Avalilação PREreview de Deep Research is the New Analytics System: Towards Building the Runtime for AI-Driven Analytics

This paper explores an important convergence between two emerging paradigms in AI-driven analytics: optimized execution via semantic operators and the flexible, agent-based reasoning found in Deep Research systems. The authors clearly articulate the limitations of each approach when used in isolation - semantic operators suffer from high execution costs and limited interactivity, while Deep Research systems lack explicit query planning and optimization. The proposed runtime aims to bridge this gap by enabling Deep Research agents to generate and execute optimized semantic operator programs.

A key strength of the work is its system-level perspective, which frames AI-driven analytics as a runtime problem rather than solely a modeling or prompting challenge. The prototype demonstrates meaningful improvements in both quality and efficiency, achieving higher F1-scores than baseline Deep Research agents and substantial reductions in cost and runtime through optimized execution. These results suggest that combining agentic planning with declarative, optimized operators is a promising direction for scalable AI analytics.

The evaluation, while encouraging, is limited to a small set of basic queries, and broader workload diversity would strengthen confidence in generalizability. Nevertheless, this paper provides a compelling architectural direction for AI-driven analytics systems and offers a valuable foundation for future research on runtimes that balance flexibility, interactivity, and execution efficiency.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.