2026 · 05 · 20 · WED
15 STORIES · UPDATED 23:53 SGT
▸ TODAY · 2026·05·20 WED
THE BIG THING TODAY

Google unveils Gemini 3.5 at I/O 2026 — frontier reasoning, fully agentic Search, and a bet that chatbots are already old news.

GOOGLE AI BLOG · 22H AGO · 13 MIN READ →

I/O 2026: Welcome to the agentic Gemini era

Editor’s note: Below is an edited transcript of Google CEO Sundar Pichai’s remarks at Google I/O 2026, adapted to include more of what was announced on stage. See all the announcements in our collection.

It’s been an extraordinary year since our last I/O, a period of relentless shipping, technology advances and hyper progress. We’re now in the part of the AI cycle where people want to see the value in the products they use every day. We’ve been really focused on that, and you’ll see that in the products and features we’re announcing today at I/O.

Ten years since we pivoted the company to be AI-first, we still see AI as the most profound way to advance our mission and improve people’s lives at scale. That’s why we’ve been taking a differentiated, full-stack approach to AI innovation, from our custom silicon and secure foundation, to our world-class research and models, to our products and platforms that touch billions of people. This approach enables us to iterate and innovate faster in ways that are lighting up every part of the company.

What’s incredible is how people are using AI, whether it’s students prepping for final exams with the Gemini app, musicians and artists using generative AI models like Lyria and Veo as part of their creative flow, or developers coding and bringing their ideas to life.

AI momentum across the full stack

These stories of how people are using AI are the best measure of progress. To understand the scale at which people are adopting AI, there is another great proxy — tokens, the fundamental units of data our models process, many representing a problem being solved.

Two years ago, we were processing 9.7 trillion tokens a month across our surfaces — a huge number. Last year at I/O, that grew to roughly 480 trillion tokens. Fast forward to today, that number jumped 7x to over 3.2 quadrillion per month.

It tells an important story about our products and how others are building as well — especially developers and enterprises:

- Over 8.5 million developers are now building new apps and experiences with our models monthly.

- Our model APIs are now processing roughly 19 billion tokens per minute.

- Over the past 12 months, over 375 Google Cloud customers each processed more than one trillion tokens, representing incredible demand for AI from across industries.

Momentum with our products

Today we have 13 products with over a billion users each. Five of those have more than 3 billion users.

Our Gemini models are a big reason more people are using our products, and why they're using our products more.

It all starts with Search, which is bringing the benefits of generative AI to more people than any other product in the world. AI Overviews now has over 2.5 billion monthly active users. And AI Mode has been a revelation, our biggest upgrade to Search ever. People love it, and in just a year, it’s already surpassed 1 billion monthly active users.

When people use our AI-powered features in Search, they use Search more. Search has become less about individual queries and feels more like an ongoing conversation, giving you deeper insights and connecting you with the vastness of the web.

Another place where we’ve been rapidly innovating is in the Gemini app. Last year at I/O, the Gemini app had 400 million monthly active users. Today, we’ve surpassed 900 million, more than doubling in a year. In that same time, daily requests have grown over seven times.

We’ve been adding a lot of unique features like Personal Intelligence, which make responses more customized and helpful. And to date more than 50 billion images have been generated with our Nano Banana image generation models. It was a breakout star this past year, showing how much latent creativity there is in the world.

Natural, conversational AI in products

There’s also a lot of latent productivity to be unlocked. Over the last year, we’ve been bringing the ability to have more natural conversations with Gemini directly inside our products. Recently, Maps got its biggest upgrade in a decade, including a new feature called Ask Maps. People are using Ask Maps for more complex, and much longer questions.

Now we’re bringing more natural conversational AI to more products.

Ask YouTube

People come to YouTube everyday to ask a lot of questions. There’s a lot of great videos, but sometimes it’s hard to know where to start.

Ask YouTube entirely reimagines the experience, making information much more digestible and easy to navigate. You’ll see videos that best match your interest, and most importantly, it jumps right to the part of the video most relevant to you.

We’re starting to test Ask YouTube now, and it will roll out broadly in the U.S. this summer.

Voice-powered Docs Live

There are a lot of times I want to get things done at the speed of my voice. That is much more possible today thanks to technical leaps in our audio models.

A new feature called Docs Live takes this to another level. To create a doc with Gemini before, you had to type out a precise prompt. With Docs Live, you can just verbally “brain dump” whatever is on your mind, and let Gemini do the rest. Here’s a demo in real-time:

In the future, you’ll be able to create new docs and edit them directly, all with your voice. Docs Live is rolling out for subscribers this summer, and powerful voice capabilities will come to Gmail and Keep then too.

Infrastructure supporting innovation at scale

It’s incredible to see the pace of innovation rolling out across our products. Supporting all of this scale for our users, while also serving enterprises and developers around the world, requires massive investments in infrastructure. We’ve been investing for now and for the future. In 2022, we were spending $31 billion annually in capex. This year, we expect that number to be about six times that, approximately $180 to $190 billion. A key part of this investment is our custom silicon.

A decade ago, we announced our very first commercial tensor processing unit, or TPU, on the I/O stage. Since then, we have transformed how the industry builds for AI. We recently announced our 8th generation of TPUs at Cloud Next. For the first time, we’ve taken a dual chip approach with specialized architectures for training and inference: TPU 8t and 8i.

- TPU 8t is optimized for large-scale pretraining, and it’s nearly three times the raw computing power of our previous generation. We’ve taken a fundamentally different approach with our training infrastructure. With JAX and Pathways, our training is no longer constrained by the limits of a single, massive data center. Instead, we can now seamlessly distribute training across multiple sites, scaling training across more than 1 million TPUs globally. This gives us the ability to create the largest training cluster in the world. For model builders, this means training larger, more capable models in weeks rather than months.

- TPU 8i is designed for inference. We have dramatically improved speed at every step. Because if we learned anything in 27 years of working on Search, it's that latency matters.

In addition to speed, we’re also thinking about scaling sustainably. Both chips are more energy efficient, delivering up to two times better performance-per-watt.

Gemini Omni

This progress with TPUs is how we can make compute advances across models, coding and agents. With world models, AI is moving from predicting text to simulating reality. We have been working to push the boundaries of what these models can do.

Gemini Omni is our new model that is capable of generating samples in any output modality from any input. We’re starting with video outputs, and over time we’ll enable image and text. This new model combines Gemini’s intelligence with our generative media models — a huge leap forward in world understanding. We’re launching the first model in the Omni family: Gemini Omni Flash.

Gemini Omni Flash is available starting today. You will be able to try it on the Gemini app, Google Flow and on YouTube Shorts. We'll also be rolling it out to developers and enterprise customers via APIs in the coming weeks.

