Welcome back. Microsoft’s decision to open-source its agent security tools is a signal that the industry recognizes agents are moving faster than safeguards. Figma is taking a smarter path by putting agents inside designers’ existing workflows, rather than treating AI as another chat box. And Google’s latest Search overhaul shows how quickly AI is being pushed into products that billions already trust. That reach is Google’s advantage, but it will also create potential problems when AI-generated answers blur the line between useful information and unreliable responses. All in all, AI is clearly becoming much more embedded. —Jason Hiner

IN TODAY’S NEWSLETTER

1. Google remakes Search with AI once again

2. Why Microsoft released agent security for free

3. Figma’s AI agent strategy breaks from rivals

BIG TECH

Google Search's AI overhaul gets even bigger

For years, a Google search meant one thing: a long page of blue links. That is, until AI Overviews changed the game. Now, Google is shaking up Search again.

At Google I/O, the company unveiled what it is calling "the biggest upgrade to our Search box in 25 years," and it's meant to make the experience more intuitive for users by more natively infusing AI. For instance, the Search box now expands to give users more space to describe what they want and provides AI-powered suggestions.

Users can now also enter more into the Search box, including text, images, files, videos, or Chrome tabs. The new AI Search box is rolling out to users today in all countries and languages where AI Mode is available, Google said in its release. The handoff between Search and AI Mode was also improved, with it now being easier to ask a follow up in the AI overview and go straight to AI Mode. That's also live today across desktop, mobile, and online.

The rest of Search updates focused on the topic of the year: agents. The announcements include:

• Information agents: These agents operate in the background and reason across information, including the web, blogs, news sites, and social posts, to find you the information you need. It can send you updates as it finds more information, for example, surfacing more apartments that match your criteria. It is rolling out to Google AI Pro and Ultra subscribers this summer.

• Agentic booking capabilities: Now expanding in Search to new tasks such as local experiences and services.

• Agentic coding in Search: Google is integrating Antigravity and Gemini 3.5 Flash’s coding capabilities right into Search to build ideal responses for your queries, including visual tools and simulations such as tables or graphs. It will be available this summer.

Lastly, Personal Intelligence in AI Mode is expanding to more people in nearly 200 countries and territories across 98 languages. You can connect apps like Gmail and Google Photos, and soon Google Calendar, for more tailored responses. Google reminds users that they are always in control of their data and can choose whether or not they want to connect the apps.

Google is uniquely positioned to dominate the AI race, largely due to the sheer scale of users who already rely on its products daily, Search is, of course, the prime example. With billions of queries processed every day, Google has an unparalleled pipeline for exposing users to its AI products. The changes to Search make strategic sense on two fronts: it dramatically expands AI's reach, and it enhances a product that people already trust and depend on. The challenge, however, is that AI-generated results are inherently prone to hallucinations, a fundamental limitation of how these systems work. Embedding AI into a product as established and trusted as Google Search risks blurring the line between authoritative information and AI-generated results that can be unreliable.

TOGETHER WITH MODE MOBILE

Apple’s Starlink Update Sparks Huge Earning Opportunity

Most people brushed off Apple’s new Starlink integration for iPhones.

Mode Mobile saw something bigger: billions of new users suddenly within reach.

Mode’s EarnPhone already reaches 490M+ users that have earned over $1B, and that’s before global satellite coverage. With SpaceX eliminating "dead zones," Mode's earning technology can now reach billions more in unbanked and rural populations worldwide.

Their global expansion is perfectly timed, and investors like you still have a chance to invest in their pre-IPO offering at $0.50/share. But that window is fading fast as their share price is expected to move on 05/29.

With their recent 32,481% revenue growth and newly reserved Nasdaq ticker, Mode is edging closer to a potential IPO.

⏰ Secure your shares at $0.50/share before price changes on 05/29!

GOVERNANCE

Why Microsoft released agent security for free

Though agents are capable of handling increasingly complex tasks, greater agency brings more risk. 

To manage more operations, these agents require increased access and permissions to your data, whether that means your email, records, or code repositories. But the more access you give these machines, the deeper the consequences are when they behave in unintended ways. It’s why Microsoft’s AI red team has unveiled two agentic security tools to address this issue before it gets out of control. 

On Thursday, the company announced two new open-source security tools for developers: 

• RAMPART is an open source framework that embeds AI red-teaming techniques into the development workflow. This tool allows teams to continuously test agent behavior, both under common and adversarial circumstances.

• Clarity is a complementary tool to RAMPART, helping engineers and product teams reason through what they're building as they build it. This pressure-tests ideas and assumptions, and weeds out potential risks before the code is even written. 

In a blog post, the company pointed to three reasons that it’s investing in these tools: To help scale red-teaming practices industry-wide, make incidents reproducible and verify mitigation techniques, and allow engineers to “think through the “why,” before the “how” of software building.” 

