This weekly update covers major developments in how artificial intelligence agents are being used in science and discovery. An agentic AI is a computer system that can make decisions and complete tasks on its own, like having a digital helper that works 24/7.

One of the biggest stories this week was that OpenAI shut down Sora, which was a tool that created videos using AI. The company said it was too expensive to run. This shows that even the smartest companies have to think carefully about which AI tools actually work well for real use. However, this setback did not slow down other important developments in the field.

Another huge milestone happened when the Model Context Protocol (MCP) reached 97 million installations. Think of MCP like a universal language that lets different AI agents talk to tools and databases around the world. This is important for science because it means researchers can now build AI helpers that can access scientific data, run experiments, and search through millions of research papers automatically.

Companies like Oracle announced new AI tools for scientists and businesses. These tools help AI agents understand and work with real data from companies and research centers. Meanwhile, researchers published a new tool called OpenScientist, which is an AI agent that can help with medical research by coming up with ideas, running tests, and finding patterns in data.

Scientists also published important research about how AI thinking works differently from human thinking. AI is good at finding tiny patterns in huge amounts of data and never getting tired, but humans are better at understanding tricky situations and explaining why things matter. The best science happens when AI and humans work together as a team, with AI handling the hard math and humans making the important decisions.