AI Weekly: Agents Get to Work, Chatbots Start Selling, and Regulators Step In

Welcome back to your weekly roundup of everything shaping the fast-moving world of artificial intelligence. From tireless AI coworkers to chat-based shopping, here’s what you need to know this week.

Agentic Tools & Productivity Go Mainstream

The race for productive AI heats up as Anthropic and Microsoft push agentic tools into everyday workflows.

Anthropic’s Claude Sonnet 4.5 can now run autonomous tasks for up to 30 hours, writing tens of thousands of lines of code and analyzing live dashboards for users like Canva (The Verge). Anthropic calls it its most capable agentic model yet—basically a tireless digital coworker. Early impressions suggest that while the raw code quality isn’t dramatically different, Claude 4.5’s ability to sustain long, complex tasks without losing focus is a major leap forward.

Meanwhile, Microsoft’s new “Agent Mode” brings prompt-driven creation to Excel, Word, and PowerPoint. Powered by GPT-5, the Copilot Office Agent can build a slide deck while researching online and logging sources automatically (The Verge). It’s part of the new Microsoft 365 Premium plan. The concept sounds promising, even if early expectations are tempered—it might not be amazing right out of the gate, but there’s clear potential for it to become a powerful productivity tool as it matures.

Chatbots Are Becoming Shopping Assistants

OpenAI and Stripe quietly launched the Agentic Commerce Protocol, letting U.S. ChatGPT users buy Etsy products directly from chat (OpenAI, TechCrunch). No tabs, no checkout page—just “buy this,” and done. Analysts say this could shift e-commerce traffic away from search engines toward chat-driven experiences. This is a huge step, and personally something I’m excited about—imagine delegating shopping to an agent that orders healthy groceries automatically. It might not be perfect right away, but the potential is undeniable.

Personalization Meets Privacy

In a move that will surprise absolutely no one, starting Dec 16, Meta will mine its Meta AI chats for ad-targeting data—talk about hiking boots, and Facebook will know (Reuters). Sensitive topics like religion or health are excluded, and the rollout skips the UK, EU, and South Korea.

After a recent lawsuit, OpenAI added new parental controls for ChatGPT (Reuters), letting parents link accounts, restrict memory, and limit whether conversations train models. Meta followed with similar safeguards for minors. There’s a lot to unpack here—it’s a positive move toward protecting younger users, but it raises the question: will these controls actually be used? That could be a discussion worthy of its own post.

Creative Rights & Deepfakes Take Center Stage

OpenAI announced that Sora video creators will be able to block or monetize use of their IP and even join a revenue-sharing model (Reuters). It’s a meaningful step for artists trying to regain control, but it also raises questions about where this road leads. The technology itself is impressive, but is combining AI generation with social media distribution really the direction we want to go? For many, AI and social media are already polarizing forces—putting them together could feel like pouring gas on a fire.

Elsewhere:

• SAG-AFTRA condemned the debut of a synthetic actress at Zurich, calling it a “direct threat to human performers” (Reuters).

• Bollywood stars Abhishek Bachchan and Aishwarya Rai sued over deepfake videos and urged platforms to stop such content from training models (Reuters).

Infrastructure, Funding & The Compute Race

The cloud wars escalated: Meta signed a $14.2 B deal with CoreWeave for Nvidia-powered infrastructure through 2031 (Reuters), while Microsoft committed $33 B to smaller “neocloud” partners (Data Center Dynamics).

Chipmaker Cerebras raised $1.1 B at an $8.1 B valuation (Reuters), signaling IPO readiness. And Samsung + SK Hynix joined OpenAI’s “Stargate” initiative to co-build AI data centers in South Korea (OpenAI).

Startups & Research Labs Push Boundaries

Meet Periodic Labs, a new venture from ex-Google, DeepMind, and OpenAI researchers, now armed with $300 M to automate scientific discovery via robot labs—starting with new superconductors (TechCrunch).

Opera Neon, an AI-native browser, dropped this week with tools to run code, fill forms, and analyze data directly on-page (Reuters). Think of it as a browser that does the homework for you.

Smart Homes Get Smarter (and Creepier?)

Amazon rolled out upgraded Echo, Fire TV, and Ring/Blink devices featuring the next-gen Alexa+ assistant (Reuters). The new cameras can even tell whether a visitor is a delivery or a threat. Personally, this is sweet—I’ve been waiting for some meaningful advancements in the home space, which feels woefully behind the rest of AI’s evolution. I’m genuinely excited for the promise of Jarvis-like smart home solutions and hope this is the start of that next leap.

Google’s “Gemini for Home” replaces Assistant on Nest devices, capable of running complex conversational routines and summarizing events (Google Blog, TechCrunch).

Laws, Fraud & the Coming AI Crackdown

California passed SB 53, the first state law for frontier-AI developers, requiring governance frameworks and whistleblower protections (FPF.org).

In Brazil, police arrested a deepfake scam ring using fake Gisele Bündchen videos to lure investors (Reuters). The Supreme Court warned platforms could face penalties if they ignore such fakes.

Oddities & Future Gazing

Jeff Bezos predicted orbiting data centers within two decades, powered by uninterrupted solar energy (Reuters). He compared today’s AI boom to the early Internet bubble—“some hype, some revolution.” It’s an intriguing idea, and it seems plausible—especially alongside Microsoft’s earlier experiments with ocean-based data centers, which explored similar ideas about environmental efficiency and cooling.

The Takeaway

This week, AI agents stepped out of the lab and into the boardroom. E-commerce is being rewritten in chat windows, creative industries are fighting for their digital likenesses, and regulators are—finally—writing the fine print. The next phase of AI isn’t just about intelligence; it’s about agency, accountability, and access.

✌️Steven