Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• Study: AI shrinking entry-level jobs

• Google antitrust breakup?

• Microsoft's new home-grown AI models

• Google Nano Banana

Let’s get right into it!

Sign Up | Advertise | Tools

CAREERS

Study: AI shrinking entry-level jobs

A new Stanford study analyzing millions of ADP payroll records suggests that generative AI is disproportionately hurting entry-level job prospects in the U.S., particularly for workers aged 22–25 in AI-exposed fields like customer service, accounting, and software development, where employment has dropped 13% since 2022. By contrast, more experienced workers in the same fields and younger workers in less-exposed jobs, such as health aides or production supervisors, have seen stable or rising employment.

The researchers argue that AI is more likely to replace “book-learning” knowledge gained through formal education rather than experience-based skills, helping explain stagnant job growth among young workers despite overall labor market resilience. While not peer-reviewed, the study highlights uneven effects: AI adoption appears to reduce jobs in some sectors while complementing and improving efficiency in others.

Link to 57 page study PDF

LEGAL

Google antitrust breakup?

A year after a Federal District Court ruled that Google illegally maintained a monopoly over online search, the court is now set to decide on remedies that could reshape both Google’s business and the future of AI competition. Options include forcing Google to sell Chrome or Android, requiring it to share valuable click-and-query data, or otherwise restricting its ability to pay companies like Apple to favor its products.

Even before the ruling, regulatory scrutiny has already curbed Google’s ability to use massive payouts to secure AI deals, opening space for rivals like OpenAI and Perplexity. The outcome could significantly affect how new AI competitors emerge, much as past antitrust actions against AT&T and Microsoft fueled new waves of innovation. Ultimately, the remedy’s impact will be measured by whether it enables smaller, more innovative companies to challenge Google’s dominance in the AI-driven information ecosystem.

BIG TECH

Microsoft's new home-grown AI models

Microsoft, long reliant on OpenAI’s models to power AI features in products like Bing and Windows 11, is now working to reduce that dependence by publicly testing its own in-house foundation model, MAI-1-preview. The model, trained on about 15,000 Nvidia H100 GPUs and supported by a new GB200 cluster, is being evaluated on LMArena, where it currently ranks 13th for text workloads behind offerings from Anthropic, DeepSeek, Google, Mistral, OpenAI, and xAI. Microsoft plans to roll MAI-1-preview into Copilot text use cases in the coming weeks, with early access available to developers, marking its first end-to-end homegrown model.

While the company remains a major backer of OpenAI—investing over $13 billion and providing critical cloud infrastructure—it has also listed OpenAI as a competitor alongside Amazon, Apple, Google, and Meta, as the two increasingly vie for market share. OpenAI, meanwhile, has diversified to other providers like CoreWeave, Google, and Oracle to support ChatGPT, now used weekly by 700 million people. The effort is being led by Mustafa Suleyman, CEO of Microsoft’s AI unit and former co-founder of DeepMind and Inflection, who has brought in colleagues from both startups as Microsoft expands its AI talent base.

BIG TECH

Google Nano Banana

Google DeepMind’s new AI image editing model, nicknamed “nano banana” (officially Gemini 2.5 Flash Image), has topped the LMArena leaderboard and is now rolling out in the Gemini app. Unlike earlier models, nano banana offers unrivaled consistency by remembering details across edits, ensuring subjects retain their appearance even through multiple changes.

This enables more reliable transformations, such as altering a person’s style or attire, or merging multiple images—like combining separate photos of a person and a dog into a realistic cuddle shot. The model’s improvements open the door to more creative and practical uses, while all outputs include both visible AI watermarks and invisible SynthID markers. Google plans to expand access soon through the Gemini API, AI Studio, and Vertex AI.