New SynthID updates and partners

As generative AI gets better, so does the need for greater transparency. Research shows people can correctly identify high-quality deepfake videos only about a quarter of the time. Three years ago, we launched SynthID, our watermark that is invisible to the naked eye. Since launch, SynthID has now watermarked over one hundred billion images and videos, along with sixty thousand years of audio assets.

Millions of people are using our SynthID detector in the Gemini app to verify AI-generated content. And now we’re going a step further and adding Content Credentials verification across products. This will show you if the origin of the content was AI or a camera, and if it’s been edited with generative AI tools. We want more people to have easy access to these tools, so we’re expanding both Content Credentials and SynthID verification to Search and Chrome.

Of course, this only works at scale if more partners decide to watermark their own AI-generated content. Nvidia signed on to SynthID last year. And today, we are thrilled to announce that OpenAI, Kakao and Eleven Labs are adopting SynthID, too. It’s great to see the cross-industry collaboration. We’re looking forward to expanding to more partners and setting the standard of transparency for the AI era.

Gemini 3.5 Flash

Gemini 3 launched a few months ago, with a full family of models. It’s our most adopted series yet. We've loved seeing developers use Flash as their daily driver, and build incredible experiences with Pro's deep reasoning and multimodal capabilities. We’ve been hard at work on improving these models, especially focused on agentic coding, long-horizon tasks and real-world workflows.

Today, we’re introducing Gemini 3.5 Flash, our first in a series of models combining frontier intelligence with action. Two things I’d highlight:

- When compared to 3.1 Pro, 3.5 Flash is better across almost all benchmarks. It’s made huge progress in coding — and look at the extraordinary jump in GDPVal. This captures many real-world economically valuable tasks.

- Gemini 3.5 Flash is a very capable model, at the frontier and comparable to the best models, but it’s still very fast. Which is why when you look at the intelligence versus output speed, it’s in a league of its own in the top right quadrant. When looking at output tokens per second, it is four times faster than other frontier models.

The new model has been a game changer for us internally at Google. We’ve been using 3.5 Flash with a reimagined version of our agent-first development platform Antigravity, and it’s dramatically accelerated how we build. In March we were processing half a trillion tokens a day internally across our AI developer tools, and we’ve been doubling every few weeks. Now, we’re processing more than three trillion tokens a day. This scale created a powerful feedback loop helping us improve 3.5.

What’s amazing about Flash is how it delivers frontier-level capabilities at less than half the price of comparable frontier models. We’ve heard that many companies are already blowing through their annual token budgets, and it’s only May. If companies used a mix of Flash and other frontier models they could save a lot of money. To put this in perspective, top companies are processing about 1 trillion tokens a day. If they shifted 80% of their workloads from other frontier models to 3.5 Flash, they’d save over $1 billion dollars annually. That is real savings they can pour back into their company.

Gemini 3.5 Flash is available for everyone today across our products and APIs. We’re also excited for Gemini 3.5 Pro. We are using it internally, it’s showing great improvements, and it will be coming next month.

Antigravity 2.0

We’re also bringing 3.5 Flash to developers in Antigravity.

Antigravity is expanding beyond the coding environment, turning it into a platform to develop and manage cohorts of autonomous AI agents. This includes Antigravity 2.0, a new standalone desktop application that acts as a central home for agent interaction, where anyone can orchestrate agents for all sorts of tasks. And we developed an even more optimized version of Flash: not just 4x but 12x faster than other frontier models.

Users in Antigravity can get a taste of this experience starting today. Read more about Antigravity 2.0 here.

Gemini Spark is your 24/7 agent

Gemini 3.5 and Antigravity are unlocking a new world of agents and agentic capabilities. We’ve been bringing agents to developers and enterprises for a while. Now we are super focused on bringing the power of agents, safely and securely, to consumers so that it works for everyone. You’ll see agentic experiences across many of our products today.

I’m particularly excited for Gemini Spark, your personal AI agent in Gemini app that helps you navigate your digital life, taking action on your behalf and under your direction.

- It runs on dedicated virtual machines on Google Cloud. And it’s 24/7 so you don’t need to keep your laptop open.

- It’s powered by Gemini 3.5 and the Google Antigravity harness, which allows it to perform long-horizon tasks easily in the background.

- Spark will integrate seamlessly with tools, starting with our own, and in the coming weeks with third-party tools through MCP.

- And you can work with Spark however is most convenient: in the Gemini app or soon, through email and chat.

- On Android, you will be able to view live updates and task progress of agents like Spark through a new UI space called Android Halo, coming later this year. Later this summer, Spark will operate directly within Chrome, acting as your agentic browser across the web.

We’re starting to roll out Gemini Spark to trusted testers this week and the Beta is coming to Google AI Ultra subscribers in the U.S. next week.

Search in the agentic era

Gemini Spark is the first experience made possible by 3.5 models and Antigravity. This combination gives us new ways to accelerate our mission and transform our products to be radically more helpful.

As we enter this agentic era, Search will be more helpful and powerful than ever. Today, we’re introducing information agents in Search. These are personalized AI agents you can set up to work in the background, 24/7, to find what you need at exactly the right moment, and help you take action. Information agents are rolling out this summer starting with Google AI Pro and Ultra subscribers.

Another way we’re building a truly agentic Search is by infusing it with agentic coding capabilities. With the power of Gemini 3.5 Flash and Google Antigravity, Search will build custom experiences just for your individual questions, like dynamic layouts and interactive visuals. These generative UI capabilities will be available for everyone in Search this summer, free of charge.

And for longer running tasks that you need to keep coming back to, Search can go a step further — building persistent, custom dashboards or trackers that you can return to and make progress on. You can think of these like mini apps for your own specific tasks. You’ll be able to build custom experiences with Antigravity, right in Search in the coming months, starting first for Google AI Pro and Ultra subscribers in the U.S.

More from our agentic Gemini era

Here’s a look what else we shared at I/O:

- Daily Brief is another out-of-the-box agent coming to the Gemini app. It gives you a personalized digest and synthesizes information from your inbox, calendar and tasks to find the most important things to be aware of. And it’s not just summarizing data: it’s prioritizing, organizing and suggesting the next steps, so it’s easy for you to take action. All in this super concise morning digest that’s built for skimming.

- Google Flow is rolling out a new agent today to everyone that can plan and reason through complex tasks with your inputs, under your control. Built with Gemini models, it brings expertise and a deep understanding of your project to help with early brainstorming, creating and editing. You can also vibe code any creative tool, right in Flow — like tools for designing video effects, hand-drawn animations or layering text.

