Researchers at Shanghai Jiao Tong University have built an AI system that can take a single typed instruction and autonomously design and execute a full laboratory experiment across more than 20 pieces of equipment, according to a study published in Nature.
The system, called AutoDNA, uses AI agents to translate a researcher's written goal into a detailed experimental plan and then carry it out without step-by-step human direction. In tests, the system matched yields achieved by human experts in enzymatic DNA synthesis and ran concurrent experimental protocols roughly 3.2 times faster than sequential, manually directed execution. In one demonstration, it completed a full DNA-based data-storage cycle — encoding digital information into DNA strands and then reading it back exactly — across more than 11,000 individual hardware steps over roughly 203 hours. An earlier preprint version of the work cited a somewhat lower step count, a discrepancy that appears to reflect refinements made before formal publication.
Execution, not discovery
The result is significant mainly as a demonstration of autonomous execution of pre-designed protocols, not autonomous scientific discovery — the system carries out experiments a human has effectively specified at a high level, rather than independently generating novel hypotheses. Outside researchers quoted in press coverage described the result as encouraging but offered measured, "cautiously positive" reactions rather than unqualified endorsement.
A key limitation is that these results come from the authors' own laboratory, and independent replication by other research groups has not yet occurred. That is standard for a newly published system, but it means the reported speed and accuracy gains should be treated as preliminary until other labs can reproduce them with their own equipment and reagents.
Why it matters
If the approach holds up under outside testing, systems like AutoDNA could reduce the person-hours needed for repetitive experimental work in fields such as synthetic biology and materials science, freeing researchers to focus on experimental design rather than manual execution. The authors say they plan to expand the range of experiment types the system can handle beyond the DNA-synthesis and data-storage tasks demonstrated so far.
