This weekly update presents remarkable advances in agentic AI systems—intelligent computer programs that can work independently to solve difficult scientific problems and make new discoveries.

Understanding What Agentic AI Does

Agentic AI refers to smart computer systems that don't need constant instructions from humans. Instead, they can think about problems, make decisions, and take actions on their own. These systems use large language models (powerful AI that understands and generates text) combined with special frameworks that let them reason through complex tasks. Think of an agentic AI agent like a helpful robot scientist that can read books, think about what it reads, come up with ideas, test those ideas, and keep improving.

Solving Impossible Puzzles

Researchers at Cognizant's AI Lab achieved something extraordinary with their MAKER system. They tackled the famous "Towers of Hanoi" puzzle—an ancient brain teaser that computers find very difficult when it gets large. While the smartest AI models in the world (like Claude 3.7 and DeepSeek-R1) fail when the puzzle gets bigger than eight disks, MAKER solved the puzzle with 20 disks and over one million steps without making a single mistake. The secret was breaking the giant puzzle into thousands of tiny pieces, having small AI agents solve each piece, and checking that every answer was correct. This breakthrough proves that agentic AI systems can handle real-world problems that need careful thinking and patience.

Training AI Smarter Using Evolution

Cognizant's AI Lab also introduced a brand new way to teach large AI systems. Scientists discovered they could use something called evolution strategies—inspired by how plants and animals improve over millions of years—to train enormous AI models better than older methods. This approach makes training faster and helps create agentic AI systems that think more like humans.

AI Scientists Making Real Discoveries

Perhaps the most exciting news involves AI co-scientists developed by major research organizations. These agentic AI systems can do the thinking work that scientists normally do. Google's Coscientist is one example—it reads thousands of scientific papers, understands what scientists already know, and creates completely new ideas for experiments. When researchers asked Coscientist to suggest treatments for liver disease, it found that a medicine already used for eyes could also help the liver. Scientists then tested this idea in the lab and confirmed it actually works.

Another breakthrough involved a system called Robin, built by researchers at FutureHouse. Robin represents something revolutionary: an agentic AI agent that can handle every step of scientific discovery all by itself. It searches for information, creates ideas, proposes experiments, analyzes the results, and refines its thinking. Robin demonstrated this ability by working on dry age-related macular degeneration, a serious eye disease that causes blindness. Without any human scientists guiding it through each step, Robin found a drug candidate called ripasudil that could help treat this disease. This usually takes scientists years of work, but Robin completed it in a fraction of that time.

Helping Engineers Design Better Technology

At Stanford University, scientists developed MetaChat, a new agentic AI platform designed to help engineers solve problems in advanced optics design—the science of light and how it bounces off special materials. This agentic AI system helps engineers think through complicated problems and find solutions much faster than before. MetaChat shows that agentic AI agents aren't just useful for biology and chemistry; they can help improve technology in many fields.

Why These Breakthroughs Matter

These achievements represent a major shift in how science and technology work. Instead of scientists and engineers having to do every step themselves, agentic AI agents can handle the thinking, planning, and testing parts of the work. This frees up human scientists to focus on the big-picture questions and deciding what experiments are worth trying. The result is that important discoveries that used to take decades might now happen in months or weeks, potentially helping more people by bringing new medicines and technologies faster.