A team of artificial intelligence agents has planned and carried out molecular biology experiments in a laboratory in China with almost no human hands involved, according to a paper published in the journal Nature on 4 November 2026.

The system, called AutoDNA, was built at Shanghai Jiao Tong University by a team led by computer scientist Haibo Chen. It coordinates more than 20 types of laboratory instrument across more than 11,000 steps, handling the core operations of working with DNA: synthesising it, transcribing it into RNA, amplifying it and sequencing it. A human operator gives the system a simple written prompt, and the agents design and run the experiment from start to finish. People are then needed only to refresh reagents and service the hardware.

Laboratories have used automation for years, but usually to carry out one fixed part of a protocol, such as mixing or heating chemicals. The Shanghai team set out to build something different: a laboratory designed around AI from the start, rather than a conventional lab with AI added on top. The aim was a self-driving lab, in which the AI system designs experiments, runs them, fixes problems and coordinates equipment across the whole room.

That is a harder problem in molecular biology than in chemistry, because the workflows are complex and demand high precision across many different machines. In most existing systems, human experts are still required to configure the instruments, and AI handles only one part of the work. AutoDNA moves samples between instruments with robotic arms and brings the planning, the execution and the checking together in one system.

The researchers tested it on four applications. In one, it detected a cancer-linked mutation, known as KRAS G13D, in a DNA sample. In another, it wrote a message into DNA itself, encoding a sentence using the four nucleotide bases, A, C, G and T, then read the DNA back to recover the message. The sentence chosen was a quotation often attributed to cosmologist Stephen Hawking: Intelligence is the ability to adapt to change.

That DNA data-storage test was the longest run in the study. According to the team's own account, reported by outlets including 36Kr and Chosun Ilbo, it lasted about 203.5 hours, roughly eight and a half days, and took 11,382 hardware steps. The system produced 78 DNA strands carrying the sentence and read them back without error.

Across the four applications, the paper reports that AutoDNA ran concurrent protocols 3.2 times faster than the comparison, matched the yields of expert-optimised enzymatic DNA synthesis, and completed a full DNA data-storage cycle with exact recovery. The code has been made public on GitHub, and the sequencing data and agent outputs are available on Zenodo, so other laboratories can inspect the work rather than take the claims on trust.

The system is not fully autonomous, and the paper is open about that. Fourteen hardware errors occurred during the experiments. The agents resolved eight of them on their own, and six needed a person to step in. The team also warns that the randomness built into large language models can change whether the same plan produces the same result when it is run again, a reproducibility problem that matters before such a lab is trusted with high-stakes work.

Even so, outside experts quoted by Nature called the work remarkable, both for the degree of autonomy and because the system can readily take on a diverse range of lab equipment. A commentary in Nature by James Zou of Stanford University said the work shows how AI systems can begin to close the gap between proposing an experiment and actually doing it.

Self-driving labs have been discussed for years, but most of the progress so far has been in planning and analysis, while the physical work stayed with people. AutoDNA is one of the clearest demonstrations yet of agents carrying out physical science, not just summarising literature, and it suggests a future in which a researcher's main job in the lab is to ask the question, refill the chemicals and check the result.