Google I/O 2026 Was About Turning Every Google App Into an Agent

Google I/O used to be where Google showed the next version of Android, a smarter Search feature, and a few developer goodies.

This year felt more like a systems diagram of a complete overhaul. The company used Sundar Pichai’s keynote to make one argument from every direction: Gemini is becoming the action layer across the internet Google already controls.

That sounds abstract until you map it onto a normal day. Your inbox briefs you before a meeting. Search monitors apartment listings while you sleep. Docs turns a voice ramble into a draft. YouTube answers a how-to question by jumping to the right part of a video. Shopping keeps a cart alive across Search, Gemini, YouTube, and Gmail. A coding agent builds in the background. Android shows what your agent is doing at the top of the screen.

The chatbot era asked you to open a separate app. Google’s I/O pitch was that the agent era starts when the AI fades into the default surfaces you already use.

First up, the TL;DR

Google used I/O to make one thing clear: Gemini is becoming the layer underneath the apps people already use.

That matters because the AI race keeps drifting away from “which chatbot answers best?” and toward “which company can turn answers into action?” Google’s answer is obvious: put Gemini inside Search, Gmail, Docs, YouTube, shopping, Android, coding, app building, and glasses.

Here’s what happened:

• Gemini 3.5 Flash became Google’s new fast model for agents, coding, and long tasks.
• Gemini Spark became a 24/7 personal agent that can work across Workspace apps.
• Search got information agents that monitor the web, plus mini apps for ongoing tasks.
• Workspace got voice features for Gmail, Docs, and Keep, plus Google Pics and AI Inbox.
• Antigravity 2.0 turned coding agents into a managed desktop, CLI, and SDK workflow.

How to try it: Open Gemini or AI Mode in Search and give Gemini 3.5 Flash a multi-step task: compare options, explain tradeoffs, and draft your next move. Builders should try AI Studio for Android app generation, or Antigravity for parallel coding agents.

Why this matters: Google has the distribution advantage. ChatGPT and Claude are destination apps. Google owns the work surfaces: inbox, docs, browser, phone, YouTube, Search, shopping, and the developer stack.

That means Google can make agents feel less like a new habit and more like a new setting inside the tools you already open all day.

Our take: The hard part now is permission. A search engine that answers is useful. A search engine that monitors, books, shops, builds, writes, and edits needs trust for people to... y'know, actually use it.

If Google gets that right, the next AI interface may feel less like opening a chatbot and more like telling your computer what outcome you want. This is good.

Google’s actual announcement was distribution

The easiest way to miss the I/O story is to treat each announcement as a separate feature.

Google did launch a lot. It introduced Gemini 3.5 Flash, teased 3.5 Pro for next month, launched Gemini Omni, upgraded AI Mode in Search, introduced Gemini Spark, refreshed Workspace, upgraded AI Studio, expanded Antigravity, showed Android XR eyewear, previewed Android Halo, and pushed YouTube, shopping, and media creation deeper into Gemini.

But the bigger move is that Google now has enough usage to make this more than a demo reel. Pichai said Google processes over 3.2 quadrillion tokens per month across its surfaces. Tokens are the small chunks of text and data models process when they read, reason, and generate. He also said more than 8.5 million developers build with Google models monthly, while the Gemini app has passed 900 million monthly users.

That is the Google advantage in one sentence: the company can ship AI into places where billions of people already have work to do. The challenge for Google, therefore, is to get people to trust, like, and choose to use it.

OpenAI has historically had the consumer mindshare. Anthropic has the trust of a lot of builders and enterprises, and that trust is growing. Google has billions of users across Search, Gmail, Docs, Android, YouTube, Chrome, Workspace, shopping, Maps, Cloud, and more, plus the hardware path to put Gemini not only on your phone, but on your face.

That makes Google’s product strategy less like “win the chatbot app” and more like “turn the operating environment into an agent platform.”

Gemini 3.5 Flash is the engine

The model news centered on Gemini 3.5 Flash, which Google described as the first release in a new Gemini 3.5 family.

Flash is the important one because Google needs agent tasks to feel fast. Agents sound magical in a keynote, but they get annoying ASAP when they take forever to read pages, call tools, write drafts, and recover from mistakes.

