AI Agents Ran an 11,382-Step DNA Experiment; Humans Just Refilled Reagents

What happened

The paper, titled "An AI-native experimental laboratory for autonomous biomolecular operations," was submitted on Oct 1, 2025, accepted on Sep 30, 2026, and published in Nature on Nov 4, DOI 10.1038/s41586-026-11228-6. A multi-agent system orchestrates more than 20 types of instruments, covering DNA synthesis, RNA transcription, amplification and sequencing; the abstract says human intervention is limited to hardware maintenance tasks such as reagent and labware replenishment, and the code is available on GitHub.

The team tested the system on four tasks: DNA synthesis output matched human experts; running several protocols in parallel was 3.2 times faster than running them one after another; DNA data storage read back without errors, with the longest run lasting 203.5 hours and 11,382 hardware steps, producing and reading back 78 DNA strands that encoded the sentence "Intelligence is the ability to adapt to change." with no errors; and the system found and confirmed the cancer-linked KRAS G13D mutation in a patient sample.

The limits are in the paper too: the experiments recorded 14 hardware errors, of which the system fixed 8 on its own and 6 needed a person; the team warns that randomness in large language models can change whether the same plan gives the same result when run again, and the system cannot yet run experiments in live animals. Milad Abolhasani, a chemical engineer at North Carolina State University who did not work on the project, called AutoDNA "remarkable both for its degree of autonomy and because it can readily incorporate a diverse range of lab equipment." The paper first appeared as a preprint in July 2025, when the longest run was 162.9 hours and 9,365 steps.

Key facts

1. Paper: "An AI-native experimental laboratory for autonomous biomolecular operations," Nature, Nov 4, 2026, DOI 10.1038/s41586-026-11228-6
2. Team: Shanghai Jiao Tong University; corresponding authors Zhaoguo Wang, Fei Wang, Chunhai Fan and Haibo Chen
3. System: a multi-agent system orchestrates more than 20 types of instruments, covering DNA synthesis, RNA transcription, amplification and sequencing
4. Human role: the abstract says human intervention is limited to hardware maintenance such as reagent and labware replenishment
5. Parallel speed: running several protocols in parallel was 3.2 times faster than running them sequentially
6. Longest run: 203.5 hours and 11,382 hardware steps, producing and reading back 78 DNA strands with the encoded sentence recovered without error
7. Encoded sentence: "Intelligence is the ability to adapt to change."
8. Clinical task: the KRAS G13D mutation was found and confirmed in a patient sample
9. Errors: 14 hardware errors in total; the system fixed 8 on its own and 6 needed a person
10. Limits: LLM randomness affects repeat runs; the system cannot yet run experiments in live animals; the July 2025 preprint's longest run was 162.9 hours and 9,365 steps