Zilliz, the company behind Milvus, the world’s most widely adopted open-source vector database, today forecasts a major shift in enterprise data interaction: natural language interfaces powered by AI agents will become the default way organizations query databases for AI-driven workloads by 2026, with SQL moving into a secondary role for traditional analytics and reporting.
We’re entering an era where talking to your database will be more productive than scripting against it,” said James Luan, VP of Engineering at Zilliz. “SQL will still matter, but it no longer defines how people interact with data. Natural language is becoming the default interface for AI workloads because it lets people focus on intent instead of syntax.”
AI Agents Turn Data Access into Dialogue
With natural language interfaces, users describe what they want—”show me customers whose behavior changed most in the last 30 days”—while AI agents automatically translate those requests into the appropriate execution plan. That plan may involve structured filters, vector similarity search, hybrid queries, or a combination of techniques that would otherwise require deep SQL expertise.
This shift expands access to operational data across product, analytics, and business teams, reducing reliance on specialists to translate requirements into executable SQL.
SQL Falls Short for Vector and Multimodal AI Workloads
AI applications increasingly operate on vector embeddings—semantic representations of text, images, audio, logs, and multimodal content. These workloads cannot be expressed cleanly in relational syntax or executed efficiently by SQL query planners.
Internal Zilliz benchmarks show that purpose-built Milvus outperforms PostgreSQL with pgvector, delivering 60% lower latency and 4.5 times higher throughput under identical vector search conditions. As organizations scale into billions of embeddings and millisecond-level latency requirements, the performance gap becomes even more pronounced.
“SQL was never designed for similarity search across thousands of dimensions,” Luan added. “AI workloads require a semantic retrieval layer, not a relational one.”
Enterprise Adoption Shows the Shift Is Already Underway
More than 10,000 organizations globally now use Milvus and Zilliz Cloud (managed Milvus) for AI applications, including semantic search, recommendation systems, retrieval-augmented generation (RAG), and agentic AI architectures. These deployments routinely operate at billion-vector scale with sub-10ms query latency across AWS, Google Cloud, and Microsoft Azure.