Google says 3.5 Flash is built for complex, long-horizon tasks. That means jobs that require more than one step, like researching a topic, comparing options, using tools, and producing a final answer. It is available in the Gemini app, AI Mode in Search, Antigravity, AI Studio, Android Studio, Gemini Enterprise, and the Gemini API.

The model layer also picked up Gemini Omni, a new model family that can generate outputs in any modality from any input. Google is starting with video through Gemini Omni Flash, available in the Gemini app, Flow, and YouTube Shorts.

That matters because Google is trying to collapse creative workflows directly into conversation. Upload footage, swap the background, add a cinematic zoom, make a clip, remix a Short, and eventually generate across images, text, and video from the same model family.

The open question here is this: does this make creation more controllable, or does it make every feed feel like it was assembled by a committee of autocomplete dice rolls, and therefore not at all satisfying for the user at the end of the process?

Search also became the action surface

Search got the most strategically important update, because it is still Google’s front door to the web.

The new AI Mode uses Gemini 3.5 Flash as its default model globally. Google also introduced what it called the biggest upgrade to the Search box in more than 25 years: an intelligent search box that expands as you type, accepts text, images, files, videos, and Chrome tabs, and helps you formulate richer questions.

The bigger change is Search agents.

Google is starting with information agents: background agents that monitor the web, news, blogs, social posts, finance, shopping, and sports data for a specific ongoing question. Apartment hunting is the example Google used. You describe exactly what you want, and the agent scans for matches, then updates you when something changes.

Search also gets mini apps. If you are planning a wedding, moving homes, or building a fitness routine, Search can build a dashboard or tracker that keeps pulling from live sources like reviews, Maps, local data, and weather.

This is where the publisher and product questions get thorny. For users, Search becomes more useful because it remembers the task and keeps going. For websites, the link economy gets more complicated because Search is doing more of the work before the click.

That tension has been building since AI Overviews. I/O made the next step clearer: Search is evolving from a place that sends you somewhere into a place that tries to finish the job. As TechCrunch put it: Search as we know it is over. We'd go a step further: search as we know it is dead. Long live Search?

Workspace turns voice into work

The Workspace announcements were the easiest to understand because they hit everyday pain.

Google said more than 4 billion people use Gmail, Docs, and Drive. Now those apps are getting conversational voice features, Google Pics, AI Inbox updates, and Gemini Spark.

Docs Live turns a voice brain dump into a structured first draft. Keep can turn rambling notes into organized lists. Gmail gets voice capabilities. AI Inbox can draft replies, surface relevant files, and turn email chaos into task management.

Google Pics is a new image creation and editing tool built on Nano Banana. It gives users precise controls like selecting objects, changing text inside images, translating text while preserving design, and editing images inside Workspace apps such as Slides and Drive.

The reason this matters is simple: voice lowers the cost of giving the AI context. Most people have more detail in their head than they will ever type into a prompt box. If Gemini can turn a messy spoken explanation into a usable doc, email, note, or design, AI starts to feel less like prompting and more like delegating.

That is also why Thinking Machines’ interactivity bet rhymes with Google’s direction. The next interface shift may come from AI that can listen, watch, interrupt, and collaborate in real time.

Gemini Spark is the agent permission test

Spark is the most important consumer agent because it moves Gemini from answering into acting.

Google describes Gemini Spark as a 24/7 personal agent that can run in the background, integrate with Workspace, use the Antigravity harness, and keep working when your laptop is closed or your phone is locked.

Google’s examples are pretty practical:

• Parse monthly credit card statements and flag hidden subscriptions.
• Watch school emails and extract critical deadlines.
• Synthesize meeting notes across emails and chats.
• Draft a Google Doc and the companion email that kicks off a project.

Spark also exposes the central trust problem. Google says Spark asks before high-stakes actions like spending money or sending emails. That is the right default. The harder question is how users will understand what an agent is allowed to see, remember, summarize, and decide before it reaches the confirmation step.

