Over billions of years, nature has evolved an extraordinary array of proteins—molecules responsible for virtually everything that occurs within a living organism, such as transporting substances, transmitting signals, fighting infections, building cellular structures, and carrying out chemical reactions essential for life. Each of these proteins begins with an amino acid sequence organized according to instructions stored in DNA.

However, there is a crucial detail: knowing the sequence alone is not enough to understand what a protein does; one must also discover the shape it assumes in space. This is because a protein does not remain a simple chain of amino acids; it folds into complex three-dimensional structures, and the way these structures are organized largely determines the protein's functions. In the past, determining the structure of a single protein could require months or even years of experimental work; it was precisely in addressing this challenge that artificial intelligence brought about one of the most significant recent transformations in biology.

In 2020, AlphaFold—developed by DeepMind—demonstrated an extraordinary ability to predict protein structures from their amino acid sequences. Two years later, DeepMind and the European Bioinformatics Institute made structural predictions available for more than 200 million proteins. Suddenly, researchers around the world gained access to information that might otherwise have required decades of experimental research—yet predicting a protein's structure is not the same as fully understanding how it functions.

A structural prediction does not automatically reveal all the molecular interactions, movements, or chemical reactions a protein can undergo; nevertheless, it offers something extremely valuable: a starting point for investigating biological mechanisms that were previously very difficult to study. It was precisely this capability that gave rise to an even more ambitious question: if artificial intelligence can learn how proteins are organized, might it also help create entirely new proteins? That is why experiments remain indispensable.

Artificial intelligence can accelerate discovery, but it is actual biology that determines whether such a discovery works. This combination of computational models and physical experiments is beginning to transform how science investigates organisms. This means that AI tools focused on the study and design of proteins can also contribute to the development of new genetic-modification tools; in fact, researchers are already exploring this possibility, including through systems that use AI to design protein variants associated with DNA editing. And here, the story begins to come full circle.

The experiment involving "Cloud" agents represents an AI searching for biological mechanisms that are still poorly understood; AlphaFold represents the ability to predict molecular structures; and generative protein models represent the next stage: proposing new structures for scientists to investigate. Discovering, understanding, and beginning to design—these are three distinct capabilities that are converging within laboratories. This leads us to a trajectory that extends well beyond protein research: if we can harness artificial intelligence to study genes, design molecules, and refine tools capable of modifying cells, how far might this technology go when applied to the human body itself? After all, using genetic engineering to treat a disease is one thing—but it is quite another to begin asking whether we might modify human traits that were never considered diseases in the first place.