PRODUCTS

Gemini 3.5 and Spark get Google in the agent race

The biggest name missing from the AI agent race so far in 2026 had been Google. Not anymore. 

On Tuesday at Google I/O, the search giant announced the new Gemini 3.5 family of models, which specialize in agentic capabilities. Rolling out now is Gemini 3.5 Flash, Google's quickest model, which it claims is four times faster than other frontier models while rivaling them in intelligence. It's available starting today in both the Gemini app and AI Mode in Google Search.

While Google is only releasing Flash today, the emphasis is on cost savings as well as speed. Google believes people will find that Gemini 3.5 Flash can process workloads in a fraction of the time and often at less than half the cost of other frontier models. 

Google also said Gemini 3.5 Pro will roll out in June to handle more complex and challenging workloads. However, Gemini 3.5 Flash is already optimized for long-horizon agentic tasks.

Coding tools and personal AI agents have been the dominant trends in AI this year, with OpenClaw, Claude Code, and OpenAI's Codex capturing the industry's imagination. 

Google also jumped into the personal AI agent race on Tuesday with the launch of Gemini Spark, running on Gemini 3.5 Flash. It rolls out to a private group of testers today and will come to Google AI Ultra subscribers in the US in beta starting next week.

The company called Gemini 3.5 Flash "our strongest agentic and coding model yet, outperforming Gemini 3.1 Pro" on major benchmarks, including Terminal-Bench, GDPval-AA, MCP Atlas, and CharXiv Reasoning. 

Other related announcements included: 

• Gemini Omni: Building on the success of Nano Banana, Google is rolling out Omni as its multimodal content-creation solution, starting with video. Omni allows you to "combine images, audio, video and text as input and generate high-quality videos grounded in Gemini's real-world knowledge."

• Gemini app redesign: Google overhauled the Gemini experience using a new design language called Neural Expressive that "features fluid animations, vibrant colors, new typography and haptic feedback."

• Android Halo: Provides at-a-glance visibility into what your Gemini Spark agent is working on at any given time, and allows you to communicate with your agent when needed. 

• Desktop app for macOS: Gemini's installed Mac app will add Gemini Spark, allowing it to run on your local machine and match the capabilities of OpenAI's Codex, Anthropic's Claude Cowork, and Perplexity's Personal Computer. 

Google I/O 2026 will go down as a major catch-up moment for Google as it makes up ground on the two hottest AI trends of the year: agents and coding. Of course, that will only be the case if Gemini Spark and the long-horizon coding capabilities deliver the same kind of value that AI enthusiasts are getting from rival tools right now. From a frontier-model standpoint, it's smart for Google to try to outpace its rivals in speed and token costs. It still has the extensive compute resources to play the long game and win.

IN PARTNERSHIP WITH LAMBDA

Most AI teams treat compute as a commodity. It’s not.

Compute looks interchangeable on a spec sheet. At scale, the facility, the cluster design, and the engineering behind it determine whether you ship on schedule or burn budget.

The teams pulling ahead know that compute infrastructure shapes what you can build, how fast, and at what margin. Lambda was built by researchers who learned this firsthand.

Find out why.

CONSUMER

Google's AI glasses look ready to take the lead

AI smart glasses have gained traction for their ability to give AI a first-person view of the world, but Google's entry into the space could push them much further into the mainstream.

Google gave the public its first look at the finished product of its AI-enabled smart glasses on Tuesday at Google I/O, the company's annual developer conference. The glasses, unveiled in partnership with Samsung and eyewear brands Gentle Monster and Warby Parker, come in two styles, both intended to bridge users' smartphones with their physical surroundings.

The multi-company effort is meant to combine Google’s leading AI technology with Samsung’s hardware expertise, along with the stylish designs of Gentle Monster and Warby Parker, which target different audiences: the former more bold and daring, the latter more classic. 

Beyond the aesthetics, the glasses pack powerful capabilities, including: 

• Navigation: Users can ask Gemini for directions by voice and see turn-by-turn directions in the in-lens display. It can also be used for recommendations on nearby places to go and to place orders. 

• Notifications: As an extension of your phone, users can get summarized notifications for texts and more. 

• Real-time translation: The audio will match the speaker’s voice in another language, and it can translate text within the user's line of sight. 

• Take photos: True to the "hands-free and heads-up" philosophy, users can also perform everyday tasks, such as snapping photos, using their voice. 

Ultimately, all of these features have Gemini at the core. The focus of the smartglasses category is to facilitate how people access AI assistance in everyday situations, which the smartglasses form factor is ideal for because it gives AI a window into what you see and interact with. 

