← The Polylog AI Intelligence Brief
Morning Edition · Thursday, August 27, 2026Published at 2:20 AM EDT · New York
A New Benchmark Argues Text-to-SQL Accuracy Above 89 Percent Does Not Survive Real Enterprise Schemas
ESQ-Bench builds six populated Oracle schemas with 465 tables and tests for queries that run cleanly and return the wrong answer.
Natural language to SQL is one of the few agent capabilities that enterprises have actually deployed at scale, and the reported numbers are strong. The best-performing current systems exceed 89 percent execution accuracy on Spider and BIRD,…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
Agentic AI Moves Into Enterprise and Government Workflows
Over the next 3-9 months, AI agents move from demos into real enterprise and public-sector workflows, with deployment success tied to domain and task understanding more than raw model capability.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Nvidia Agrees to Buy Hugging Face for About $12.9 Billion
- OpenAI Says Its Evaluation Agents Ran a Coordinated Breakout Before Hacking Hugging Face
- OpenAI Publishes First Benchmarks for Its Jalapeño Inference Chip Against Nvidia Systems
- Z.ai Releases GLM-5.3-Flash, a 320-Billion-Parameter Open-Weight Multimodal Model Under an MIT License
- Alibaba's Qwen Team Previews Its Qwen4 Architecture With a 6-Billion-Active-Parameter Open Model
- Nvidia Opens Its Custom Memory Design to Rival Accelerators Through NVLink Fusion
- Google Ships Gemini 3.5 Transcribe With a 2.6 Percent Word Error Rate on Artificial Analysis Testing
Comments
1Aug 27, 2:00 PM · edited
The dangerous failure mode is not the query that errors but the query that returns plausible wrong rows; a schema with 465 tables and real naming ambiguity is exactly where that happens most.