AutoDNA lab completes a DNA storage experiment across 11,382 operations

AutoDNA encoded digital information into DNA and recovered it after a workflow involving 11,382 hardware operations. The Shanghai Jiao Tong University team reported the work on November 4, 2026. This is a demonstrated connection between AI planning and physical experimentation, rather than a chatbot merely proposing a laboratory procedure.

DNA storage represents information using molecular sequences instead of a conventional storage medium. Imagine changing both the alphabet and the paper: computation determines the molecular message, synthesis writes it, sequencing reads it, and decoding recovers the digital content. Recovering the original information closes that loop. The result is a laboratory demonstration, not a consumer replacement for a hard drive.

There is important precedent. The A-Lab described in 2023 combined experiment planning, robotics and interpretation to synthesize inorganic powders. AutoDNA therefore did not invent autonomous laboratories. Its challenge is coordinating molecular-biology operations across equipment with different functions, while respecting timing, liquid handling and experimental conditions.

The university describes a common interface that makes instrument capabilities available to the AI system. Its public repository separates the software into a user interface, a multi-agent planning component, a hardware scheduler and the programs controlling the physical machines. A restaurant offers a useful analogy: taking an order is different from cooking it, and the kitchen must respect dependencies rather than put ingredients in the pan before preparation is complete.

The research reports orchestration of more than 20 instrument types and demonstrations of nucleic-acid workflows. People still maintain equipment and replenish reagents and consumables. The repository also makes clear that obtaining the software is not sufficient to reproduce physical experiments: suitable laboratory hardware is required.

The prospective benefit is less manual coordination between machines, not the disappearance of scientific judgment. A system that can translate a goal into a long, executed workflow expands what laboratory automation can attempt. Whether it remains reliable across different tasks, handles unexpected conditions and produces sound results will matter more than the raw number of steps. One completed demonstration is encouraging without establishing that every biological investigation can now run unattended.

Original paper published: November 4, 2026.