“This intelligent eyewear marks an important step in Samsung’s vision for AI,” said Jay Kim, Head of Customer Experience at Samsung, in the post. “With this new AI form factor, we are further expanding the Galaxy device ecosystem, where each device is optimized to deliver unique AI experiences that best fit each form.” 

Beyond an expected fall launch, full product details have yet to be revealed, with additional information to arrive in the coming months.

These glasses have a competitive advantage over other products on the market because of the power of the Android ecosystem and deep phone integration. The Meta Ray-Bans, which have dominated the category thus far, have been mostly valued for their function as a hands-free content-capturing tool. The in-lens displays on Google’s devices make a world of difference, offering users an experience much more akin to glancing at their phones. The biggest challenges will be comfort, which, based on my try-ons of the prototypes, the company has achieved, and then, of course, addressing people’s privacy concerns.

IN PARTNERSHIP WITH ORACLE NETSUITE

Turn KPI Insights Into Smarter Decisions and Measurable Business Growth

In this field guide, Bernie Smith, Director of Made to Measure KPIs and KPI Academy, shares how to turn KPI insights into real business action.

Learn how to build trusted metrics, improve decisions, align teams, and drive measurable performance.

Download the Field Guide

HARDWARE

OpenAI targets enterprise with guaranteed compute

OpenAI wants to cut itself a bigger slice of AI’s so-called five-layer cake.

On Tuesday, the company unveiled OpenAI Guaranteed Capacity, an offering that guarantees customers “long-term access” to its compute. Customers can choose commitments of 1 to 3 years, with discounts based on the duration. 

Guaranteed Capacity offers certainty of access to compute-based “spend levels,” OpenAI said in its announcement. Customers can also spend from this commitment across OpenAI’s products. This can also be used across some supported cloud providers and model plans. 

The goal of Guaranteed Compute is to allow businesses to run critical workflows without concerns about uptime or predictability, and to offer the compute companies need to expand their products or adoption plans. 

CEO Sam Altman said in a post on X that this will be available until OpenAI sells out of its allocated compute for this program, leaving enough capacity remaining for ChatGPT, Codex and its other products. Altman also noted that OpenAI intends to offer this again in the future, aiming to build “as much compute as fast as we can.” 

“Customers are increasingly asking us for certainty on capacity,” Altman said on X. “As models get better, we expect that the world will be capacity-constrained for some time.” 

It isn’t the first sign that OpenAI wants to be known for more than its models. Guaranteed Capacity follows the company's unveiling of Multipath Reliable Connection, or MRC, a networking protocol designed to make GPU clusters faster and more efficient by mitigating network congestion and failures. Taken together, OpenAI may be trying to become a foundational part of the AI compute stack at a time when the shortage of compute is holding back practically every lab in the industry.

OpenAI isn’t just trying to dominate a larger piece of the industry with this new offering. Instead, it may be trying to prove itself. While consumers, startups and small- to medium-sized enterprises rely on OpenAI’s products, large, regulated industries face a far higher barrier to entry. With Guaranteed Compute, OpenAI is offering customers more peace of mind that they will have the uptime, capacity and performance they need to embed AI into mission-critical infrastructure and regulated environments. It’s the same reason the company partnered with 1Password for secure authentication of Codex, and is feeding the open models ecosystem, as The Deep View’s Jason Hiner pointed out: To give customers access to powerful models that fit within their regulatory and security requirements. In short, with a wealth of opportunities for major contracts and important work, OpenAI wants to be trusted by the big kids. 

LINKS

• Andrej Karpathy joins Anthropic to work on research and development

• Anthropic acquires Stainless, a startup that automates developer tools

• Investors don’t trust AI to handle their money, survey finds

• Sports Illustrated deleted articles by author after AI plagiarism accusation

• Spotify adds verification badges to ban AI-generated podcast impersonators

• OpenAI adds C2PA standard for detecting AI-generated images

• Meta lays off 8,000 employees amid its AI transition

• Claude Design: Anthropic doubled the token limits on its AI-powered design tool. 

• Claude Managed Agents: The company launched self-hosted sandboxes in public beta and MCP tunnels in research preview. 

• Omnia: The AI visibility tool just added step-by-step insights that allow you to improve your brand’s visibility in AI engines. 

• Index: Parallel Web Systems unveiled a platform for content owners and creators to understand how AI uses their work.

• Microsoft: Research Intern - Applied Speech Research

• Meta: AI Research Scientist - Meta Superintelligence Labs

• OpenAI: Researcher, Misalignment Research

• Cartesia: Researcher, Models

GAMES

A QUICK POLL BEFORE YOU GO