Google I/O 2026: The Gemini Agent Era Is Official — But Controversies Have Already Exploded
On May 19, 2026, from the Shoreline Amphitheatre in Mountain View, Sundar Pichai opened Google I/O with a statement that summarizes the company's entire strategic direction: we have entered the era of Gemini agents. It is not an incremental update, nor is it a new feature to be enabled in settings. It is the complete redesign of how Google wants people to interact with their information, their devices, and their daily work. The announcements were numerous, ambitious, and technically impressive. However, the reactions were not solely enthusiastic — and understanding why is as important as understanding what was announced.
The protagonist model of the event is not the largest in the family, but the fastest and most efficient. Gemini 3.5 Flash becomes the default model in both the Gemini app and AI Mode within Google Search, available globally starting now. The choice is significant: Google is not aiming to impress with extreme reasoning capabilities, but rather to integrate a fast, capable, and low computational cost model into every user touchpoint.
Google announced what it defined as the most significant update to the search bar since its introduction over 25 years ago. The new search box integrates AI suggestions that go far beyond autocomplete: the system helps the user formulate questions more precisely, adds nuances and context the user might not have considered, and fluidly connects traditional search with AI Mode. The transition from query to AI response happens seamlessly, without interface interruptions. Google specified that the classic search bar will not disappear — and this clarification is not accidental, as we will see in the section dedicated to criticisms.
One of the most structurally relevant announcements concerns search agents: it is now possible to create, customize, and activate AI agents that work in the background 24/7 to find specific information and notify the user when they find it. You no longer search only when you have a question: the system searches proactively based on defined interests and goals. Personal Intelligence — the feature that allows Gemini to access the user's personal data to contextualize responses — is being extended to nearly 200 countries in 98 languages, without requiring a subscription.
Gemini Omni is the new multimodal generator capable of creating content — video, text, images — from any type of input. Video management was showcased as a high-end capability: direct editing through chat, generation from text, and contextual understanding of existing clips.
Gemini Spark is the new agentic mode designed for daily life. Built on Gemini 3.5, it manages practical tasks autonomously: bookings, calendar management, comparative research, and purchasing decision support. It doesn't just answer; it acts — integrating the user's personal information to contextualize every action.
Among the hardware announcements, Googlebook stands out: a new line of laptops designed from the ground up with Gemini Intelligence at the core, not as an add-on but as the operating system for the user experience. Available starting this autumn, they represent Google's attempt to build a hardware identity based on vertical integration with AI, in direct competition with Apple and Microsoft.
Google confirmed upcoming Android XR products, including smart glasses with AR displays. The presentation showed integrated AI functionalities: real-time translation, contextual search in the physical environment, and contextualized notifications. Availability is expected for autumn. On the software front, Ask YouTube brings the ability to ask questions about any video and get immediate answers and summaries, available in the USA from summer, while in Google Workspace, Gemini integration deepens with new features in Docs, Gmail, and Drive, providing personalized responses that adapt to the user's tone and style.
The announcements were technically solid and visually convincing. Yet, in the hours following the keynote, critical reactions multiplied — and they come from very different directions. Ignoring them would be an analytical error.
The loudest and most structured criticism concerns the impact of the new AI search on website traffic. This isn't a new controversy — AI Overviews have already existed and data on the decline of clicks has been documented for months — but Google I/O 2026 made official and irreversible a direction that many still hoped could change. The numbers are already heavy: according to Semrush data, 93% of searches performed in AI Mode end without a single click to an external site. Global organic traffic from Google dropped by 33% between 2024 and 2025, peaking at 38% in the United States. When an AI Overview appears, the click-through rate on links is halved: only 1% of users click on a link within the AI summary.
With Google I/O 2026, this dynamic is not corrected — it is accelerated. The new AI search bar, background agents, and AI Mode as the default experience: every single announcement pushes in the same direction, which is to keep the user within the Google ecosystem without ever sending them to the third-party sites that produced the information used to build the answers. FIEG — the Italian federation of publishers — had already filed an appeal with Agcom against AI Overview, accusing Google of violating the Digital Services Act. The Independent Publishers Alliance has reported Google to the European Commission for abuse of a dominant position. After Google I/O 2026, these legal cases have found concrete new arguments.
The logic of the problem is brutally simple: if you don't authorize Google to use your content for AI responses, you risk disappearing from results. If you do authorize it, the user stops at the AI answer and never reaches your site. There is no third option — and this asymmetry is at the heart of all ongoing legal disputes in Europe and the United States.
The second critical front concerns privacy. Personal Intelligence — the function that allows Gemini to access Gmail, Calendar, Drive, Photos, and other Google services — is being rolled out globally without a subscription. The fact that it is free is not good news for those concerned about their data: it means the business model is, as always for free Google services, profiling and advertising. An AI assistant that knows your emails, your calendar, your documents, your photos, and your search history has a profile of you that is enormously more detailed than any previous digital service.
Google presented these integrations as a benefit to the user — and it is, from a utility standpoint — but the implications for privacy were dismissed in a few lines during the keynote, with generic references to data protection systems that critics judge as entirely insufficient. The European GDPR places strict limits on this type of personal data processing, and several digital law experts have already pointed out that some of the announced functions might be difficult to reconcile with European regulations in their current form. It is not the first time Google has announced global features that then require significant modifications — or temporary blocks — to be released in Europe.
There is a deeper criticism that many analysts are raising. Google already controls about 90% of the search market in Europe and many other markets. Until now, this monopoly concerned traffic distribution: Google decided which sites were visible and which were not. With AI Mode as the default experience, the monopoly moves to a higher level: Google decides not only who is visible, but what is told to users. When the answer is not a link to a source but a synthesis produced by an AI model trained on others' content, the original source disappears from the user experience. Google becomes the sole narrator. This concentrates a power over information that no private company has ever held before, and which European regulators are beginning to examine with growing and worried attention. The risk is not just economic for publishers: it is systemic for the quality and plurality of information available to citizens.
A more technical critical thread, but one very present in developer conversations, concerns the distance between the demos shown at the keynote and the operational reality of the products. It is not a problem exclusive to Google — all major AI companies present optimized demos — but the extent of the Google I/O 2026 announcements has made this gap particularly visible. Gemini Spark, background agents, deep integrations with Workspace: all functions that appear fluid and reliable in demos, but in practice depend on an agentic reasoning capacity that current models still possess in an inconsistent manner. The developer community has learned in recent years to distinguish between what a model does in a controlled demo and what it does in production with real, unpredictable, and often poorly formulated inputs. Many are asking for real release times and conditions, not just indicative launch dates.
Google I/O 2026 confirmed that Google has a clear and consistent vision of where it wants to take its products: an AI ecosystem that accompanies the user at every moment of the day, on every device, with access to their personal data to maximize the relevance of responses. From a user experience perspective, this vision is genuinely powerful and in many cases desirable. The problem is that this vision comes at a cost paid by others: the publishers who lose traffic, the creators who see their content used without adequate compensation, the European users whose data is processed in ways potentially non-compliant with local regulations, and anyone concerned about having a plural information ecosystem not controlled by a single private actor.
Google I/O 2026 was not just a technical conference. It was a strategic policy statement on how digital information will be organized in the coming years. Understanding its implications — not just the features announced — is the most important reading to do right now.
