AI as Tool vs AI as Inventor: Why the Distinction Will Decide the Next Decade of Patents

There are two versions of "AI in drug discovery" being marketed right now, and they are not the same thing. Version one: AI as a tool. A scientist uses structure prediction, screens a virtual library, runs molecular dynamics, and proposes a lead. The AI is a powerful instrument; the scientist remains the inventor, and patent claims downstream are defensible because there is a documented chain of human inventive contribution at every key decision point. Version two: AI as an agent. A model is given a target and an objective and generates candidates through automated pipelines with minimal human intervention; the discovery is logged as a system output, and the patent claims downstream are murky.

Patent offices increasingly draw a hard line: an AI system cannot be a named inventor, and AI-generated outputs are patentable only when a natural person can demonstrate significant inventive contribution. The boundary is real, and the cost of getting it wrong can be the loss of patent protection on assets that cost a great deal to generate.

Three Practices That Matter Now

For biotech founders, three practices matter now. First, document human inventive contribution at every decision point — target selection, screening modality, scaffold choice, linker design for a degrader — captured, dated, and attributed, so the patent file reads like a scientific notebook, not a software changelog. Second, separate the tool from the invention: your models, pipelines, and cloud architecture are not your IP unless you have separately built proprietary, patentable methods; your IP is the molecules, methods, and biological insights that come out the other end. Third, maintain an IP-defensible audit trail: version-control code, log model versions, and keep the chain of custody from hypothesis to molecular design to validation result.

Special Consideration for PROTACs and Beyond

This is especially critical for heterobifunctional modalities — PROTACs, LyTACs, MoDE-A platforms — where the combinations of warhead, linker, and ligand are the inventive step. AI is excellent at generating combinations; demonstrating which were inventive and which were obvious to try requires scientific judgement captured in the record. The lesson for teams building AI-accelerated pipelines: AI is your tool; the invention is still yours — but only if you can prove it.