Google provided a shared NotebookLM with 45 sources of everything covered during this year’s recent 2026 Google IO.
Here is the link to the notebook shared by Google:
https://notebooklm.google.com/notebook/87ae4230-9dda-445a-9775-df61ad7044dc
I used NotebookLM to create a summary of the opening keynote’s top news, top developer announcements and highlights, and what Demis Hassabis talked about.
Request:
Create a summary of top news in the opening Google IO 2026 keynote.
Response:
Google’s I/O 2026 opening keynote marked a decisive shift into the “agentic era,” focusing on AI systems that move beyond answering questions to autonomously acting on users’ behalf [1, 2]. CEO Sundar Pichai noted the staggering scale of AI adoption, revealing that Google now processes over 3.2 quadrillion tokens per month [3].
Here are the top announcements from the keynote:
Core AI Models: Gemini 3.5 and Gemini Omni
- Gemini 3.5 Flash & Pro: Google introduced its next-generation Gemini 3.5 family, starting with Gemini 3.5 Flash, an exceptionally fast model built for agentic workflows and coding [4]. It outperforms previous models, operates four times faster than other frontier models, and is available at less than half the cost [5-7]. Gemini 3.5 Pro is currently in internal testing and will launch next month [8].
- Gemini Omni: Billed as a massive step toward artificial general intelligence (AGI), Omni is a new “world model” that can create anything from any input [1, 9]. Starting with video, Omni allows users to naturally edit videos via conversation, transform environments, and accurately simulate physics like gravity and kinetic energy [10-12].
The Era of Personal Agents: Gemini Spark Google announced Gemini Spark, a 24/7 personal AI agent that navigates your digital life and takes actions on your behalf [13]. Because it runs on dedicated cloud virtual machines, it continues to work in the background even when your devices are closed [13]. Spark integrates seamlessly with Google Workspace to handle multi-step tasks—such as tracking RSVPs for a party, parsing emails to create checklists, and eventually booking services—all while remaining under user control [14-17].
Agentic Search and Commerce
- Search Agents & Generative UI: Google Search received its biggest upgrade in 25 years with a completely reimagined, AI-powered search box [18]. Search is introducing Information Agents that monitor the web 24/7 to notify you of specific market movements, real estate listings, or product drops [19-21]. Additionally, Search now features agentic coding capable of generating custom, interactive UIs—like playable astrophysics simulations or personalized weekend planning apps—on the fly [22-25].
- Agentic Commerce: Google is laying the groundwork for a new shopping experience with the Universal Cart, which works across Search, YouTube, and Gmail to track prices and flag product incompatibilities [26-28]. This is powered by new open-source standards like the Universal Commerce Protocol (UCP) and the Agent Payments Protocol (AP2), which allow your agent to securely make purchases within strict budget guardrails [29-31].
Developer Platforms: Google Antigravity 2.0 Google unveiled Antigravity 2.0, a standalone, agent-first development platform [32]. To demonstrate its power alongside Gemini 3.5 Flash, Google tasked Antigravity with building a functioning computer operating system from scratch [33]. Over 12 hours, an autonomous team of 93 subagents wrote the code, ran tests, and successfully produced an OS capable of running the game Doom, costing less than $1,000 in API credits [33-35].
Intelligent Eyewear Google provided a first look at its upcoming audio glasses, arriving this fall [36]. Created in partnership with Samsung, Warby Parker, and Gentle Monster, the glasses feature discreet, over-ear speakers and a camera [37, 38]. Users can tap the frames or speak to Gemini to get walking directions, translate signs, order coffee, and summarize missed texts without taking out their phones [39-41].
Everyday Workspace and Creative Tools
- Voice-Powered Creation: New features like Docs Live and Gmail Live allow you to “brain dump” spoken ideas while Gemini organizes them into structured documents and emails in real-time [42-44].
- Gemini App Redesign: The Gemini app (which has now surpassed 900 million monthly active users) was completely overhauled with the vibrant Neural Expressive design, which replaces walls of text with dynamic layouts and interactive widgets [45-47]. It also includes a new Daily Brief agent that synthesizes your calendar, tasks, and emails into a morning digest [48].
- Google Pics & Flow: For creatives, Google announced Google Pics, a highly precise Workspace tool for editing and segmenting images [49]. Additionally, Google Flow users can now use Gemini Omni to conversationally edit and iterate on video projects [50].
Hardware, Science, and Trust
- TPU 8th Generation: To power these advancements, Google announced a dual-chip approach with TPU 8t (for training large models) and TPU 8i (optimized for lightning-fast inference) [51-53].
- AI for Science: DeepMind CEO Demis Hassabis introduced Gemini for Science and highlighted tools like WeatherNext, an AI model that successfully predicted the historic Category 5 Hurricane Melissa striking Jamaica five days in advance [54-56].
- SynthID Expansion: Google’s digital watermarking technology for AI content, SynthID, is expanding its industry footprint, with OpenAI, Kakao, and ElevenLabs adopting the standard [57].
