Good morning, AI enthusiast.

Google made a significant impact at its I/O 2025 event, unveiling a suite of powerful AI advancements. Key among these are the general availability of its Gemini 2.5 models and the introduction of a new premium 'AI Ultra' subscription tier.

These updates demonstrate Google's strong commitment to embedding advanced AI across its ecosystem. With the 'AI Ultra' plan offering top-tier model access and early experimental tools, what new possibilities will this unlock for professionals and how will it shape the broader AI landscape?

In today’s AI recap:

• Google's Gemini 2.5 models and AI Ultra plan

• Google's Gemma 3n for on-device AI

• Klarna's AI boosts revenue per employee

Google's I/O AI Barrage: Gemini 2.5, AI Ultra, & More Future Tech

The Recap: Google I/O 2025 showcased a significant AI push, with Gemini 2.5 models achieving general availability and a new premium "AI Ultra" subscription tier being introduced. These updates signal Google's commitment to embedding powerful AI across its ecosystem and offering advanced capabilities to power users.

Unpacked:

• Gemini 2.5 Flash becomes the new default model, offering enhanced reasoning and coding, while the Pro version gains a "Deep Think" capability for tackling complex problems.

• The new Google AI Ultra plan, priced at $250/month, provides top-tier access to Google's most capable models, higher usage limits, and early access to experimental tools like Veo 3.

• Google is also experimenting with Gemini Diffusion, a novel approach to generate entire blocks of tokens simultaneously, promising faster and more accurate outputs for text and code.

Bottom line: Google's I/O announcements underscore a major strategic focus on deploying and monetizing its most advanced AI technologies. These developments provide professionals with more powerful tools for automation and productivity, while also indicating a future where cutting-edge AI becomes increasingly accessible through premium services.

Klarna's AI Drives $1M Revenue Per Employee

The Recap: Fintech leader Klarna reports its fourth profitable quarter, crediting AI for driving revenue per employee towards $1 million, according to its latest results. This highlights AI's significant impact on operational efficiencies and overall financial health.

Unpacked:

• Klarna's AI adoption has 96% of its employees using the technology, significantly boosting operational efficiency and contributing to a 152% increase in revenue per employee since Q1 2023.

• A key driver was AI in customer service, where Klarna initially planned to replace nearly 700 contractors with AI chatbots to reduce costs.

• However, adapting its strategy, Klarna recently reintroduced the option for customers to connect with human agents for support, balancing automation with customer needs.

Bottom line: Klarna's results showcase AI's tangible power to significantly enhance productivity and profitability within established financial services. This case offers a compelling blueprint for how businesses can leverage AI for substantial operational gains and improved customer interactions.

Google's Gemma 3n Takes AI Mobile

The Recap: Google just unveiled the preview of Gemma 3n, its latest open model engineered for powerful, efficient, and mobile-first AI experiences. This new architecture also lays the groundwork for the next generation of Gemini Nano, set to enhance on-device capabilities across Android and Chrome later this year.

Unpacked:

• Its innovative Per-Layer Embeddings (PLE) technology dramatically cuts RAM usage, enabling larger models (5B and 8B raw parameters) to operate with the memory footprint of significantly smaller ones (2GB and 3GB dynamic footprint), making them practical for direct use on your mobile devices. You can learn more from the official documentation.

• Gemma 3n expands multimodal understanding, processing audio, text, and images with enhanced video capabilities—all designed to be processed locally. This includes high-quality speech recognition and translation, enabling rich, interactive experiences that respect user privacy and function offline.

• The model boasts many-in-1 flexibility thanks to MatFormer training, which includes a nested, smaller submodel within the larger one. This design allows you to dynamically trade off performance and quality on the fly without needing to host separate models.

Bottom line: Gemma 3n empowers developers to build a new wave of intelligent, real-time applications directly on user devices, making sophisticated AI more personal and broadly accessible. This advancement pushes AI towards more private, offline-capable solutions that can seamlessly integrate into everyday mobile experiences.

The Shortlist

SAG-AFTRA filed an Unfair Labor Practice complaint against Epic Games' Llama Productions over AI-generated Darth Vader voice usage in Fortnite without union consent.

President Trump signed the Take It Down Act, establishing federal penalties for creating and distributing non-consensual explicit AI-generated deepfakes.

Builder.ai enters insolvency proceedings as the Microsoft-backed AI startup faces financial difficulties amid sales figure audits.

Elon Musk announces xAI and Tesla collaboration on Colossus 2, aiming to create the first gigawatt-scale AI training supercluster powered by Tesla Megapacks.

Hearst distributed an AI-generated summer guide through King Features syndication that contained fake book titles and non-existent experts, affecting multiple major newspapers.

Signing off,

David, Lucas, Mitchell — The Recap editorial team