Three stories landed this weekend that tell you exactly where AI is headed in 2026: Apple finally admitted defeat on building its own models, Anthropic made agents accessible to everyone, and Microsoft’s own data revealed an uncomfortable truth about who’s actually using this technology.
Apple Picks Google to Fix Siri
After years of watching Siri become a punchline, Apple made a billion-dollar admission: they can’t do this alone. The company announced a multi-year partnership with Google to power Apple Intelligence with Gemini, the same technology running Google’s search and Pixel devices.
The financial terms are staggering. Apple will pay approximately $1 billion annually for access to Google’s 1.2 trillion parameter model. That’s an eight-fold increase from the 150 billion parameter model currently powering Apple Intelligence. The overhauled Siri is expected to launch this spring.
What makes this deal interesting isn’t just the money. Apple tested competing technologies from OpenAI and Anthropic before selecting Google, suggesting the company prioritized raw capability over diversifying its AI partnerships. Their existing ChatGPT integration for complex queries remains unchanged for now, but the message is clear: when it came time to rebuild Siri from the ground up, Apple went with the model that powers two billion Android devices.
The privacy angle deserves scrutiny. Apple maintains that user data won’t be accessible to Google, with processing occurring on-device or through Apple’s Private Cloud Compute. Whether that architecture holds up under a model this large remains to be seen. What’s certain is that Alphabet’s market cap briefly touched $4 trillion after the announcement. Elon Musk called the alliance an “unreasonable concentration of power.” He’s not entirely wrong.
Anthropic Brings Agents to Non-Coders
While Apple was signing checks, Anthropic quietly launched something that might matter more in the long run. Cowork is essentially Claude Code for people who’ve never touched a terminal. Built into Claude Desktop, it lets you point Claude at a folder on your Mac and say “organize this” or “build a report from these notes.”
The pitch is simple: designate a folder, give instructions through regular conversation, and Claude handles the rest. Anthropic offers examples like sorting cluttered downloads folders, generating expense spreadsheets from receipt photos, or drafting reports from scattered documents. The feature integrates with existing connectors for Asana, Notion, PayPal, and other services.
Under the hood, it’s the same Claude Agent SDK powering Claude Code. Cowork uses Apple’s Virtualization Framework to run a sandboxed Linux environment, which means your files stay local while Claude operates in an isolated container. According to company insiders, the team built the entire feature in about a week and a half using Claude Code itself.
The catch: it’s currently limited to Max subscribers ($100 or $200 per month) with a waitlist for everyone else. Anthropic also warns about prompt injection risks and the potential for file deletion with vague instructions. They’re explicitly telling users to be precise about what they want. Windows support and cross-device sync are coming, but no timeline.
This feels like a preview of where all AI assistants are headed. The command line was never going to be how most people interact with autonomous agents. Cowork is Anthropic’s bet that a folder and a chat window is all the interface most people need.
Microsoft’s Data Reveals the AI Divide
Meanwhile, Microsoft published data showing AI adoption grew throughout 2025, but with a troubling pattern: the gap between who’s using AI and who isn’t is widening, not closing.
The numbers paint a stark picture. Roughly one in six people worldwide now use generative AI tools. But adoption in the Global North (24.7% of working-age population) grew nearly twice as fast as in the Global South (14.1%). The UAE leads globally with 64% adoption, followed by Singapore at 60.9%. The United States? Just 28.3%, ranking 24th globally.
Enterprise adoption tells a similar story. Over three-quarters of organizations claim to use AI, but only about 1% have mature deployments delivering real value. OpenAI’s own data shows “frontier workers” (the top 5% by usage intensity) send six times more messages than the median worker. The productivity gains aren’t being distributed evenly.
This matters because Microsoft is betting the company on enterprise AI. Over 90% of the Fortune 500 uses Copilot, but internal documents paint a different picture. CEO Satya Nadella reportedly criticized Copilot’s Gmail and Outlook integrations as tools that “don’t really work” and are “not smart.” Carnegie Mellon researchers found AI agents fail to complete real-world office tasks 70% of the time.
Nadella has taken an unusually hands-on role in fixing the product, essentially becoming Microsoft’s top product manager for AI. The company is pulling Copilot from WhatsApp and other messaging apps starting January 15th to refocus efforts. As Nadella put it in a recent post: “We are still in the opening miles of a marathon. Much remains unpredictable.”
What This Week Tells Us
These three stories share a common thread: the AI industry is recalibrating expectations. Apple is paying Google to solve a problem Apple couldn’t solve alone. Anthropic is making agents accessible because the command line isn’t how most work gets done. Microsoft is admitting, internally at least, that their flagship AI product needs serious work.
The billion-dollar partnerships and research previews will get the headlines. But the Microsoft data might be the most important signal. AI adoption is growing, but it’s growing unevenly. The gap between early adopters and everyone else isn’t closing. And the tools that promised to transform work are, by their own creators’ admission, still figuring out how to actually work.
2026 is shaping up to be the year AI gets real about what it can and can’t do. That’s probably healthy.
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.