“That shift from 'generate text' to 'do things in the world' changes the safety equation entirely, because an agent that can act can also potentially act in ways nobody intended,” Ram Shankar Siva Kumar, founder of Microsoft’s AI red team, said in a blog post. 

The goal of these tools is to treat security as an ongoing consideration in building and deploying software, rather than something that can be configured or assessed once and left behind. 

It’s a critical concern given that AI has the potential to create an incredibly dangerous cybersecurity landscape. Along with allowing engineers to ship faster than ever, potentially opening the door for vulnerable, vibe-coded apps and products to hit the market, increasingly powerful models have also put cyberattack capabilities into the hands of practically anyone with the motivation to do harm.

Microsoft is joining a growing chorus of tech giants calling out security concerns around AI, with OpenAI and Anthropic each announcing their own cybersecurity measures as they race to bring more powerful models to market. With these two tools, however, Microsoft has taken it to another level: by open-sourcing them, the company’s red team is enabling other security professionals and developers to build on its work. This feeds a stronger security ecosystem and could enable defense techniques and safeguards to spread through the industry faster at a time when software development is moving at such a rapid (and potentially dangerous) clip. And, of course, this benefits Microsoft, too. Given that it’s not particularly standing out in the AI industry for models themselves, feeding the security market allows it to differentiate itself in the AI ecosystem.  

TOGETHER 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

PRODUCTS

Figma’s AI agent strategy breaks from rivals

Amid a plethora of AI-powered design tools, Figma is embedding agents where designers will actually use them. 

On Wednesday, the company unveiled an AI agent built specifically for product design, embedded into the Figma canvas and connected to user context. Rather than just embedding a simple chat interface, the company said this agentic canvas is a “natural extension” of the way design teams already work. 

In a press release, the company said the agent was built to understand Figma workflows, such as components, design system logic and brand standards.  

• Figma’s agent can help designers generate and edit design layers, generate new directions with “infinite canvas,” and get real-time feedback. The agent is a natural language tool, aimed to “[lower] the barrier to entry for anyone who wants to design and build,” the company said in its release. 

• The agent calls on fine-tuned AI models trained to understand product design, patterns and Figma-specific use cases. Teams can also work with multiple agents in parallel, and navigate between AI-assisted design and manual manipulation without switching tools. 

• Because the agent is connected with the existing team context, product designers can automate busywork such as component setup, layout fixes, and bulk edits. 

“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,” said Loredana Crisan, Figma’s Chief Design Officer, in a statement. 

The agent is now available in beta and will gradually roll out to Figma Design paid users in the coming weeks, eventually expanding to other Figma products. It’s the latest in Figma’s broader AI transition, which includes the natural language “prompt-to-prototype” tool Figma Make, advancements to its MCP server, and the ability to connect third-party coding agents.

Though design is part of Figma’s DNA, it’s not the only company that’s paying attention: Anthropic launched Claude Design in mid-April, allowing users to collaborate with Claude to create “polished visual work.” Meanwhile, both OpenAI and Google collaborate with the popular design tool Canva. As AI firms have long fought over the hearts and minds of software developers with AI coding tools, design may become the next battlefield for AI expansion in business. However, given that design is inherently a creative field, deploying AI in this market could raise the same ethical qualms that arise when AI is used in fields like filmmaking and music. While coding was a natural starting place for AI adoption, it remains to be seen whether design will have the same uptake.

LINKS

• OpenAI could go public as soon as September, reports Wall Street Journal

• Intuit to reportedly cut 17% of its staff to streamline operations with AI

• Nvidia reports quarterly revenue up 85% year-over-year at $82 billion

• OpenAI lobbyist seeks state AI regulation for de facto national standard

• Trump Administration executive order demands access to advance models

• GitHub confirms breach impacting roughly 3,800 internal code repositories

• Stability Audio 3.0: Stability released a family of audio models capable of generating songs up to three minutes long. 

• Fastlane: An AI marketing engine that can automate the process of shortform content generation.

• Granola Briefs: An agent that searches your emails, the web and previous meeting notes to provide users with concise, need-to-know bullets. 

• Perplexity Computer: The computer use agent now connects to Snowflake, allowing users to run work against their data warehouse.

• Nvidia: PhD Research Intern, Security and Privacy - Fall 2026

• Salesforce: Adversarial AI & Research Engineer

• DoorDash: AI Research Fellowship, (Summer and Fall 2026)

• Zoom: Research Scientist - AI Incubation

GAMES

POLL RESULTS

Have you, or would you, wear a pair of AI glasses?

Yes, I already do (7%)
Yes, I’d consider it (58%)
I’m not sure (23%)
No, I would not (10%)
Other (2%)