Competition in AI for Science is extending toward a closed research loop.
I
n
protein design, for example, models are becoming stronger, yet researchers still need
AI for Science Enters the Research Loop: Protein Design Firm Turns Agent Into a Scientific Workbench
Lévin Institute, a Hangzhou-based AI protein design company, has released Lévin Harness for the research community, placing data, models, plugin tools, computing power, and workflows into a single Agent workspace. Researchers define scientific goals and key decisions, while the Agent handles investigation, tool calls, and compute orchestration. As AI moves beyond one-off inference into continuous work in real research settings, the ability to organize models, tools, and compute becomes...