Data Intelligence Agents match or beat SQL benchmarks across all seven tests
Researchers introduce Data Intelligence Agents (DIA), a three-agent system of autonomous coding agents that generates, executes, validates, and repairs data artifacts to compress enterprise data workflows, matching or surpassing published results on all seven SQL benchmarks tested.
Score breakdown
DIA replaces the multi-party, lossy handoff workflow of enterprise data integration with a fully autonomous, execution-grounded agent system that generalizes across SQL dialects and task categories without task-specific engineering.
- 01DIA is authored by Anoushka Vyas, Aarushi Dhanuka, and Sina Khoshfetrat Pakazad and is deployed in production for enterprise customers.
- 02The system comprises three autonomous coding agents: Data Interpreter, Schema Creator, and Query Generator.
- 03Agents generate, execute, validate, and repair concrete artifacts rather than emitting plain text.
Anoushka Vyas, Aarushi Dhanuka, and Sina Khoshfetrat Pakazad present Data Intelligence Agents (DIA), a system that addresses a core bottleneck in enterprise data integration: the repeated, lossy handoffs between data owners, engineers, and analysts who must collaboratively discover, structure, and query enterprise data. DIA treats autonomous coding agents (ACAs) as a first-class abstraction, meaning the agents do not merely emit text but instead generate, execute, validate, and repair concrete artifacts. A shared memory layer enables experience reuse across tasks, and each artifact is surfaced for review by domain experts. The system is deployed in production for enterprise customers.
DIA consists of three agents: the Data Interpreter, the Schema Creator, and the Query Generator.
DIA consists of three agents: the Data Interpreter, the Schema Creator, and the Query Generator. The paper focuses its evaluation on the Query Generator, testing it in fully autonomous mode across seven SQL benchmarks that span four task categories and four SQL dialects. The Query Generator matched or surpassed the best published results on all seven benchmarks, with the authors arguing that an architecture grounded in execution — built on ACAs and shared memory — generalizes across the data intelligence workload, requiring only natural-language instructions for adaptation rather than task-specific engineering.
Key facts
- 01DIA is authored by Anoushka Vyas, Aarushi Dhanuka, and Sina Khoshfetrat Pakazad and is deployed in production for enterprise customers.
- 02The system comprises three autonomous coding agents: Data Interpreter, Schema Creator, and Query Generator.
- 03Agents generate, execute, validate, and repair concrete artifacts rather than emitting plain text.
- 04A shared memory layer enables experience reuse across the three agents.
- 05The Query Generator was evaluated in fully autonomous mode across seven SQL benchmarks.
- 06Benchmarks span four task categories and four SQL dialects.
- 07DIA matched or surpassed the best published results on all seven SQL benchmarks.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 18, 2026 · 10:40 UTC. How this works →