- Google Pics is our new AI image creation and editing tool, built on our latest Nano Banana model, that helps you create just about anything with the creative controls you want. Whether you’re building a design from a blank canvas or editing an existing photo, Pics treats every element as an individual object rather than a flat, static image. This allows you to create, swap or perfect specific details, so you can bring your exact vision to life. Google Pics is available now to trusted testers and will be rolling out later this summer to Google AI Pro and Ultra subscribers in Workspace.

- We also shared more about our intelligent eyewear, which we first gave a glimpse of last year, including audio glasses that offer spoken help in your ear and display glasses that show you the information you need, right when you need it. Both let you stay hands-free and heads up, with help from Gemini just by asking. Audio glasses are launching first, coming later this fall.

- Gemini for Science brings together a number of AI tools to help accelerate scientific research. Building on the deep reasoning and research capabilities of Gemini as well as Deep Think and Deep Research, it includes new experiments on Labs as well as Science Skills to connect agentic platforms like Google Antigravity to over 30 major life science databases and tools. Users can express interest to try Gemini for Science experiments on Google Labs, and Science Skills is available today on Github and directly in Antigravity.

As we look across the full stack of innovation, from the infrastructure behind TPU 8i to the frontier capabilities of Gemini 3.5 and Antigravity, it’s clear we’re firmly in our agentic Gemini era. I’m excited to see how it will unlock new ways to accelerate our mission and transform our products to be radically more helpful, for everyone everywhere.

See everything we announced here.

Read on blog.google →

models

Gemini 3.5 Flash arrives tuned for agents — Google is trading raw chatbot speed for tool-use horsepower.

TECHCRUNCH AI · 22H AGO · 3 MIN →

Google launched on Tuesday Gemini 3.5 Flash, a new AI model that the company says is its strongest yet for coding and autonomous AI agents. The model, which was introduced at the company’s annual Google I/O developer conference, can independently execute coding pipelines, manage research projects, and, in internal tests, build an operating system entirely from scratch.

The release signals Google’s shift from pitching AI as a conversational tool to AI as an agentic tool. It’s not just answering questions, but planning, building, and iterating on real work with minimal human input.

Koray Kavukcuoglu, DeepMind’s chief technologist, told reporters on Monday ahead of the public launch: “3.5 Flash offers an incredible combination of quality and low latency. It outperforms our latest frontier model, 3.1 Pro, on nearly all the benchmarks,” including coding, agentic tasks, and multimodal reasoning.

He added that it is 4x faster than other frontier models, a speed that’s ideal for coding and agentic tasks, but that Google has “taken it to another level” by developing an optimized version of Flash that’s 12x faster with the same quality.

That speed is central to Flash’s design for agentic work, where multiple AI agents run at the same time on long-running tasks, according to Kavukcuoglu. Onstage at I/O, Google engineer Varun Mohan, demonstrated agents spawning off to work on separate components before coming together to build a full operating system inside Antigravity, the company’s agentic development platform and IDE.

Kavukcuoglu said Flash 3.5 was co-developed with Antigravity so that agents could have a “native environment where they can live, work, and execute.”

At I/O, Google released Antigravity 2.0, a stand-alone desktop application designed around agent-first development.

The gains are showing up beyond demos. Google says 3.5 Flash’s agentic capabilities are already creating impact among partners, like banks and fintechs automating multi-week workflows, or data science teams finding insights in complex data environments.

The model can run autonomously for multiple hours, though Tulsee Doshi, Google’s senior director and head of product, said it will at times pause and ask for user input when it hits a decision point or permission issue that requires human judgment.

When Google releases its forthcoming 3.5 Pro model, the two are designed to work in tandem.

Doshi told TechCrunch that “3.5 Pro becomes your orchestrator, your planner, and then it actually can leverage Flash to be the various sub-agents. I think it really comes down to where do you really want that reasoning power, where you actually want that larger model that can really push on the reasoning side versus where do you have tasks that really do merit good brute force tool use capabilities?”

Now, 3.5 Flash is the default model in the Gemini app and in AI Mode in Search globally. At I/O, Google also announced agentic capabilities coming to Search, letting users create, customize, and manage AI agents directly on the platform. The new model will also power Gemini Spark, Google’s new personal AI agent designed to run 24/7 to help consumers manage their digital life.

Providing that level of AI capability for average consumers comes with scrutiny. Google is currently facing a lawsuit after a man nearly committed a mass casualty event and died by suicide following weeks of chatting with Gemini last year.

The implications for harm only grow when making powerful autonomous agents available more broadly. Google says Gemini 3.5 has strengthened cyber and CBRN (chemical, biological, radiological, and nuclear) safeguards and is better calibrated to engage with sensitive questions rather than refuse them outright.

Gemini 3.5 Flash is available generally today via Antigravity, the Gemini API, and Gemini Enterprise, as well as through the Gemini app and AI mode in Search.

Read on techcrunch.com →

NEW Stability AI ships a new audio model that generates coherent six-minute songs from a single text prompt.

TECHCRUNCH AI · 53M AGO · 2 MIN →

Stability AI, the company behind Stable Diffusion, is releasing a new family of audio models, called Stability Audio 3.0. The top model can generate professional-grade music of more than six minutes long, the company claimed.

The company is releasing four new models under the Stable Audio 3.0 name: small SFX (459M parameters), small (459M parameters), medium (1.4B parameters), and large (2.7B parameters). The duo of small models is suitable for on-device sound and music generation of up to two minutes.

Both medium and large models can create full compositions of 6 minutes 20 seconds long that can maintain musical structure and melodic tone. This is more than double the length of what Stable Audio 2.0, released in 2024, was capable of generating.

Stability AI is making small SFX, small, and medium models available with open weights for anyone to use and modify. In 2024, the company released Stable Audio Open, which allowed for music generation of up to 47 seconds. The new family of models is a big step up from the previous open versions.

The large model is available only through the API and self-hosting paid services. Plus, companies with more than $1 million in revenue would need to get an enterprise license.

Many companies, including Google and ElevenLabs, are releasing models and tooling around music generation. However, as Suno and Udio’s ongoing court battles have proved, licensing of data and partnerships with music labels could become a key part of the long-term survival of these services.

Last year, Stability AI inked deals with Warner Music Group and Universal Music Group to develop models and music creation tools. The company said that its latest set of audio models is built on fully licensed data.

The AI startup is developing a new suite of products for professional musicians, but didn’t give more details on its features. Ethan Kaplan, former chief digital officer at Universal Audio and Fender, is joining the company to lead Stability’s professional music offering.

A number of AI companies are trying to bolster their credentials by hiring music execs. Earlier this year, Suno hired former Merlin CEO Jeremy Sirota as chief commercial officer. ElevenLabs has also hired Derek Cournoyer from indie music publisher Kobalt as a strategy lead for its music business.

