Agentic AI can perform tasks on the user’s behalf, such as booking a restaurant and planning a wedding. In three years, the internet has moved from the search era, in which users searched for information, to the AI era, in which AI systems generate content, and now to the agentic era, in which AI can act.

Google Search, owned by Alphabet, illustrates the shift. In the search era, it returned a list of blue links to the user query. In the AI era, it generated an answer grounded in sources.

The shift has already reshaped behaviour. According to Google, AI Mode—launched just a year ago—has changed how people search: queries are longer and asked through various channels, such as images, video, and voice. It has also changed what they search for. For instance, planning-related queries on AI Mode have grown 80% faster than the rest over the past six months—a preview of the agentic era to come.

Google is now pressing further in. On 19 May, Google unveiled new agentic capabilities for search, booking and shopping, letting users complete tasks without leaving the page. It also launched Gemini Spark, a 24/7 personal agent that runs continuously in the background even after users shut down their devices.

Google moves fast because competition is fierce. Agentic AI is reshaping the boundaries of digital markets. Newcomers like Perplexity, OpenAI and Anthropic have entered the search and are now offering agentic features. For instance, Perplexity Computer, launched in February 2026, already runs background tasks across a user's folders, apps and the web. Over time, such systems could evolve into personal AI-driven operating systems, reminiscent of the AI assistant in the movie Her. If that happens, these operating systems will compete directly with the existing ones from Apple, Google, and Microsoft for control of the next interface.

Search and operating systems are not the only markets that agentic AI reshapes. E-commerce is already agentic. In the United States, Google and Amazon can now shop for customers, intensifying a heated fight over shopping queries. China's giants are pressing on, too. On 11 May, Alibaba put its chatbot, Qwen, on Taobao, its e-commerce platform. Since March, ByteDance has been testing its chatbot, Doubao, on Douyin, China's TikTok e-commerce services.

Europe is another matter. Alphabet, Amazon, Apple, Booking, ByteDance, Meta, and Microsoft must do so under the watchful eye of EU competition rules—antitrust law and the Digital Markets Act (DMA), the bloc's bespoke regime for regulating large digital platforms. Among other things, the DMA forbids them from favouring their own services over rivals (“self-preferencing”) and obliges them to open their hardware and software to third parties (“interoperability”).

These rules are already impacting AI in the EU. Google delayed its AI features, including AI Mode, for nine months owing to regulatory concerns. Months after launch, the scrutiny has begun. In December 2025, the European Commission opened an antitrust probe into how Google uses publishers and YouTube content to fuel its AI products. In February 2026, news publishers lodged a formal antitrust complaint before the Commission, accusing Google of harvesting their work without fair payment. In France, a national neighbouring rights law that requires Google to bargain with news publishers before using their material still keeps AI Mode out of the country. A separate DMA probe, running since March 2024, alleges that Google is self-preferencing. That could complicate plans to plug agentic tools, such as shopping and hotel booking, into Google Search, where they would directly compete with rivals like Booking.com.

Apple has likewise delayed launching Live Translation on its AirPods in Europe, citing the difficulty of implementing a DMA-compliant interoperability fix that meets the DMA requirements for product readiness by launch day. Google is also under an ongoing DMA proceeding into interoperability requirements for Android with third-party AI services, which may lead the company to delay or abandon the deployment of its own AI services if no workable interoperability fix is ready for launch.

The effect is that Europeans might wait longer to use agentic AI. Brussels is right to police the competition process; it is wrong to impose obligations that distort it with mounting legal uncertainty and disproportionate compliance burdens and costs.

If Europe wishes to quickly join the agentic era rather than merely regulate it, it must give its rulebook a measure of legal certainty and proportionality.