Google just shipped the playbook for the next decade. Here are the 10 moves from I/O 2026 you cannot ignore.

Sundar Pichai gave a 4-year AGI countdown, shipped 24 announcements in 35 minutes, and ended the app era on stage.

May 23, 2026

5 minutes. 24 announcements.

1 thesis:

Agents replace apps. Search becomes software. Glasses replace phones for the small things. AGI is the destination, not the slogan.

I watched the full Google I/O 2026 keynote so you do not have to.

Here are the 10 that matter.

1. Gemini Spark runs on Google Cloud 24/7

Then it ran Doom on it.

The most important developer moment of the keynote was a 1990s shareware game.

Antigravity 2.0 is a standalone desktop application. Agent-first, not editor-first. The harness gained sub-agents, hooks, and asynchronous task management. Gemini 3.5 Flash was co-optimized with it.

The agents built an OS. The OS ran Doom. That is the whole thesis in one image: agents now produce non-trivial, working software systems end-to-end.

Ships today, globally: full CLI, an SDK, native voice with Gemini audio models, and integrations with Android, Firebase, and Google AI Studio.

Cursor defined the category. Claude Code scaled it. Google made it a first-class platform with native voice, an SDK, and a working OS demo on stage. Developer mindshare is the prize, and Google brought the proof.

For where the coding-agent race stands now, see the complete guide to AI coding in 2026, the AI code review checklist that prevents the next $1M production incident, and Claude Cowork: the tool that triggered a $285B software selloff.

2. Antigravity 2.0 built a working operating system from scratch

The most important developer moment of the keynote was a 1990s shareware game.

“Build a working operating system from scratch. But clearly, this isn’t a real OS. Unless I can play Doom.”

Antigravity 2.0 is a standalone desktop application.

Agent-first. Not editor-first.

The harness gained sub-agents, hooks, and asynchronous task management. Gemini 3.5 Flash was co-optimized with it.

The agents built an OS. The OS ran Doom.

That is the whole thesis in one image. Agents now produce non-trivial, working software systems end-to-end.

What ships with Antigravity 2.0 today:

Full CLI experience
Antigravity SDK
Native voice support with Gemini audio models
Integrations with Android, Firebase, Google AI Studio

Available globally today.

Why it matters: Cursor defined the category. Claude Code scaled it. Google made it a first-class platform with native voice, an SDK, and a working OS demo on stage. Developer mindshare is the prize and Google brought the proof.

3. Gemini 3.5 Flash runs 4x faster than every frontier competitor

Speed is the new intelligence.

Google shipped a frontier-grade model that runs four times faster than everything at the top.

Flash beats 3.1 Pro across nearly every benchmark. Coding gains are large. GDPval, which measures real-world economically valuable tasks, jumped sharply. Pro lands next month.

Speed is the unlock for the agentic era:

▫️ Agents run hundreds of inference calls per task. Speed compounds.

▫️ Background execution needs throughput, not just intelligence.

▫️ Generative UI in Search needs sub-second response to feel real.

Available across products and APIs today.

Output tokens per second is the new latency. The gap between Flash and the fastest frontier tier from OpenAI or Anthropic is the metric to watch this quarter. Whoever wins speed at frontier quality wins the agent layer.

For the distillation thesis behind this, see Demis Hassabis named his AGI year and Marc Andreessen on why the AI moat is not the model.

4. Gemini Omni generates a claymation explainer of protein folding from one sentence

Google says video generation just had its image-editing moment.

Omni combines Gemini reasoning with Veo, Nano Banana, and Genie. The result understands physics well enough to generate a claymation explainer of protein folding from a one-line prompt.

The real product is iterative editing. Drop in a selfie. Talk to it. The scene morphs. Style, camera angle, characters, environment. The performance holds.

For paid subscribers in the Gemini app today. What changes for creators: style transfer on existing footage without losing the performance, camera angle changes from a single clip, new characters added while everything else holds.

The cost of motion content collapsed again. A business that depended on the production gap between an idea and a video just lost that gap.

For where this leaves the content business, see Grok Imagine and the video leaderboard nobody saw coming and the SaaS defense playbook for the AI era.

5. Google rebuilt the search box for the first time in 25 years

And it codes custom interfaces per query.

The new box accepts text, images, files, and video together. AI Overviews and AI Mode collapse into one continuous experience. Context follows you down the rabbit hole.

Then it gets stranger. Generative UI lands in Search this summer. Ask how black holes affect spacetime, get a custom interactive visual built in real time. Follow up about binary black holes, get a brand-new visual built for the follow-up.

Agentic coding deployed at the scale of Search. Free, to everyone.

The ten blue links monetized the last 25 years of the internet. If the answer is a custom interactive experience built per query, the link economy enters a different conversation. Publishers should be planning for it now.

For the playbook on surviving the search shift, see the creator economy in 2025: the data nobody wants to admit and the AI GTM playbook: what is actually working in 2026.

6. Search agents now run the web for you 24/7

The apartment hunt runs itself.

Search no longer waits for you to ask. You brain-dump your criteria. The agent scans the web, social, forums, and the shopping graph, then pings you when something matches.

Apartment hunts. Sneaker drops. Job leads. Investment signals. Multiple agents run simultaneously on different threads of your life, alongside Spark. Rolling out this summer.