Read on techcrunch.com →

Allen AI releases OlmoEarth v1.1, a smaller and more efficient family of open Earth observation models.

HUGGING FACE BLOG · 21H AGO · 3 MIN →

OlmoEarth v1.1: A more efficient family of Earth observation models

We released OlmoEarth (v1) in November 2025. Since then, partners have applied it across a wide range of tasks, from tracking mangrove change to classifying drivers of forest loss to producing country-scale crop-type maps in days, scaling deployments to national, continental, and global areas. Every release moves us closer to our mission: bringing state-of-the-art AI to organizations and communities working to protect people and our planet.

When OlmoEarth processes satellite imagery to make predictions across tens to hundreds of thousands of square kilometers, efficiency shapes what’s possible. Over the full lifecycle of running OlmoEarth – data export, preprocessing, inference, and post-processing – compute is by far the highest cost. A more efficient model means we can support more partners on the OlmoEarth Platform, and that anyone running OlmoEarth on their own can leverage this technology faster and at lower expense.

That’s why we built OlmoEarth v1.1: a new family of models that cuts compute costs by up to 3x while maintaining OlmoEarth v1's performance on a mix of research benchmarks and tasks we’ve constructed with partners.

Increasing efficiency by decreasing sequence lengths

The OlmoEarth models are transformer-based models, one of the dominant architectures in machine learning today. To process remote sensing data, we first convert it into a sequence of tokens the model can ingest.

Two important levers control efficiency in transformer-based models: model size (this is why we release a family of models, so users can pick the size that fits their compute budget) and token sequence length. Compute costs scale quadratically with the token sequence length, so even small reductions can meaningfully cut the cost of running the model.

MACs, or multiply-accumulate operations, estimate the computation needed for one model forward pass; lower MACs generally mean cheaper, faster inference. The y-axis is inverted because lower average rank is better. Labels show model family and size. All plotted points use the pasted MAC/rank values.

Designing the token

This raises an important question for transformer-based remote sensing models: what should a token represent?

Take Sentinel-2 imagery, a common modality we process. A Sentinel-2 input will be some tensor with a height and width (H, W representing the latitudinal and longitudinal pixels), a temporal dimension T, and 12 Sentinel-2 channels ([H, W, T, D=12]).

Currently, we split the data into resolution-based patches. Concretely, this means that we will pick some spatial patch size p, and split our overall Sentinel-2 image into patches of size p x p:

For each patch, we create a token per timestep per resolution. So a Sentinel-2 input with 2 timesteps yields 6 tokens per patch (2 timesteps x 3 resolutions, 10m, 20m, and 60m).

In total, a[H, W, T, D=12] Sentinel-2 input will yield H/p x W/p x T x 3 tokens.

Using a unique token per resolution is a common technique when processing Sentinel-2 data—Galileo and SatMAE both take this approach, and SatMAE shows significantly better results when doing it. However, it is not universal: CROMA is a model that only uses a single token for all bands, regardless of resolution. Because token counts compound multiplicatively, collapsing resolutions into a single token produces three times fewer tokens and material savings across pretraining, fine-tuning, and inference.

Naively combining the tokens in this way leads to significant performance drops, including a 10 ppt drop on m-eurosat kNN (a common benchmark task for remote sensing models). We hypothesize that separating Sentinel-2 bands into different tokens makes it easier for OlmoEarth to model important cross-band relationships.

Merging tokens without impacting performance required us to modify our pre-training regimen. We describe those changes in detail in our paper.

For developers

The result is a model family that does more with less. At every size, OlmoEarth v1.1 runs up to three times cheaper than OlmoEarth v1, making frequent, planet-scale map refreshes more affordable for every team running OlmoEarth. If you're using a model from the original OlmoEarth family, try OlmoEarth v1.1. It provides similar performance to OlmoEarth v1 while requiring one third of the compute, though we have seen some regressions (see our technical report for more details). If it works for your task, you should see a significant speedup during fine-tuning and inference.

For researchers

Pretrained remote sensing models have many degrees of freedom, which makes them hard to study. When performance shifts, is it the architecture, the dataset, or the pre-training algorithm?

We train OlmoEarth v1.1 on the same dataset as OlmoEarth v1, so any differences between the two isolate the effect of methodological changes. We hope this advances understanding of scientific principles when pretraining models for remote sensing.

Get started

Check out the OlmoEarth v1.1 weights and training code, including the weights for our Base, Tiny, and Nano models.

Read on huggingface.co →

Google's Genie world model can now simulate real streets using Street View data as the seed.

TECHCRUNCH AI · 22H AGO · 4 MIN →

We’ve all pulled up Street View on Google Maps to show a friend what our childhood home looked like, or dropped that little person icon onto the streets of Paris to see if we booked a hotel in a cool neighborhood. Imagine being able to do that, but in a more immersive, interactive way that allows you to really simulate the street and its environs, and even do things like adjust the weather or see what it would look like in a “Day After Tomorrow” scenario.

That’s one of the goals of Google’s latest integration. Starting today, Google DeepMind is connecting Street View to Project Genie, the company’s general-purpose world model that can generate diverse, interactive environments. The new feature launched during the Google I/O 2026 developer conference.

“It’s really powerful for both the agent [and robotics] use case and for humans to play with, and that’s always been the thesis of Genie,” Jack Parker-Holder, a research scientist on DeepMind’s open-endedness team, told TechCrunch.

He gave the example of a new robot being deployed in London, which rarely sees the sun. Genie could, Parker-Holder says, simulate those scarce occasions when the sun glints off the Victorian housing, so the rays don’t shock the robot when it happens.

“Simultaneously, you might say, ‘I’m going to New York City, but not this time of year,’” he continued. “‘It’s going to be snowy. I want to see what that block looks like in the snow.’”

Google has been collecting Street View data for 20 years via cars with cameras and individuals strapped with “tracker backpacks.” The tech giant has collected north of 280 billion images across 110 countries and seven continents.

“With Street View, we have imagery from a large quantity of the world,” Jack said. “You can imagine how potentially powerful it is to combine this rich source of real-world information and data with an ability to simulate worlds.”

Google released its latest world model Genie 3 for research preview last August and opened up access to the tool to Google AI Ultra subscribers in the U.S. in January, allowing customers to create interactive game worlds from text prompts or images. The goal is to use Genie for educational experiences, gaming, and robotics training.

Genie 3 is already helping to power one of Waymo’s simulators to train its self-driving cars on “exceedingly rare events” like tornadoes or casual elephant encounters. Adding Street View data to that could help Waymo prepare to launch in more cities around the globe.

Waymo has its own simulator that it relied on to scale to 11 U.S. cities and test its AI driver in several more. The difference with Genie, says Parker-Holder, is that those are all from the car’s point of view. Street View allows for not only simulating a world anchored to a real place, but also shifting the point of view to other types of agents, like a human or a robot.

