Scientific Research & Discovery Weekly AI News

November 17 - November 25, 2025

This week brought exciting news about AI agents - special computer programs that can think and make decisions on their own. These are not just tools that follow orders; they are more like helpful teammates that can plan, learn, and fix problems without someone telling them exactly what to do every single time.

Microsoft's Big Announcements at its Ignite conference showed how this technology is moving into businesses everywhere. The company introduced new products with names like Work IQ and Fabric IQ that help AI agents understand what humans are doing and find the information they need quickly. Microsoft also launched something called Agent 365, which is like a control center where companies can build, manage, and watch over their AI agents.

Scientists and researchers have discovered something important: AI agents that can learn from corrections work much better than the old kind. Instead of people having to learn how to use stiff software, the software now learns how to work with people. This makes it possible to handle messy, real-world problems that old computer systems couldn't handle before.

The biggest tech companies are racing forward. Google announced that its next super-powerful AI model, called Gemini 3, will arrive before the year ends. Meanwhile, Microsoft partnered with other companies like Anthropic and NVIDIA to bring even more powerful AI tools to their cloud service called Azure.

One really cool finding is that 35% of businesses are already using agentic AI, and 44% more are planning to start soon. However, nearly half of companies still don't have a clear plan for what to do with this new technology. Experts say that companies treating AI agents as only cost-cutting tools are missing the bigger picture - these agents can actually become smarter and more helpful over time, making them valuable learning partners.

Research also shows that having outside experts help build AI agents works about twice as well as trying to build them alone - around 67% success rate compared to internal attempts. The future looks like special AI agents made just for certain jobs, like one for hospitals, another for banks, and another for stores.

Extended Coverage