From Queries to Agents: The Next Era of Data Retrieval on PostgreSQL | POSETTE 2026

Abe Omorogbe explains how PostgreSQL is evolving into a backbone for production AI agent workflows, focusing on reliable and safe data retrieval. He covers MCP-based agent patterns, common failure modes when agents generate SQL, and emerging approaches like context correction and blended retrieval across relational, vector, and graph techniques.

Overview

In this POSETTE 2026 talk, Abe Omorogbe (Microsoft) discusses how AI agents are moving beyond simple RAG patterns and why the hard part in production is dependable, context-aware retrieval rather than the model itself.

Key themes covered in the session:

PostgreSQL as an agent retrieval backbone

• PostgreSQL is positioned as a core system for agent workflows, not just a query engine.
• The talk frames retrieval as a multi-step, multi-tool process where agents need controlled access to data.

Model Context Protocol (MCP) for agent-to-data/tool connectivity

• MCP is presented as a standard way for agents to interact with tools and data sources (including PostgreSQL) via MCP servers.
• The session describes how agents interact with Postgres today using MCP servers and why standardization matters for building agent systems without bespoke glue code.

What goes wrong when agents generate SQL blindly

• The talk highlights failure modes when agents attempt to generate SQL without sufficient grounding or constraints.
• It calls out issues such as hallucinations, semantic mismatches, and the need for guardrails.

Context correction and trustworthy retrieval

• Retrieval is described as increasingly requiring robust context correction so agents can produce reliable results.
• The session emphasizes building retrieval layers that provide controlled, high-quality access to Postgres.

Blended retrieval: combining multiple retrieval modes

• The talk describes blended retrieval approaches that combine:
  • Relational SQL
  • Vector similarity search
  • Graph-aware traversal
• The goal is to give agents a more complete and dependable view of data than any single retrieval method.

Azure Database for PostgreSQL AI capabilities and HorizonDB direction

• Abe references Microsoft’s work in Azure Database for PostgreSQL, including:
  • Vector Search
  • GenAI capabilities
  • DiskANN-based vector indexing
  • MCP integrations
  • Semantic operators
  • Azure HorizonDB (described as newly announced)
• The session outlines a roadmap toward more unified retrieval layers (including a concept of a unified AI.search() capability in HorizonDB).

Resources

• POSETTE conference site: https://posetteconf.com
• POSETTE playlist: https://aka.ms/posette-playlist