Google is launching Street View in Genie to some Ultra users in the United States starting today, with access rolling out at scale over time. Global Ultra users will gain access over the next few weeks, per the company.

The researchers’ goal is to put this new capability into as many hands as possible, per Diego Rivas, a product manager at DeepMind. He cautioned that Street View in particular and Genie in general is still an experiment, so there’s much to improve upon in terms of accuracy.

In the samples the Google team showed me — including an underwater simulation of a neighborhood I used to live in — the results are impressive and recognizable, but still video game quality rather than photorealistic. The models are also not yet physics-aware, meaning they don’t yet understand cause and effect. For example, in a simulation of a woman running through a snowy Joshua Tree, she ran right through cacti and bushes.

Compare that to, say, Google’s image generator Nano Banana — which can now generate perfect text in infographics — or its video generator Veo — which understands that paper boats drift on water currents, smoke disperses into the air, and fabric drapes over forms.

Physics isn’t hard-coded into these models; they learn it intuitively over time through passive observation, as a living being would.

“I think for this kind of model, it’s maybe six to 12 months behind video in terms of the accuracy and quality, so I think it’s something we will solve,” Parker-Holder said.

Jonathan Herbert, director of Google Maps who started on the Street View team as an intern 12 years ago, said that Genie can’t yet create a faithful reconstruction of a street. He thinks the real breakthrough is the AI’s spatial continuity. If you turn 360 degrees, the AI correctly remembers and simulates the environment behind you. From that point on, the model can build a new environment on top of that.

“We have long thought about how we can build out the best and richest model of the world on top of Street View data,” Herbert said. “It’s definitely been an idea of ours to use Maps Data in new ways and for new kinds of AI research for a pretty long time.”

Catch up on the rest of Google IO 2026’s big news

Google Search as you know it is over

Google updates Gemini app to take on ChatGPT and Claude

Google introduces Gemini Spark, a 24/7 agent assistant with Gmail integration

Read on techcrunch.com →

research

Demis Hassabis tells press he sees 'foothills of the singularity' as Gemini hits new reasoning marks.

THE VERGE AI · 17H AGO · 1 MIN →

Welcome to a “profound moment for humanity,” according to Google DeepMind CEO Demis Hassabis, who closed out Google I/O’s keynote presentation on Tuesday, saying:

Demis Hassabis said this might be the ‘foothills of the singularity.’ What?

After a couple of hours of apps and itineraries, Google’s big AI presentation turned philosophical.

After a couple of hours of apps and itineraries, Google’s big AI presentation turned philosophical.

Google’s cutting-edge research and products will help unlock AGI’s incredible potential for the benefit of the entire world. When we look back at this time, I think we will realize that we were standing in the foothills of the singularity.

It will be a profound moment for humanity. This technology will be a force multiplier for human ingenuity and usher in a new golden age of scientific discovery and progress, improving the lives of everyone, everywhere. We look forward to building the future with all of you. Thank you, and enjoy the rest of Google I/O.

Just before announcing we’ve possibly arrived at “the foothills of the singularity,” Hassabis introduced Gemini for Science, a set of tools and experiments in Google Labs and Google Antigravity intended for helping with scientific research. According to Hassabis, with tools like these, Google hopes to “reimagine drug discovery with the goal of one day solving all disease.” Tech executives often discuss AI this way, like Microsoft CEO Satya Nadella referring to AI as “cognitive amplifier tools” and Luma AI CEO Amit Jain’s claim last year that AI is the key to saving Hollywood.

In an interview with Bloomberg just a few months ago, Hassabis said that “the singularity is another word for a full AGI arriving,” which is not the same definition we’re used to from the likes of Vernor Vinge and Ray Kurzweil.

At the time, even by his measurement, he claimed that “we’re nowhere near that.” When asked about his timeline for achieving AGI, Hassabis said he was standing by his prediction that we have a “50 percent chance of getting there by 2030.”

Read on theverge.com →

Two AI science assistants succeed at drug-retargeting tasks in independent trials, hinting at real lab utility.

ARS TECHNICA AI · 20H AGO · 1 MIN →
ARS TECHNICA AI

Both tools generate hypotheses; one goes on to analyze some of the data.

Read on arstechnica.com →

Demis Hassabis pushes back on AI-driven layoffs: companies shedding talent now will regret it inside two years.

WIRED AI · 22H AGO · 3 MIN →

Demis Hassabis, the CEO of Google DeepMind, is keen to talk about the coding skills of his company’s newest model, Gemini 3.5 Flash. The model has been trained to perform complex agentic coding tasks: translate large code bases from one language to another; find and fix bugs lurking deep in knotty code; and even write entire operating systems from scratch.

Hassabis does not, however, think this spells doom for software developers. “I have no idea why people are going around talking with certainty about that,” Hassabis tells WIRED ahead of the new model reveal at today’s Google’s I/O event.

“Perhaps there is an ulterior motive for putting those messages out; raising money or whatever,” Hassabis says. “From my point of view, from DeepMind and Google's point of view, if engineers are becoming three or four times more productive, then we just [want to] do three or four times more stuff.”

The striking coding abilities of the latest models has led to widespread fear that AI may be on the brink of eliminating programming roles and other white-collar jobs. Executives at some AI companies have predicted widespread job displacement, while some prominent tech companies, including Amazon, Salesforce, and Block, have blamed recent layoffs on the use of AI.

Hassabis thinks that Alphabet, which oversees several companies besides Google, may be well positioned to take advantage of a revolution in software productivity. “I have a million ideas, from lab drug discovery to game design,” he says. “I'd love to have some free engineers to go and do those kinds of things.”

Hassabis says that companies looking to replace developers with AI may be making a big mistake. “I think it's a lack of imagination—and a lack of understanding of what's really going to happen,” he says.

Google revealed a raft of AI stuff at its annual developer event. Through a coding tool called Antigravity, Gemini 3.5 Flash offers frontier coding and reasoning capabilities but is faster and cheaper than the offerings of its competitors, Google says. Gemini 3.5 Pro, a more powerful new version of its flagship model, will debut next month.

The company needs to catch up when it comes to AI coding, which has emerged as a crucial and lucrative application for the latest AI models. Anthropic and OpenAI lead developer adoption with their respective tools, Claude and Codex, according to a 2025 Stack Overflow survey.

The company also demoed an agentic assistant called Spark that lives in Google’s Cloud and has access to its apps. The design is meant to be safer than something like OpenClaw because it has limited access to personal data, Google says.

Other agentic demos included a version of Android with an AI agent built in and a refreshed version of Google Search that uses agentic coding to generate a site or app on the fly in response to a search query.