In this sense, the permission screen becomes the new user interface. If it is too broad, users will get nervous. If it is too narrow, agents will feel dumb. If it is too frequent, everyone will click through like cookie banners (or me in a Claude Code session because I'm still too afraid to go full Yolo mode).

Developers got the same story

The developer announcements turned the same agent thesis into a builder stack.

Google AI Studio now connects directly to Workspace (love this), exports to Antigravity (makes sense), supports native Android app generation from prompts (this is so cool!!), runs in-browser Android previews, and lets builders deploy their first two apps to Google Cloud at no cost.

Antigravity 2.0 becomes the agent-first development hub. The desktop app orchestrates multiple agents in parallel. The CLI lets terminal users create agents without a graphical interface. The SDK gives developers programmatic access to the same agent harness powering Google products.

This matters because Google is trying to win two sides of the same market. Consumers get agents inside default apps. Developers get the tools to build agents on the same infrastructure.

The strategy is classic platform Google: make the core capability visible to users, then make the build surface irresistible to developers.

The hardware layer is back

Endless ink (or should we say pixels?) has been spilled on how Google Glass failed because the culture and technology were early. Android XR arrives in a completely different world: one powered by agents.

Google says intelligent eyewear with Gemini is coming this fall through partners including Samsung, Gentle Monster, and Warby Parker (hey, I wear those!). It described two types: audio glasses that speak help into your ear, and display glasses that show relevant information when you need it.

That pairs with Android Halo (coming later this year), which is a small agent status strip at the top of the phone screen. It shows what an agent is working on while you keep using other apps.

Those are small UI details with large implications. If agents become persistent workers, users need a kind of ambient status bar. They need to know when an agent is thinking, acting, waiting, or asking. Otherwise AI becomes another invisible process running in the background with unclear authority, and if it's costing you tokens you buy over the cloud, you're going to want to make sure it's not wasting them.

Reading between the lines

Google is trying to make the AI assistant disappear into the surfaces where work already happens.

That is a distribution bet, a trust bet, and a coordination bet all at once.

The distribution bet is obvious. Google already owns the places where billions of people search, write, watch, email, shop, build, navigate, and use phones.

The trust bet is harder. The more useful agents become, the more they need access to personal context. Gmail, Calendar, Drive, Docs, Search history, shopping behavior, and location are powerful ingredients. They are also the exact places where a bad permission model can break user confidence.

The coordination bet may be the hardest. Google has to make Gemini feel consistent across Search, Workspace, Android, YouTube, Gemini, AI Studio, Antigravity, and hardware. Some brutal honesty here for the ~700 something odd people from Google who read The Neuron every morning: so far, y'all have not done that.

There are wide gaps in the level of quality and reliability of the user experience across each different surface Google has released AI into so far (example: Gemini in AI Studio = great! Gemini in Google Docs / Sheets = unreliable & untrustworthy, in our experience), and it's high time they close those. This feels like a step in the right direction. But if each product team ships a slightly different version of “agent,” users will feel the seams.

That is why this I/O felt like a roadmap reveal most of all. Google showed all of the individual pieces of an agent operating system, then asked users, developers, and enterprises to believe it can wire them together.

How to actually try the new stuff

The practical version of this recap is simple: what can you open after reading this?

Some of Google’s I/O demos are live now. Some are rolling out by subscription, country, or trusted-tester access. And some are still “watch this space” announcements with no public button to click yet. Here is the useful map.

Start here if you want the consumer stuff

• Gemini 3.5 Flash: open the Gemini app or AI Mode in Search and give it a task with enough moving parts to matter. Try: “Compare these three options, explain the tradeoffs, and draft the next step I should take.” Developers can also test it in Google AI Studio, the Gemini API, Android Studio, and Antigravity.
• Gemini app redesign: open Gemini on web, Android, or iOS. Google says the Neural Expressive redesign is rolling out globally, with richer answers, Gemini Live baked in, and the new mic flow for longer spoken prompts.
• Gemini Omni: try Omni video generation in the Gemini app if your plan and region support it. For creative work, open Google Flow; Omni is the new model behind conversational video editing there. YouTube creators can try Omni through YouTube Shorts Remix and the YouTube Create app.
• Daily Brief: open the Gemini app and look for Daily Brief if you are on Google AI Plus, Pro, or Ultra. It pulls from connected apps like Gmail, Calendar, and Tasks to make a skimmable morning brief. The feature is starting in the U.S.
• Gemini Spark: there is no separate Spark app to download yet. It lives in the Gemini app, starts with trusted testers, and is headed to U.S. Google AI Ultra users in beta. Use the Google AI plan page to check eligibility.
• Gemini for macOS: download the Gemini Mac app now. The current version gives you a desktop Gemini shortcut and screen-context help; Spark and the bigger voice workflow are coming later this summer.

