Plan and run lab experiments on automated equipment
An agent reads documentation, writes code for lab robots, and plans experiments, with scientists approving each run.
The use case
Automated labs need code to drive liquid handlers and instruments, and writing it is a bottleneck. Coscientist, described in Nature by Carnegie Mellon researchers, searched documentation, wrote control code for lab hardware, and planned and ran chemistry experiments such as optimizing reactions. It shows what agents can do in self-driving labs. In practice, scientists set the goals, review generated protocols and code, check safety constraints, and approve runs on real equipment. Most labs that try this begin in simulation or with cheap, safe reagents, and keep a person at the instrument during early runs until the generated protocols prove reliable.
Example workflow
- Define the experimental goal and constraints
- Agent reads instrument docs and drafts code
- Scientist reviews protocol and code
- Run in simulation or with safe reagents first
- Approve and monitor real runs
Tools
- Coscientist
- Opentrons
Risks and guardrails
- Hardware errors can waste reagents or cause hazards
- Approve each run on real equipment
- Results need replication before claims
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Nature: autonomous chemical research with LLMs: www.nature.com/articles/s41586-023-06792-0