AI coding has captivated the AI world in recent months, even inspiring hope that models could one day rewrite their own code in a self-improvement loop. Hassabis says it’s possible but doubts that it will immediately lead to superhuman-level AI.

Progress in other areas of science might require AI models to have a deeper understanding of the physical world and even an ability to perform experiments within it, he says.

Even within the seemingly solved world of coding, Hassabis says it is notable that AI has yet to produce a blockbuster app or video game without human help. “I think there's something missing,” he says.

Read on wired.com →

industry

Musk v Altman trial wraps — evidence shows Musk's own early OpenAI plans mirrored the structure he later sued over.

TECHCRUNCH AI · 19H AGO · 4 MIN →

The jury’s speedy decision to reject Elon Musk’s lawsuit against the other founders of OpenAI and Microsoft confirmed what we saw in the courtroom: Musk’s case was a weak one, in part because he waited so long to file it.

Watching the closing arguments last week, OpenAI’s attorneys detailed point by point how the law was on their client’s side, while the plaintiff’s team focused on Sam Altman’s apparent lack of credibility and expressed disbelief that anyone would disagree with Musk’s accusations.

The final effect was that, after the verdict, some found it hard to believe Musk had lost — including the man himself. In a post he later deleted, Musk called Judge Yvonne Gonzalez Rogers a “terrible activist Oakland judge,” then announced his plans to appeal, declaring, “There is no question to anyone following the case in detail that Altman & Brockman did in fact enrich themselves by stealing a charity.”

But Altman and Brockman weren’t the only figures who benefited from OpenAI’s non-profit investments. As much as Musk and his legal team tried to make the trial about Altman, the proceedings revealed just as much about Musk.

One incident that came out in court showed Musk benefiting from OpenAI in an uncomfortably familiar way. Greg Brockman testified that in 2017, Musk asked him to bring a team of OpenAI researchers down to Tesla’s headquarters to help with the autopilot team for a few weeks. “It was pretty clear that was not something we could say no to,” Brockman said.

Brockman described taking a team of leading scientists, including Andrej Karpathy, Ilya Sutskever, and Scott Gray, to consult with the “demoralized” Tesla workers. They helped come up with ideas to improve the vehicle’s self-driving technology, with Sutskever telling the team that if they could find 10,000 images of a tricky corner case, they would be able to fix their software. Musk even asked Brockman to recommend employees to fire, which he declined to do.

Another person familiar with the episode confirmed Brockman’s account and said Tesla did not reimburse OpenAI for the time and effort of its employees. Musk’s family office, Excession, didn’t reply to a request for comment.

The heart of Musk’s case is that Altman, Brockman, and OpenAI committed a “breach of charitable trust” — that Musk donated funds for a specific charitable purpose, and his co-founders instead used them for something else. He also accused them of “unjust enrichment” due to stock and other benefits from OpenAI’s for-profit.

In the case of the OpenAI scientists parachuting into Tesla, Musk’s charitable donations were intended to hire scientists focused on securing the benefits of AGI. Instead, he had them work for free at his for-profit company.

Dorothy Lund, a Columbia Law School professor and the co-host of the Beyond Unprecedented podcast, told TechCrunch that this arrangement wouldn’t be legal, calling it “a bit rich for Musk to be suing for breach of a charitable trust, when he appears to have been redirecting assets in a way that was inconsistent with that mission.”

It’s true that the self-driving work involved artificial intelligence, but witnesses for Musk emphasized that Tesla’s self-driving project was very different from OpenAI’s research agenda. That’s in part because Karpathy left OpenAI for Tesla shortly after this incident. OpenAI’s attorneys portrayed the departure as Musk violating his duty to the lab, where he was co-chair of the board, by recruiting one of its key researchers to his own company.

The other fact that no doubt influenced the jury was the amount of time Musk spent trying to gain sole control of a potential OpenAI for-profit affiliate in 2017. Musk deployed good cop, bad cop tactics in an attempt to convince his co-founders to let him have total control of OpenAI’s for-profit affiliate — giving them free Teslas and threatening to withhold his donations.

His efforts put his attorneys in a tricky spot, facing a need to convince the jury there was a significant difference between what Musk envisioned and the for-profit that was ultimately created. They suggested a “small adjunct” for-profit would be permissible, though OpenAI’s witnesses showed non-profits with large commercial arms are common.

Indeed, there’s a very plausible counter-factual where Musk took one of the offers his co-founders made to split their equity more evenly, and finds himself today as one of OpenAI’s largest shareholders — just not the controlling one. But several times during the trial, Musk’s associates testified that he refuses to invest in any business he could have sole control over.

The failure of Musk’s claims because he filed them too late has been cited as a technicality, but the statute of limitations has substance behind it: People and businesses make important decisions and spend resources based on their understanding that what they are doing is permissible. If someone like Musk waits too long to sue, then the cost of unraveling all those decisions can outweigh a just reimbursement.

No members of the jury have spoken about how they arrived at their verdict. However, they were asked to consider if, before August 5, 2021, Musk should have known that OpenAI was spending resources outside its mission or launching a for-profit affiliate. The answer to that is clear: Musk himself was doing those things.

Read on techcrunch.com →

Meta employees scramble to burn through benefits as a fresh layoff wave is widely expected to land this week.

WIRED AI · 20H AGO · 2 MIN →

Ahead of Meta’s latest round of mass layoffs tomorrow, some employees are deserting offices, abandoning their work, and loading up on perks they might soon lose, several people at the company tell WIRED.

Two employees describe a widespread rush to use up an annual $2,000 flexible benefit, which can cover a variety of expenses including health and wellness activities. A separate triennial credit of $200 toward the purchase of audio gear has led to a scramble to purchase Apple AirPods and other headphones. Another source says Meta offices have been largely empty this week, as people prioritize polishing their résumés and gather offsite to commiserate with friends for what may be their final time as colleagues. Employees are variously “paralyzed,” “coasting,” and “panicked,” sources say.

Meta plans to lay off about 10 percent of its nearly 80,000 employees on Wednesday, with notices going out to affected workers’ personal and corporate email addresses at 4 am Singapore, London, or San Francisco time depending on their location, according to a company-wide memo sent on Monday. The cuts are coming at a time when the social media giant behind Instagram, WhatsApp, and Facebook is enjoying record-high profits.

But CEO Mark Zuckerberg insists that the company must free up cash to invest in AI data centers, and that Meta can perform just as well with fewer employees because of AI technologies that augment human labor.

Meta didn’t immediately respond to a request for comment for this story. The company has undergone three previous large rounds of layoffs since 2022, including as part of Meta’s one-time “year of efficiency” drive in 2023. But even though the latest round is smaller than a couple of those, it is drawing widespread scrutiny because it comes at a time of societal anxiety about AI’s impact on jobs.