Push notifications were the attention primitive of the mobile era. Background agents are the attention primitive of the AI era. The surface that owns background agents owns the next decade of consumer attention.

For the agent-reliability discipline this demands, see stop blaming the model, fix the architecture and I built a second brain in 10 minutes with Granola + Claude.

7. The Universal Cart turns every Google surface into a storefront

Amazon just got a problem.

Google built one shopping cart that works across Search, Gemini, YouTube, and Gmail. It runs in the background the moment you add a product. Tracks deals. Watches price drops. Surfaces price history. Pings you on restocks. It gets smarter as the Gemini models improve.

Search and the Gemini app at launch. YouTube and Gmail this summer.

Amazon’s moat was logistics plus the buy box. Google is attacking the buy box with intelligence layered across attention surfaces it already owns. Merchant economics shift. Affiliate models shift. Track the next earnings cycle closely.

For the VC lens on these platform shifts, see the most valuable VC-backed startups in the world and Coatue’s 18-chart AI report.

8. Google AI glasses ship this fall

Google did a live AI glasses demo on stage. It worked.

The live demo ordered coffee through DoorDash. On stage. It worked.

The glasses launched walking navigation to a remembered location, suggested a coffee stop, opened DoorDash in the user’s pocket, clicked through every option screen automatically, and prepared the order for confirmation. Tip added by voice.

Two product lines arrive this fall: audio-only glasses with voice-in-ear assistance, and display glasses paired optionally with a watch for glanceable previews. Both pair with Android and iOS.

Apple Vision Pro went heavy. Meta went content-first with Ray-Ban. Google is going for the agent on your face and pricing it for the mainstream. The wearable category just got a third serious contender with the agent layer the other two are missing.

For the vertical-integration play behind hardware bets like this, see Elon Musk and the outer limit of vertical integration.

9. Docs Live Lets You Talk To A Blank Document (And Walk Away With A Draft.)

And walk away with a draft.

A user opened an empty doc. Said they had a career day talk tomorrow. Pulled a resume from Drive by voice, asked for funny analogies, pointed to an email for logistics, iterated on table formatting, and added a personal story in bold at the top. All by voice. In real time. Not sped up.

Docs Live for Pro and Ultra this summer. Gmail and Keep in the same window. Gemini for Mac with multimodal file context this summer.

Voice was a feature. Voice is now an interface. A team still typing prompts into chat boxes a year from now has lost a productivity tier.

For the voice-first writing stack, see your voice is the only AI moat that compounds and why ChatGPT and Claude keep disappointing you.

10. Sundar named 2030 as the AGI year

And pointed to drug discovery as the proof.

The closing was not a product. It was a position. “We were standing in the foothills of the singularity.”

The proof points: Isomorphic Labs is in preclinical stage on multiple programs across immune disorders and cancer. Alpha Earth Foundations is positioned as a digital twin of the planet for deforestation and food security. CodeMender automatically finds and fixes critical software vulnerabilities.

The thesis stack: AGI is the goal, stated plainly. Health is the highest-value application. Cybersecurity is the highest-risk surface. Science acceleration is the proof point.

A company that publicly bets on AGI prices, hires, ships, and partners differently than one that does not. The 4-year countdown started May 22, 2026.

For the parallel AGI timelines from the other labs, see Demis Hassabis named his AGI year, Dario Amodei and the long game of safe AI, and what Sam Altman and Greg Brockman finally said out loud.

The 5 principles to steal from I/O 2026

Google did not show up to compete on chat. It showed up to redefine the surface layer of computing. Agents, not apps. Generative interfaces, not links. Glasses, not phones. AGI as horizon, not slogan.

Founders

The agent layer is the new platform. Thin wrappers on someone else’s model got thinner today. Assume your user has Spark, an information agent, and a Universal Cart running in parallel. Audit your product for the agentic stack within 30 days.

▫️ 50 game-changing AI agent startup ideas for 2026

▫️ 70 startup ideas YC wants you to build

▫️ The SaaS defense playbook for the AI era

Investors

Watch three indicators: output tokens per second across frontier models, background-agent adoption inside Search and Gemini, and merchant economics on the Universal Cart. Re-underwrite any consumer AI thesis that assumes the chat-window paradigm.

▫️ What top-tier VCs actually look for in 2026

▫️ Where VC money is going in AI

▫️ The full investor lists archive

Operators

Voice is now a primary interface. Multimodal input is the default. Pilot a voice-first workflow on your highest-volume internal process this month.

▫️ The single best productivity decision you can make with Claude right now

▫️ I built a second brain in 10 minutes with Granola + Claude

The 5 principles

1️⃣ Agents replace apps as the unit of attention. Build, invest, and operate accordingly.

2️⃣ Speed is the new intelligence. A frontier model that runs 4x faster is a different product.

3️⃣ Interfaces will be generated, not designed. Custom UI per query rewrites the web.

4️⃣ Background work is the killer feature. The user closes the laptop. The agent keeps going.

5️⃣ AGI is now a public corporate strategy. Plan for the world where it ships.

The 4-year countdown started May 22, 2026. The companies that ship for 2030 win the decade. Everyone else is building for a world that ends in 48 months.

If this breakdown saved you 35 minutes, share it with one founder or investor who needs to see it.

The AI Corner

Discussion about this post

Ready for more?