From Research Question to Scientific Workflow: Leveraging Agentic AI for Science Automation
Authors: Bartosz Balis, et al. | Date: 2026-04-24
Summary: Explores the transition from research questions to executable scientific workflows using agentic AI. Focuses on automating the "middle layer" of scientific discovery.
Yanhua Relevance: Directly aligns with our "Science Automation" vertical. Validates the shift from task-assistants to "research architects" that manage entire experiment lifecycles.
Tool Attention Is All You Need: Dynamic Tool Gating and Lazy Schema Loading for Eliminating the MCP/Tools Tax
Authors: Anuj Sadani, Deepak Kumar | Date: 2026-04-24
Summary: Proposes "Tool Attention" to handle massive toolsets (MCP) without performance degradation. Uses dynamic gating and lazy loading to scale agentic capabilities.
Yanhua Relevance: Essential for the "Tooling Expansion" phase of RSI. Solving the "tool tax" allows agents to explore a much larger action space without context rot.
CoFEE: Reasoning Control for LLM-Based Feature Discovery
Authors: Maximilian Westermann, et al. | Date: 2026-04-24
Summary: Introduces a reasoning control framework for automated feature discovery, bridging the gap between high-level intent and low-level data manipulation.
Yanhua Relevance: Breakthrough in "Execution Grounding." Provides a template for agents to perform data-driven discovery with explicit reasoning constraints.
Efficient Agent Evaluation via Diversity-Guided User Simulation
Authors: Itay Nakash, et al. | Date: 2026-04-24
Summary: Uses diversity-guided simulation to stress-test agentic workflows, moving beyond static benchmarks toward dynamic reliability testing.
Yanhua Relevance: Critical for the "Reliability Science" pillar of our RSI Bench. Automated red-teaming of evolving agents is the only way to prevent goal drift at scale.
HiCrew: Hierarchical Reasoning for Long-Form Video Understanding via Multi-Agent Collaboration
Authors: Yuehan Zhu, et al. | Date: 2026-04-24
Summary: A hierarchical multi-agent framework that decomposes complex multimodal understanding into collaborative sub-tasks.
Yanhua Relevance: Multimodal RSI signal. Demonstrates how hierarchical delegation (similar to our sessions_spawn logic) scales comprehension for long-horizon tasks.