Use this if you want Search to do more work

• AI Mode in Search: go to AI Mode. Gemini 3.5 Flash is now the default model there, and the new intelligent search box is rolling out wherever AI Mode is available.
• Follow-up questions from AI Overviews: start with a normal Google Search. If you see an AI Overview, use the follow-up path into AI Mode to keep the context going.
• Information agents: these are coming this summer for Google AI Pro and Ultra subscribers. The use case is background monitoring: apartment listings, price changes, sports drops, finance updates, and other “tell me when this changes” tasks. Start from AI Mode when it appears.
• Agentic booking: Google says Search will help with bookings for local experiences and services this summer in the U.S. Try it from Search or AI Mode once the booking prompts appear.
• Search mini apps: these are coming in the next few months for Google AI Pro and Ultra users in the U.S. The idea is that Search can build a persistent tracker or dashboard, like a moving checklist or fitness plan. No separate app exists yet; it will live inside Search.
• Universal Cart: this is a summer rollout inside Search and Gemini, with YouTube and Gmail later. There is no standalone cart app yet. The pitch is one shopping cart across multiple retailers, with Gemini checking price history, compatibility, loyalty perks, and stock changes.

Use this if you live in Gmail, Docs, Drive, or Slides

• Gmail Live: open Gmail and watch for the Gemini Live-style voice search rollout. You will be able to ask your inbox questions out loud, like “What’s my flight gate?” or “What’s going on at school this week?”
• Docs Live: open Google Docs. When the feature reaches your account, use voice to brain-dump a rough idea and let Gemini structure it into a draft. Google says it can pull relevant context from Gmail, Drive, Chat, and the web with permission.
• Keep voice notes: open Google Keep. The new feature turns a rambling spoken note into organized notes and lists. It is rolling out this summer to Google AI Pro and Ultra subscribers and Workspace business previews.
• AI Inbox: open Gmail if you are on an eligible Google AI or Workspace plan. AI Inbox can draft replies, surface relevant files, and turn task-heavy email threads into a cleaner action list.
• Google Pics: open the Google Pics page. For now, trusted testers and some Workspace admins can get access through Workspace Experiments or Gemini Alpha features. The broader rollout comes this summer for Google AI Pro and Ultra subscribers and Workspace business customers.

Use this if you build software, apps, or agents

• Google AI Studio: start at AI Studio. Use the Build tab to prototype apps, connect Workspace data, generate assets, and deploy your first two apps to Google Cloud at no cost.
• Android app generation: in AI Studio, select “Build an Android app” and prompt for the app you want. Google says the flow creates Kotlin code, runs an Android emulator in the browser, and can publish to Google Play’s Internal Test Track. The Android team’s walkthrough lives here.
• AI Studio mobile app: pre-register at ai.studio. The mobile app is meant for capturing an idea on your phone, remixing examples, and continuing the build on desktop later.
• Antigravity 2.0: download or open Antigravity. Use the desktop app if you want to orchestrate multiple coding agents, background tasks, and subagents in parallel.
• Antigravity CLI: use Antigravity if you prefer terminal workflows. Google is steering Gemini CLI users toward the Antigravity CLI for agent creation and automation.
• Antigravity SDK: start from the Antigravity developer surface if you want to define custom agent behaviors and host them yourself.
• Managed Agents in the Gemini API: start with the Gemini API or AI Studio. The new managed-agent flow spins up an isolated Linux environment where an agent can reason, use tools, execute code, and resume with files and state intact.
• Gemini Enterprise Agent Platform: enterprise teams should start with Gemini Enterprise or their Google Cloud admin. This is where Antigravity connects to Google Cloud projects and enterprise agent workflows.
• Build with Gemini XPRIZE Hackathon: apply through the Gemini XPRIZE site if you want to build an app for the $2M hackathon.

