GoKawiilWe've joked for the past several years that Google I/O, the company's developer conference, is really more like Google A/I. There's a lot of truth behind it; so many of Google's apps have gotten an AI makeover. But in a turbocharged field like AI, things change quickly. So as we prepare for Google I/O 2026, we're revisiting what's changed over the past year.
Google's approach to AI is multidimensional. Gemini, the company's AI, is embedded in every part of the business, from hardware like new AI-forward Pixel phones and smart glasses to AI integrations into every piece of software like Search, Google Docs, Gmail and more. Gemini, the chatbot, has also led the pack in a competitive industry and carved out industry-first capabilities.
The AI industry as a whole has changed a lot over the past year, too. Vibe coding and claws, or agents, have taken off. Data centers to power AI are multiplying, and they've sparked significant backlash in the communities where they're proposed due to environmental and economic concerns. A computer chip shortage, sometimes referred to as RAMaggedon, has affected every part of the tech industry. There are still serious questions around AI's role in our mental health and worries about AI leading to layoffs, job loss and an internet that's overflowing with AI slop.
All of Google's AI work over the past year has set the foundation for its future work. Here are the major trends we've seen from Google over the past year and what they suggest about what may come next.
Watch this: Google I/O 2026: New Gemini, Smart Glasses and a Whole New Laptop OS. Here's What to Expect 06:44
Latest Gemini updates
Google leapfrogged ahead of competitors in November 2025 with Gemini 3. At the time, Gemini 3 had the most advanced AI capabilities of any model, sparking "code red" worries inside competitors like OpenAI and setting the technological foundation for the Gemini updates we've gotten since then. In an "icing on the cake" moment, Apple officially picked Gemini to power its smarter Siri, reportedly paying $1 billion for the pleasure.
Gemini 3.1 Pro is the latest model, building on its reasoning abilities. Google doesn't do as many iterative models as OpenAI and Anthropic, but it does regularly release new features. Over the past few months, we've gotten the ability to generate files and interactive images directly in chats, a new MacOS app, NotebookLM integrations and more.
Agentic AI is another area where Google has been investing. Gemini agents can run processes and autonomously complete tasks. But Google recently shut down its experimental Project Mariner, which could be used to run agents across the web. The project's capabilities will be used in other Gemini projects, and we will likely hear more about Google's agentic AI future at I/O -- including, possibly, more about Project Astra, Google DeepMind's tech that helps AI "see" or identify objects like in Gemini Live's camera mode.
Google is also one of the only major AI companies to give us a peek into AI's environmental cost. Unsurprisingly, it said that the cost of an individual prompt is "minuscule." But when you multiply that by the billions of prompts given each day -- and an increasing number of agents operating independently -- those numbers rise. Making AI more efficient through new hardware, like chips and software changes, has become a priority for AI companies battling rising costs while still developing new advanced models.
... continue reading