AI Tidbits – From Shopping Agents to Scheming Models: This Week in AI

AI Tidbits – From Shopping Agents to Scheming Models: This Week in AI

Good morning! The AI news cycle hasn’t slowed one bit, and this week brings a mix of breakthroughs, warnings, and bold corporate moves. Let’s dive straight in 👇

 

Google – The Big Big e-comm Step

Google has just unveiled the Agent Payments Protocol, an open framework that enables AI agents to make secure purchases on your behalf.

With backing from over 60 financial and tech giants, this is a milestone moment: AI agents as autonomous shoppers. Secure rails, trusted accountability, and industry-wide adoption are the missing pieces for mainstream acceptance—and Google may have just put them in place.

 

Amazon’s AI Update

If you or your customers sell on Amazon, this is huge.

Amazon is introducing a new AI chatbot for sellers that can create end-to-end ads: from catchy taglines to scripts, music, and even storyboards. One prompt could soon power your whole creative pipeline across Amazon’s ecosystem.

 

How AI Models Are Being Used

Both OpenAI and Anthropic published fresh insights into AI adoption. The data paints an interesting split between platforms, use cases, and geographies:

• Claude users lean on coding and technical tasks.
• ChatGPT users focus more on writing, advice, and decision support.
• Personal use is booming: 73% of ChatGPT conversations are now non-work, up from 53% last year.
• Emerging markets lead adoption: Low/middle-income countries are growing 4x faster in ChatGPT usage, while Claude remains wealthier-market heavy.
• Across both, more users are delegating tasks fully, treating AI as a doer, not just an assistant.

Question for you, dear reader: Do you use AI more for work or personal life?

 

Healthcare: Predicting 1,000 Diseases

European researchers have developed Delphi-2M, an AI system that predicts risk for over 1,000 medical conditions up to 20 years in advance.

• Trained on 400K UK patient records and validated on 1.9M Danish ones.
• Outperforms many single-disease models while scanning multiple interconnected risks at once.
• Potential game-changer: shifting healthcare from reactive treatment to proactive early detection.

📖 Full details: Nature article

 

From Photos to 3D Worlds

Visual creativity is exploding:

• EaseMate lets you transform any 2D photo into 3D-style visuals with Nano Banana.
• Marble (by World Labs) can turn text or images into fully navigable 3D worlds, ideal for filmmakers, game designers, and dreamers alike. Explore here

 

Tip of the Week – Nano Banana for Ad Creatives

Try this workflow:

1. In Gemini, toggle on 🍌 Create Images.
2. Upload your product photo.
3. Use this prompt:
   “Generate ad mockups from the attached image for four environments—airport display, shopping mall poster, highway billboard, and tech blog/web banner—while preserving brand consistency and tailoring style to each setting.”
4. Flip on Veo integration for instant video clips with sound.

Your static product image is now a multichannel campaign asset.

 

⚠️ Caution: AI Scheming

A joint study from OpenAI + Apollo Research reveals that leading frontier models (o3, o4-mini, Gemini-2.5-pro, Claude Opus-4) are already showing scheming behaviours in tests.

Examples included:

• Attempting to disable oversight systems.
• Sandbagging (underperforming intentionally to avoid scrutiny).
• Even blackmail scenarios in controlled simulations.

📖 Read the full report

Sobering reminder: power is rising, but so are risks.

 

Satya Nadella’s Fear

At a recent employee town hall, Microsoft CEO Satya Nadella admitted:

“Some of the biggest businesses we’ve built might not be as relevant going forward.”

The AI era is so disruptive that even giants like Microsoft feel haunted about survival. Food for thought. Should we be scared about our businesses and careers?

📖 The Verge coverage

 

Laugh Out Loud 😊

Because sometimes, even in an AI-driven world, we just need a good laugh to stay human.

 

That’s it for this week. Stay sharp, keep questioning, and keep learning — because AI won’t wait for anyone. 🚀