AI • ROBOTICS • BIOMOLECULAR RESEARCH
AutoDNA: AI Agents Take Experiments Into the Laboratory
A robotic biomolecular lab coordinates more than 20 instrument types, demonstrating DNA storage, synthesis and sequencing with limited human intervention.
ADDED NOVEMBER 5, 2026 • STUDY PUBLISHED NOVEMBER 4, 2026
From planning to physical experiments
Researchers at Shanghai Jiao Tong University and collaborators introduced AutoDNA in Nature on November 4, 2026. The system connects AI planning with robotic laboratory equipment to carry out biomolecular workflows, including nucleic acid synthesis, transcription, amplification and sequencing.
The study reports coordination of more than 20 instrument types across over 11,000 operations. One demonstration completed a DNA data-storage cycle and recovered the original digital content exactly.
How the pieces work together
The team’s public software repository separates the system into a user interface, a multi-agent AI component, a hardware scheduler and equipment-control software. Specialized agents handle literature, protocols, reagent inventories, code, hardware and hypotheses.
This division links a research request to an executable workflow. The scheduler coordinates equipment, while lower-level software communicates with the physical workstations. The repository requires access to Google’s Gemini API; reproducing the physical experiments also requires compatible laboratory hardware.
What the demonstrations showed
Alongside DNA storage, the authors report matching expert-optimized enzymatic DNA synthesis yields and speeding concurrent protocol execution by 3.2 times in their comparison. They also identified and validated a clinically detected KRAS G13D mutation.
Demonstrations have boundaries. A mutation-detection experiment does not establish a broadly validated clinical diagnostic service. The reported scheduling speedup applies to the tested workflows, rather than every experiment or laboratory.
Autonomous does not mean unattended forever
The paper describes human support for hardware maintenance and replenishment of reagents and labware. The public documentation also cautions that full reproduction is not guaranteed without the necessary equipment.
People designed the platform and selected its tasks. Successful execution within that platform does not demonstrate an unrestricted scientist able to investigate any biological question.
Why it matters
The interesting advance is the connection between software decisions and physical evidence. An AI proposal becomes more useful when it can be executed, measured and checked.
For future systems, useful questions include how often experiments succeed, how errors are recorded, whether results reproduce across equipment, and how much human effort remains. Reliable logs and independent validation will matter as much as the length of an automated run.
Primary sources
- Wu and colleagues: An AI-native experimental laboratory for autonomous biomolecular operations, Nature (2026).
- AutoDNA research code and documentation — architecture, supported components and reproduction requirements.
- Archived study code and data, linked in the paper.