Why AI agents might change how we do science

Scientists are now shifting their attention toward AI agents. Unlike the older tools that act like giant reference books, an agent is an AI that has been given the ability to use tools and reason through a problem step-by-step. Instead of just guessing an answer based on a mountain of training data, these agents can draft theories, criticize them like a peer reviewer, and run experiments. Recently, one such system successfully identified how antibiotic resistance spreads between bacteria, reaching the same conclusion as human researchers who had spent a decade in the lab.

Think of traditional AI as a librarian who has memorized every book in the building but can only answer questions using the exact text they were fed. If you ask them something not in their books, they are stuck. AI agents are different; they are more like a smart, tireless research assistant. They use a core reasoning engine—the brain behind tools like ChatGPT—and are granted access to digital tools, like calculators or databases. When you give them a goal, they break it into small tasks. One part of the AI might look for existing patterns, another might test those ideas, and a third might refine the results based on what it finds. Crucially, because they are digital, they keep a perfect log of every step they take. This fixes a major headache in science: the lack of clear, detailed records. If a human forgets to jot down a small detail in a lab notebook, the experiment might be impossible for someone else to replicate. An agent records everything by default.

The current system of science is straining. Researchers are drowning in an ever-growing number of papers, and finding enough qualified people to review them has become a struggle. This bottleneck slows down discovery and makes it harder to separate high-quality work from the rest. By taking over the tedious parts of the process—like checking logic, logging data, and drafting hypotheses—AI agents could act as a relief valve. While these systems are still prone to mistakes and need human oversight, they offer a way to keep up with the modern pace of research. The future of science may not just be about having more data, but about having better ways to reason through it.