Beyond Text-to-SQL: An Agentic LLM System for Governed Enterprise Analytics APIs

A new agentic system called Analytic Agent translates natural language questions into governed enterprise analytics API calls, aiming to bridge the gap between non-technical users and complex organizational data without exposing raw databases to large language models, according to a preprint posted on arXiv [1]. The system, detailed in a paper by Md Tahmid Rahman Laskar and colleagues, addresses a persistent challenge in enterprise analytics: making data accessible for decision-making while maintaining security and compliance [1][2]. Traditional business intelligence tools and Text-to-SQL systems often require technical expertise that many business users lack [1]. Recent large language model approaches that generate SQL directly promise easier access but introduce risks when applied to enterprise environments where analytics pipelines depend on governed APIs, not raw databases [1][2]. These APIs embed complex business logic to ensure consistency, auditability, and security [2]. Delegating mathematical or aggregation logic to an LLM can create reliability and compliance gaps [1][2]. Analytic Agent operates as an LLM-based agentic system that interprets user goals, validates permissions, executes governed queries, and produces compliant visualizations through multi-step reasoning and policy-aware orchestration [1][2]. The researchers evaluated the system on 90 real enterprise use cases constructed by domain experts [1][2]. The paper was submitted to arXiv on May 20, 2026, and revised on June 15, 2026 [1]. arXiv, an open-access repository founded in 1991, hosts preprints across mathematics, physics, computer science, and related fields, and as of November 2024 receives about 24,000 submissions per month [4]. Papers on the platform are moderated but not peer-reviewed before posting [4]. The work builds on the broader trajectory of large language model research, which accelerated after the 2017 introduction of the transformer architecture in the paper “Attention Is All You Need” [6]. That architecture now underpins most large language models and has been cited more than 250,000 times as of 2026 [6]. The Analytic Agent preprint appears under arXiv’s Computation and Language category and is associated with arXivLabs, a framework for experimental community projects on the platform [1].