pace charts, and recovery recommendations, all generated on the fly by Gemini 3.5 Flash. The mini apps are not static; they can be updated with new data by asking the agent to modify the design or add new features.
Antigravity is a new platform from Google that combines code generation, vector databases, and real-time data streaming. It allows Search to create apps that are not just static HTML but fully interactive with custom logic. The underlying architecture uses a plugin system that can integrate with external APIs, such as weather services, stock market feeds, or calendar apps. This means a mini app generated for tracking a wedding guest list could automatically sync with Google Sheets or send email invitations via Gmail.
Custom mini app creation will roll out to AI Pro and Ultra subscribers in the US first, while the generative UI—which allows any search result to include an interactive visual element—will become available to all users this summer. For example, searching for "compare iPhone and Android" might return an interactive chart that updates automatically as new data comes in. This feature is expected to greatly enhance the utility of search results, particularly for data-intensive queries like financial comparisons, product reviews, or travel planning.
Finally, Google is expanding Personal Intelligence, which connects Search to Gmail and Google Photos for more personalized results, to nearly 200 countries across 98 languages, with no subscription required. Personal Intelligence originally launched as a limited beta in select markets earlier this year. It allows Search to answer questions like "What time is my flight?" by pulling information from your emails, or "Show me photos from my trip to Japan" by accessing your photo library. The expansion means that billions of users will now be able to benefit from this feature without paying a cent.
The expansion also brings new capabilities to Personal Intelligence. Users can now ask more complex questions that combine data from multiple sources. For example, "What email did I send to Sarah last week about the budget meeting?" would search both Gmail and the user's calendar to find the right message. Privacy controls are front and center: users can revoke access to specific services at any time, and Google says that all personal data processing happens on-device or through encrypted channels. The company has also committed to not using personal data for ad targeting without explicit consent.
These announcements come as Google faces increasing competition from Microsoft Copilot, which integrates AI directly into Windows and Office, and from startups like Perplexity AI, which offer agentic search experiences. By embedding agents, coding, and personalization directly into Search, Google is betting that its vast ecosystem—spanning email, photos, maps, and cloud—will give it an edge over standalone AI products. However, privacy advocates have expressed concerns about the amount of data Google will be able to access through personal agents and the potential for misuse if an agent's instructions are accidentally too broad.
To address these concerns, Google has introduced an "Agent Control Panel" within Search settings, where users can view all active agents, see their recent activities, and adjust permissions. Agents also have a limited lifespan by default; they expire after 90 days unless the user explicitly renews them. This is designed to prevent forgotten agents from accumulating sensitive data. Additionally, Google has published a public security audit of the Antigravity platform, performed by a third-party firm, which found no critical vulnerabilities in the code generation pipeline.
The broader context of these updates is Google's ongoing transformation from a keyword-based search engine to an agentic assistant. The company's CEO, Sundar Pichai, framed the changes as a natural evolution: "Search is not just about finding information anymore. It's about taking action on it. These agents will help you monitor, analyze, and create—all from the search box." Analysts note that Google is essentially turning Search into a platform for building small software tools, which could have far-reaching implications for the app economy. If users can generate a custom wedding tracker in seconds, why would they pay for a dedicated app?
Nevertheless, technical challenges remain. Generating high-quality interactive apps on demand requires significant compute resources, and Google has acknowledged that some complex queries may take several seconds to process. The company is using a technique called "speculative decoding" to speed up Gemini 3.5 Flash, but for ultra-responsive interfaces, it may cache common mini app templates. Another challenge is handling ambiguous queries. If a user asks for a "tracker for my plants," the system must infer whether they want a watering schedule, a growth log, or both. Google says the system will ask clarifying questions when needed, but early testers report that it sometimes makes wrong assumptions.
The rollout of these features is phased. The redesigned search box is available today, though some users may see it gradually over the next week. Information agents are expected to arrive in August for Pro and Ultra subscribers, with generative UI following in September. Personal Intelligence expansion is already underway, with all supported countries expected to have access within the next month. Google has promised to keep developers updated through its blog, and is offering early access to the Antigravity platform for a limited number of third-party developers who want to build custom agent actions.
Source: Digital Trends News