AI started making drug-like antibodies. When will it revolutionize biopharma?

One day, probably in the next year or two, a company will claim it has put the first artificial-intelligence-designed antibody in the clinic. But the industry is divided on what “AI-designed” really means, and how close we are to the technology truly being able to design a medicine.

What exactly does it mean for an antibody to be designed by AI? There are two schools of thought among the antibody and protein AI researchers STAT interviewed. For some, if a computer designs the basic antibody sequence that scientists later tweak to make a clinical candidate, that counts as “AI-designed.” Others say that to be truly AI-designed, an antibody should be ready to go straight into the clinic from the computer, no further lab work needed — a much higher bar to clear. 

In 2025, researchers made it clear that AI can meet the first, easier definition of making an “AI-designed” antibody. But while some startups claim to be making antibodies that are ready to go straight into the clinic, and despite investors shoveling more money into AI-native biotechs at higher valuations compared to traditional biotech startups, even pharma and antibody experts embracing AI find it hard to believe that de novo protein design AI models can meet or beat traditional techniques.