Autonomy also has a physical price. Running a whole lab unattended means many parallel stations consuming reagents, samples, instrument time, and maintenance. Each cycle takes real hours, and failures inside an unattended run can waste an entire batch before anyone notices. Throughput is bounded by physics and budget, not by how clever the model is. This is why most real systems run narrow, bounded campaigns rather than open-ended unattended operation.
Can AI Run a Drug Discovery Lab on Its Own?