Request:
Create a summary of the top developer announcements and highlights.
Response:
Here are the top developer announcements and highlights from Google I/O 2026, which heavily focused on tools for building “agentic” applications:
Gemini 3.5 Flash and Managed Agents Google introduced Gemini 3.5 Flash, a model tailored for high-speed agentic workflows that runs four times faster than other frontier models and outperforms Gemini 3.1 Pro [1]. To help developers harness this, Google launched Managed Agents in the Gemini API, which allows you to spin up a custom agent with a single API call [2, 3]. These agents can reason, use tools, and execute code in an isolated, ephemeral Linux environment, and developers can easily define their behaviors and skills using markdown files like AGENTS.md and SKILL.md [3, 4].
Google Antigravity 2.0 Ecosystem The Antigravity development platform received a massive upgrade with Antigravity 2.0, a new standalone desktop application designed specifically to help developers orchestrate multiple agents to execute tasks in parallel [5]. For developers who prefer to stay in the terminal or use their own infrastructure, Google also introduced a lightweight Antigravity CLI and a new Antigravity SDK [5].
Google AI Studio Updates Google AI Studio introduced several new features to take ideas from prompt to production faster:
- Native Android “Vibe Coding”: You can now build production-quality native Android apps using Kotlin and Jetpack Compose directly in the AI Studio browser [6, 7]. It includes an in-browser emulator and allows for one-click publishing to the Google Play Console [7].
- Mobile App: A new Google AI Studio mobile app is now available for pre-registration, allowing developers to prototype, iterate on code, and share live deployments directly from their phones [6, 8].
- Seamless Integrations: Apps built in AI Studio can now natively call Google Workspace APIs [9]. Additionally, you can export your entire project—including conversation history and secrets—directly to Antigravity for local development [10].
- Free Cloud Deployment: Builders can deploy their first two apps to Google Cloud at no cost, with no credit card required [11].
Powering the Agentic Web via Chrome Google unveiled new tools to help developers prepare for the “agentic web”:
- WebMCP: An experimental open web standard that allows developers to expose structured tools (like HTML forms and JavaScript functions) so browser-based agents can interact with their websites reliably and precisely [12].
- Chrome DevTools for Agents: Coding agents now have direct access to DevTools capabilities—such as console logs and network traffic—to automatically verify, debug, and optimize code in real time [13].
- HTML-in-Canvas API: A new declarative API that lets developers integrate real DOM elements directly into a canvas with WebGL and WebGPU, enabling highly interactive and immersive 3D UIs [14].
Build with Gemini XPRIZE Hackathon To encourage developers to utilize these new models and platforms, Google launched a global hackathon featuring a $2 million prize pool—the largest ever for a hackathon—tasking builders with solving pressing global challenges [15].
Request:
What did Demis Hassabis talk about at this year’s Google IO?
Response:
At Google I/O 2026, Google DeepMind CEO Demis Hassabis discussed several major AI advancements, emphasizing his belief that Artificial General Intelligence (AGI) is just a few years away and that humanity is currently standing in the “foothills of the singularity” [1, 2].
His key announcements and talking points included:
Gemini Omni Hassabis introduced Gemini Omni, a new “world model” designed to create anything from any input, starting with video [3, 4]. He highlighted that Omni represents a step-change in simulating reality, possessing an intuitive understanding of physics, such as kinetic energy and gravity [3, 5]. He demonstrated how users can naturally edit and iterate on videos using conversational language, and officially launched the first model in the family, Gemini Omni Flash [6, 7].
Gemini for Science To help scientists accelerate their research, Hassabis announced Gemini for Science [8, 9]. He detailed how this suite of AI tools and prototypes can streamline daily scientific tasks, such as generating novel hypotheses, staying on top of newly published literature, and transforming research goals into usable code [9].
Advanced Simulations and WeatherNext Hassabis emphasized the importance of AI simulations for understanding complex, dynamic systems that are too difficult to model directly today [9]. He mentioned AlphaEarth Foundations, which acts as a digital twin of the planet to help address issues like food security and deforestation [10]. He also highlighted the real-world success of WeatherNext, an AI weather model that accurately predicted the historic Category 5 Hurricane Melissa striking Jamaica days in advance, allowing for early evacuations that saved lives [10, 11].
Healthcare and Drug Discovery Hassabis shared that the ultimate application of AI should be improving human health, noting that biological models like AlphaFold and AlphaGenome are already standard tools for millions of researchers [12, 13]. He also provided an update on Isomorphic Labs, where they are modeling molecular interactions to massively accelerate the development of new medicines [13]. He revealed that they are currently in the pre-clinical stage for potential treatments for cancer and immune disorders, working toward the ultimate goal of solving all disease [2, 13].
Request:
Which sessions were the most popular. Include links to the sessions.
Response:
I’m sorry, but the provided sources do not contain information about which specific sessions at Google I/O 2026 were the most popular, nor do they include links to the individual sessions. The sources focus primarily on the announcements made during the event, such as new products, models, and developer tools.
Google IO 2026 Announcements
https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-collection