AI Agent Harness for Autonomous Data Visualization Design

A research paper on arXiv introduces a general end-to-end agentic harness that autonomously designs custom visual analysis applications (VIS apps) from raw data and high-level task descriptions. This system represents a step toward an AI co-scientist capable of executing long-horizon tasks without human intervention. The harness coordinates multiple agents and specialized skills for exploratory analysis, planning, environment configuration, and implementation. The paper is available at arXiv:2605.21825.