Inside Meta, the imminent cuts are among several concerns that have sunk morale to unprecedented depths, according to 16 current and former employees who recently spoke to WIRED. Employees also have been frustrated by being “drafted” onto a new AI team without any choice and the rollout of surveillance software that tracks US workers’ laptop use to train AI models.

Meta also plans to internally restructure as it conducts sweeping layoffs, transferring 7,000 remaining staff to “AI initiatives” and converting more managers into individual contributors. That would bring the total number of those affected—either laid off or placed in a new role—to 20 percent of the current workforce, Reuters reported on Monday. WIRED independently confirmed this reporting. Some parts of the company have been told they won’t be affected at all.

But in recent days, employees who are bracing for changes have shared checklists internally about benefits to take advantage of, and are saving documents such as performance reviews and pay stubs, according to one worker. Some teams are meeting up at bars and restaurants near Meta offices in New York and Menlo Park on Tuesday and Wednesday to eat and drink away their sorrows, several employees said. Management has encouraged employees not to come into offices on Wednesday.

Update, May 19, 11:40 PM EDT: WIRED corrected the time zones when layoff notices will be emailed.

Read on wired.com →

OpenAI launches Education for Countries in Singapore, its first government partnership in Asia-Pacific.

OPENAI BLOG · 19H AGO · 3 MIN →

Introducing OpenAI for Singapore

Today at the ATx Summit in Singapore, we’re launching OpenAI for Singapore, a partnership with the Ministry of Digital Development and Information (MDDI) to support Singapore’s National AI strategy.

Singapore is a global leader in AI and has made it central to its plans for economic growth, public services, and how people learn, work, and build. It recognised early that AI is becoming core infrastructure for economies and societies, with the potential to drive greater productivity, creativity, scientific progress, and economic opportunity.

As intelligence becomes more like a utility, OpenAI for Singapore is designed to support Singapore’s ambition to become an AI-powered economy. Backed by a commitment of more than S$300 million, the initiative will focus on three key areas:

- Helping organisations in Singapore deploy frontier AI and solve some of their hardest problems

- Developing the next generation of AI talent locally

- Helping more people and businesses across Singapore benefit from AI

“We’re excited to partner with Singapore as it builds on its position as a global leader in AI. Singapore has strong technical talent, trusted institutions, and a clear ambition to use AI to drive long-term growth and improve people’s lives. Through OpenAI for Singapore, we want to help more organisations benefit from frontier AI, support the next generation of local AI talent, and widen access to these tools across the country.”

At the centre of the partnership is the establishment of our Applied AI Lab in Singapore, OpenAI’s first outside the United States. What this means in practice is that OpenAI will create more than 200 Singapore-based technical roles over the next few years and make Singapore one of its global hubs for Forward-Deployed Engineers.

Forward-Deployed Engineers sit at the point where frontier research meets real-world deployment. They work directly with companies on some of their hardest problems and unlock new sources of value.

Through the partnership, the Lab will support work aligned with Singapore’s AI Mission priorities, particularly in areas such as public service, finance, healthcare and digital infrastructure. As this work grows and our Singapore-based team expands, we also expect to increase our office footprint in the country over time.

“With AI reshaping economies, businesses and the workforce, Singapore's response has been deliberate: growing new sectors, anchoring global frontier companies here, and equipping our people with the skills to thrive in this new environment. This partnership with OpenAI reflects the Government’s commitment to developing Singapore's AI capabilities, strengthening enterprise adoption of AI, and securing good jobs for Singaporeans.”

“We are delighted by OpenAI’s decision to expand its applied AI engineering capabilities in Singapore through the launch of OpenAI for Singapore. This investment presents exciting opportunities for Singaporeans and underscores Singapore’s growing role as a trusted global hub for AI innovation and development in the region. We look forward to partnering with OpenAI to support AI adoption across Singapore’s economy through these capabilities.”

Through OpenAI for Singapore, we will work with government and local ecosystem partners to build the skills Singapore will need for a future shaped by AI. This includes:

- Working with the Ministry of Education and GovTech on AI-enabled learning use cases and tools, including more interactive support for Mother Tongue language learning

- Supporting educators through a Singapore chapter of the OpenAI Academy and Codex for Teachers hackathons

- Launching a Forward-Deployed Engineer training programme to help develop local AI deployment talent

- Participating in the National AI Impact Programme to deepen AI capabilities across the technology workforce, including through the use of Codex

Taken together, these efforts are intended to help build the skills Singapore will need for the next phase of AI adoption.

AI’s benefits should reach every layer of the economy, not only the biggest enterprises or the people building the technology.

Through OpenAI for Singapore, we will work with local partners to help more people and businesses across the economy benefit from AI. This includes exploring accelerator programmes for AI-native startups and collaborating on workshops for micro-entrepreneurs and small businesses.

These efforts will focus on practical adoption, helping founders build with AI and SMEs improve operations and customer service.

OpenAI for Singapore is about people as much as technology. By investing in frontier deployment, next-generation AI talent, and access across the economy, we want to be a long-term partner as Singapore builds toward an AI-ready economy in ways that are useful, responsible, and broadly beneficial.

Read on openai.com →

NEW AI search startups are seeing valuations spike as Google's full pivot to agentic search validates the category.

TECHCRUNCH AI · 48M AGO · 1 MIN →

Yesterday’s big news was Google’s plan to blow up its traditional Search in favor of an AI-powered experience — but Google isn’t the only company planning for the next generation of discoverability.

This morning, Bloomberg has news of the Andreessen-backed Exa Labs, which has raised $250 million against a $2.5 billion valuation to go after the same market. And it’s part of a wave of startups all chasing AI search, which has quietly become one of the most attractive targets in consumer AI.

From Bloomberg:

Exa is part of a wave of startups that are vying to transform the search industry, including Tavily, TinyFish and Parallel Web Systems. Led by former Twitter Chief Executive Officer Parag Agrawal, Parallel recently raised $100 million at a $2 billion valuation in a round led by venture firm Sequoia Capital, according to the Wall Street Journal.

At the same time, we’re also seeing conventional tech platforms like Amazon, LinkedIn and Reddit looking to AI to revamp their search and discoverability features — so there will be plenty of potential acquirers if any of the startups start looking to sell.

The biggest competitor is ChatGPT, which still owns the interface layer and, prior to the Google launch, was handling the vast majority of the AI-powered searches taking place on a given day. But OpenAI can’t make Search a priority and Google has an ad business to protect, which could leave room for a smaller lab like Exa or Parallel to carve out a niche for itself.

