In the rapidly evolving landscape of artificial intelligence, the race to develop AI applications and agents is intensifying. Organizations are exploring various tools to enhance functionalities ranging from customer support to internal document retrieval. However, the effectiveness of these AI solutions hinges on their access to existing enterprise data, such as customer records and support inquiries. Without this critical information, AI agents may resort to generating inaccurate responses, a phenomenon often referred to as “hallucination.”
Jensen Huang, co-founder and CEO of Nvidia, emphasized the importance of structured data during his recent keynote at GTC, describing it as the “ground truth for AI.” This sentiment is echoed by Phillip Merrick, co-founder and Chief Product Officer at pgEdge, who asserts that AI applications must reference existing data to avoid inaccuracies. He notes, “Structured data really is the ground truth for AI… It’s how you make sure your LLM and your agents aren’t hallucinating; you point them at the actual data.”
Postgres leads the pack for AI workloads
When it comes to selecting a database for AI applications, PostgreSQL stands out as a preferred choice. According to the Stack Overflow 2025 Developer Survey, 66% of respondents indicated that they have worked with PostgreSQL in the past year and wish to continue doing so. Merrick describes PostgreSQL as “elegant; it’s easy to get started with and easy to use and develop against,” solidifying its reputation as the database of choice for AI applications.
Merrick highlights several advantages of PostgreSQL, particularly its open-source model, which he argues is genuinely community-driven. He contrasts this with many proprietary databases that claim to be open source but often impose commercial limitations, potentially leading to vendor lock-in for developers. Scalability is another vital aspect for AI workloads, and Merrick points out that PostgreSQL is well-equipped to handle these demands, noting that even OpenAI’s API is built on PostgreSQL.
A one-stop shop for AI data and workloads…
Beyond its relational capabilities, PostgreSQL’s extension architecture allows it to integrate various database patterns seamlessly. As the demand for custom-built vector databases grows, Merrick emphasizes the benefits of utilizing a general-purpose database like PostgreSQL with a vector extension such as pgvector. He states, “Pgvector is just as performant as those standalone vector databases, but with the advantage that it’s standard Postgres. It’s what you’re using for all your other data.”
While some developers may opt for proprietary databases for speed, Merrick believes the need for such specialization is diminishing. He acknowledges the significance of unstructured data, such as support tickets and knowledge bases, in developing effective AI applications. PostgreSQL can accommodate both structured and unstructured data, allowing developers to feed documentation into the database, utilize pgvector for embeddings, and create chatbots leveraging retrieval-augmented generation (RAG).
He cites the pgEdge Agentic AI Toolkit for Postgres as an example of a robust infrastructure that supports the entire lifecycle of AI application development, from document ingestion to embedding generation and retrieval. Merrick explains, “The [pgEdge] docloader takes care of getting your documents into Postgres,” facilitating a streamlined process for developers.
Moreover, he urges developers to consider ongoing vector maintenance, as many overlook the necessity of keeping vectors updated when content changes. “People don’t realize they’re going to have to worry about it because, out of the box, things like pgvector don’t give you automatic updates of vectors when the underlying content changes,” he warns.
“People don’t realize they’re going to have to worry about it because, out of the box, things like pgvector don’t give you automatic updates of vectors when the underlying content changes.”
To address this, pgEdge’s toolkit includes the pgEdge Vectorizer, which automatically generates vector embeddings and maintains them as content evolves, along with a comprehensive Model Context Protocol (MCP) server that connects to both new and existing PostgreSQL databases.
…with security for mission-critical industries
Whether utilized independently or in combination, the technologies within the pgEdge toolkit significantly simplify the development of AI applications on PostgreSQL. Merrick asserts that PostgreSQL is the optimal choice for enterprises, delivering high availability, data sovereignty, and robust security essential for scaling AI applications in critical sectors such as finance, healthcare, telecommunications, and government.
He attributes these advantages to PostgreSQL’s deployment flexibility, which allows organizations to run applications on managed cloud databases, self-managed cloud accounts, or on-premises solutions. “So you’ve got that complete flexibility,” Merrick states, emphasizing that not all PostgreSQL vendors offer this level of adaptability—pgEdge does.
In addition to flexibility, PostgreSQL’s established security features are a significant draw for developers. Merrick highlights that pgEdge enhances security with user-based and token-based authentication, encrypted traffic via TLS, and a default read-only mode for its MCP server. These security measures, combined with PostgreSQL’s distributed capabilities, help enterprises mitigate data leaks and comply with regulations like GDPR, solidifying PostgreSQL’s position as a leading choice for AI application development.
As the demand for AI solutions continues to grow, the onus is now on enterprises to invest in the right infrastructure to support their ambitions.