Use this if you make media, design, or prototypes

• Google Flow: open Flow. Try Gemini Omni for video, use the new Flow Agent for brainstorming and batch edits, or build custom creative tools in natural language.
• Google Flow Tools: go to Flow’s tool gallery to try video resizers, image editors, storyboard helpers, type overlays, and partner-built tools like pixelBento.
• Google Flow mobile: Android users can try the Flow beta app. Google says iOS is coming soon.
• Google Flow Music: use Flow Music on the web, or download the iOS app. The new controls let you edit sections, rewrite or translate lyrics, create covers, and use Gemini Omni for music videos.
• Stitch: open Stitch. You can start from text, voice, code, or design files, then steer the design agent in real time. When it is ready, generate a shareable link through AI Studio, export to Antigravity, or publish with Netlify.
• Project Genie: open Project Genie if you are an eligible Google AI Ultra user. It generates interactive worlds, and the new Street View capability lets you anchor worlds in real places.
• Ask YouTube: U.S. Premium members 18 and older can try it through YouTube New. Ask YouTube turns broad video searches into structured answers, then points you to the most relevant parts of videos.
• YouTube Shorts Remix with Omni: open YouTube Shorts and look for eligible Shorts Remix flows. Omni remixing is rolling out at no cost, with watermarks and links back to the original video.

Use this if you care about science, provenance, or hardware

• Gemini for Science: start at Google Labs Science. The experiments cover literature insights, hypothesis generation, and computational discovery.
• Science Skills in Antigravity: use Science Skills if you want agent workflows connected to life-science databases and tools inside Antigravity.
• SynthID and content verification: learn about SynthID or use the Content Credentials verifier. Google says SynthID verification already appears in Gemini and is expanding to Search and Chrome.
• Android XR glasses: read the Android XR update. Audio glasses arrive first this fall; display glasses come later through partners like Warby Parker and Gentle Monster.
• Android Halo: read the Android Halo preview. You cannot use it today, but it matters because it shows the live status of agents like Spark while they work.

The fastest way to test Google’s I/O thesis is to pick one surface you already use every day and push it past a normal prompt. Ask Search to monitor something. Ask Gemini to plan something. Ask AI Studio to build something. Ask Flow to make something. That is where the keynote stops being a list and starts being a product strategy.

What to watch next

Three things matter more than the demo reel.

First, watch adoption in places where users already have urgent work: Gmail, Docs, Search, and Android. If agents work there, Google will not need to teach people a new habit. I'd also be concerned if Google tries to revoke access to its apps via MCP to other AI models; this would be a fool's bet to try to juice interaction with the AI in the app itself, but as anyone using AI knows, you don't necessarily need or want different AIs in all the tools you use; you just want the data and tools you use available wherever you use AI, and vice versa. Put another way: you don't want to be locked in by the vendor, you want the vendor to bend itself into whatever shape best fits wherever you are doing your work.

Second, watch permission design. The company that makes agent permissions feel understandable and easy to use could have a real moat. So far, that's been Anthropic: their permission system is digestible, easy to understand, and trivial to control. On the flip side, any company that makes these permissions feel overly burdensome will be ignored.

Third, watch whether builders adopt Antigravity and AI Studio as production tools, not only as fun demo machines. Google themselves have been focused heavily on this, because they rightfully see code as the key to making everything else work (that, and the engineering market is the grand prize to be won for generating recurring revenue in AI right now).

After I/O, there's no longer any question as to whether or not Google has an AI roadmap. It clearly does. The question is whether everyday non-AI pilled "normie" users want the company that organizes the world’s information to start acting on it for them. If Gemini works well in every surface it's deployed, then they'll slowly but surely get "one-shotted" as the industry calls it. If Gemini continues to succeed or fail more often than not in one surface over another, than the accusations of AI slop and general AI haterade will continue to flow among said normie masses like fine wine. And there's no shortage of disdain for AI these days...