Research harnesses for LLM agents, multi-agent systems, and agentic AI for Science. Open research on evaluation, feedback, memory, and reproducibility.