Read on techcrunch.com →

tools

Google AI Studio now ships full Android apps from a single prompt — code, build, and signed APK in under five minutes.

TECHCRUNCH AI · 22H AGO · 3 MIN →

The AI coding boom is now coming directly for Android app development. On Tuesday at Google IO 2026, the company announced new native Android app creation capabilities in its web-based Google AI Studio, shrinking a process that takes weeks of setup and coding down to minutes.

The company also said that consumers will be able to use Gemini AI to find the apps they need, both on the Play Store and the web, expanding opportunities for developers to have their apps discovered.

Google says the new capabilities could make sense for anyone from a seasoned developer looking to prototype a new app quickly to a first-time creator.

By offering the ability to essentially vibe-code Android apps via web-based tools, Google is ramping up the competition with other AI-powered development tools, like Cursor, Replit, Lovable, Claude Code, and others, while also opening up Android development to a new type of user: a non-technical creator. The news also represents an expansion of Google’s earlier addition of AI-powered coding with Gemini in its desktop version of Android Studio.

The apps are built with the Kotlin programming language using Google’s Jetpack Compose toolkit and with support integration with hardware sensors like GPS, Bluetooth, and NFC, the company says. However, the resulting creations, for now, are only meant to be used personally, as publishing for family and friends is still on the roadmap.

The company suggests the technology could be used for the creation of personal utilities and simple social apps, hardware-enabled experiences, or AI-powered experiences.

For now, would-be app developers can use the embedded Android Emulator directly in a web browser to preview and interact with the app as it’s being built. Users can then install the app on their Android phone over a USB cable connected to their computer, using the integrated Android Debug Bridge (adb).

For those looking to take their project further, AI Studio can automatically create the app record, package the bundle, and upload it to an internal testing track in Google Play Console for developers. This allows users to continue to iterate on their app while updating on their devices along the way.

Those who want to take the next steps to publishing the app more publicly can hand off this version of the project to Android Studio by downloading a zip file and exporting it directly to GitHub. In time, Google plans to allow creators to publish their apps for use by family and friends and will add support for Firebase integrations (Firestore, Firebase Auth, Firebase App Check, and other tooling).

In doing so, the company is imagining an Android app ecosystem where users find apps from among their own network of friends, not just the Play Store.

However, for the latter, Google is infusing AI into the experience here, too.

A new “Ask Play” AI-powered overlay will allow users to discover new apps by having natural conversations with AI within the Play Store.

Perhaps more importantly, apps will begin to be surfaced with users’ conversations with Google’s Gemini virtual assistant, exposing developers’ apps to millions of users. This will roll out in the weeks ahead across Gemini on the web and on Android. Later this year, Gemini will also surface over 450,000 movies and TV shows, plus where to livestream sports, which can directly link users from their queries to a developer’s Android app with the content in question.

While Google previewed a number of Android-related announcements last week, it held back on sharing the news of the native Android app development until Tuesday’s start of its annual developer conference, Google I/O. That suggests the company believes this is bigger news and more closely tied to its idea of putting AI to real-world use, as was the larger theme of this year’s event, where AI was spread across Google products, from workspace productivity apps to AI tools, search, mobile apps, and more.

Catch up on the rest of Google IO 2026’s big news

Google Search as you know it is over

Google updates Gemini app to take on ChatGPT and Claude

Google introduces Gemini Spark, a 24/7 agent assistant with Gmail integration

Read on techcrunch.com →

Google launches Antigravity 2.0 with a new desktop app and a CLI for shell-native agentic coding.

THE VERGE AI · 22H AGO · 2 MIN →

Google is announcing a major upgrade to one of its vibe coding platforms: Beginning today, you can now use AI Studio to build native Android apps.

Google can now vibe-code you an Android app

But Google is carefully suggesting that the apps might be best as more limited experiences.

But Google is carefully suggesting that the apps might be best as more limited experiences.

With Google AI Studio, you can prompt your idea for an app and preview it with an embedded emulator of Android. When you want to try it out on an actual phone, you can connect an Android device to your computer and install it. In the future, you’ll also be able to invite app testers from AI Studio.

It sounds like you won’t be able to whip up just any old app of your dreams, though; in a blog post, Google says that this “initial release” of the functionality is focused on “personal utility” apps like habit trackers and study quizzes, “hardware-enabled experiences” like apps that use your phone’s camera or GPS, and “AI-powered experiences” that rely on Gemini’s API.

And while this new feature might make it easier to create an app, if you want to publish your app on Google Play, it will need to meet Google’s rules. “App quality continues to be a top priority to Google Play and we will not be changing any of our review processes and standards,” spokesperson Mia Carter tells The Verge. “AI Studio simply lowers the barrier to entry for creating high quality Android apps. Apps created with AI Studio will still need to meet these strict quality and review standards on Google Play.”

At I/O, Google is making other developer- and app discovery-focused announcements as well. The company is launching a 1.0 version of its command-line interface for building Android apps. Google is going to start showing apps as recommendations to Gemini queries in the “coming weeks” and will surface movies and TV shows beginning “later this year.” And in Google Play, Google is rolling out a short-form video feed of “Play Shorts” about apps to users in the US and “select developers.”

Read on theverge.com →

Figma adds an AI assistant to its collaborative canvas, letting designers spawn variants and components conversationally.

TECHCRUNCH AI · 2H AGO · 1 MIN →

Over the last few months, Figma has struck partnerships with OpenAI and Anthropic to bake in support for AI CLI tools like Claude Code and Codex to allow users to use these coding environments alongside its design software. The company is now baking in its own take on AI smarts via a new AI agent that operates within its collaborative canvas.

Figma says users can employ natural language text prompts to direct its new AI agent to generate new designs, edit existing ones, or automate tasks such as generating iterations of existing designs. Users can even fire up multiple agents that can do various tasks simultaneously.

The company claims the AI assistant understands design contexts and elements since it runs on AI models that are fine-tuned for design use.

“As building software gets easier, what matters most is setting direction: deciding what to work on, how it should function, what the experience should feel like. Teams can now collaborate with agents on the multiplayer canvas to test out ideas, visualize edge cases, and refine concepts together without over-indexing on the more tedious parts,” Figma’s chief design officer, Loredana Crisan, said in a statement.

The agent is first launching in Figma Design, and the company plans to eventually make it available in its other products. Figma said that, over time, it wants to bring design and code even closer together within its apps.

Facing intense competition from the likes of Canva, Adobe, Flora, Krea and Dessn, last year Figma acquired node-based design tool Weavy, and has added new image editing features to its products.

The company has done well despite fears of AI eating into the work of designers and the demand for software they use: In the first quarter of 2026, Figma reported revenue of $333.4 million, 46% more than a year earlier.

Read on techcrunch.com →