Medra, in collaboration with DARPA, has launched AI Experimentalist, a scientific reasoning layer that translates high-level research goals expressed in natural language into executable lab workflows. The system, part of Medra's physical AI platform, aims to address one of drug discovery's most stubborn bottlenecks: scaling experimental validation to generate the data foundation models need.
Physical AI tackles the data generation gap
Medra CEO Michelle Lee, PhD, told GEN Edge that building biology foundation models capable of predicting and curing disease "will take thousands of years of data generation." She added, "The more I looked at the field, the more I realized that this data problem is actually a robotics problem." Medra positions its platform as a way to apply physical AI - intelligent robotic systems equipped with sensors - to experimental tasks, moving beyond the repeatable automation common in combinatorial chemistry and screening.
From natural language to full experimental cycles
AI Experimentalist is designed to interpret a scientist's high-level prompt and convert it into a multi-step workflow. In a blog post, Medra described an example: a prompt to "build an Epidermal Growth Factor Receptor (EGFR) blocking antibody assay cascade" prompts the system to propose optimized execution steps. These include testing linear DNA templates in parallel, adjusting expression conditions, and feeding results into subsequent runs, which can compress timelines from days to hours. The platform covers the entire cycle, from literature review and wet-lab execution to data analysis and protocol refinement.
Why subtleties matter in autonomous labs
Lee highlighted that even minor variables - pipette angle, timing of reagent mixing - can determine whether an experiment succeeds. "The artisanal nature of science is actually what makes certain experiments work and others fail," she said. Traditional industrial automation handles repeatable tasks well but cannot adapt to these nuances. Medra's approach equips standard lab hardware with sensors and decision-making capabilities, enabling the robot to adjust on the fly.
Multi-agent architecture and partnerships
AI Experimentalist uses a multi-agent architecture and a model-agnostic harness, allowing Medra to integrate new biological AI models as they emerge. Current integrations include NVIDIA Nemotron models for protocol editing and optimization, as well as the newly launched NVIDIA BioNeMo Agent Toolkit. Medra operates both on-site labs at customer facilities and remote access through its 38,000-square-foot autonomous lab, Medra Lab 001 (ML001), which runs experiments around the clock. Partners across academia, biopharma, and government are already using the platform for antibody discovery, protein engineering, gene editing, and cell biology. As physical AI platforms grow more capable, structured AI Learning Path for Research Scientists programs can help professionals adapt these tools to their own experimental workflows.
Why this matters for science and research
For research scientists, Medra's launch signals a shift from software-only AI tools to integrated robotic platforms that execute entire experimental cycles. Rather than simply recommending protocols, systems like AI Experimentalist close the loop by performing the work and incorporating results into the next iteration. This could reduce the time and manual labor required for assay development, freeing scientists to focus on higher-level experimental design and hypothesis generation. The platform's model-agnostic design also suggests that as new biological AI models emerge, labs won't need to overhaul their infrastructure - they can plug in better reasoning engines over time.
Your membership also unlocks:
