Agentic-SQL Taxonomy: A Survey of Autonomous and Interactive Text-to-SQL with LLMs

Text-to-SQL systems have transitioned from simple machine translation models to complex reasoning frameworks as database schemas grow in scale and ambiguity. Despite the impressive capabilities of Large Language Models, one-shot generation often fails to produce correct SQL in real-world scenarios. This survey introduces the Agentic-SQL Taxonomy, an autonomy-based classification [21] that reevaluates existing methods through the lens of inference complexity. We categorize research into single-turn generation, iterative refinement, and multi-agent collaboration to highlight the shift toward interactive debugging and collective reasoning. We analyze how these sophisticated pipelines bridge the performance gap on challenging benchmarks. Current evaluations show that leading models can result in incorrect execution in nearly 40 % of cases when instructions are vague or incomplete. Our work identifies executionguided feedback and modular agent architectures as the primary drivers of future progress in building robust and reliable database interfaces.