Enterprise PostgreSQL

Enterprise PostgreSQL Is More Than a Database, Says Phillip Merrick

Governance and Enterprise Control: As enterprise systems become more distributed and autonomous, managing risk and maintaining control has become increasingly complex. This series explores how organizations approach governance, security, compliance, data strategy and accountability across technology environments. It focuses on the structures, processes, and decisions that help enterprises operate responsibly while adapting to regulatory and operational change.

Enterprise data strategies are entering a new phase. As organizations scale AI initiatives, modernize legacy applications, and navigate increasingly distributed environments, CTOs are under pressure to build database infrastructure that is resilient, scalable, and free from unnecessary vendor lock-in. For many, the conversation is no longer about replacing existing databases, but about evolving them to support modern enterprise workloads without compromising flexibility.

Against this backdrop, Enterprise PostgreSQL is gaining momentum as a foundation for modern data architectures.

Phillip Merrick, CEO and co-founder of pgEdge, shares his perspective on why PostgreSQL is becoming increasingly relevant for enterprise AI, how organizations can improve resilience without sacrificing flexibility, and what the next phase of database modernization means for technology leaders.

Phillip Merrick co-founder and CEO, pgEdge, on Enterprise PostgreSQL
Phillip Merrick co-founder and CEO, pgEdge, on Enterprise PostgreSQL

Why enterprise AI needs a different database strategy

Enterprise AI is no longer just experimental. Companies are now building real systems powered by AI agents that constantly gather, analyze, and use operational data. This change puts new demands on databases, such as needing high availability, strong governance, and flexible deployment in the cloud, on-site, or in regulated settings.

While many discussions about AI focus on the models, the data platform underneath often decides if an AI project will scale or get stuck before reaching production. For companies updating old systems, the real challenge is not picking a database, but choosing an architecture that is open, resilient, and efficient.

pgEdge positions itself as an enterprise-grade, open source PostgreSQL platform built for AI applications. What differentiates your approach from the broader PostgreSQL ecosystem?

Phillip Merrick: At pgEdge, we’re focused on delivering an enterprise-grade, 100 percent open source PostgreSQL platform designed specifically for modern AI applications.

When we say enterprise grade, we mean a platform that’s hardened for production, supports high availability for mission-critical workloads, and provides the security and compliance capabilities enterprises require. Flexible deployment is equally important. Organizations should be able to run the same PostgreSQL platform in our cloud, their own cloud, on-premises, or even in completely air-gapped environments. We work with defense and intelligence customers whose systems cannot be connected to the internet, so supporting those deployment models is essential.

Beyond that foundation, we’ve invested heavily in AI tooling around PostgreSQL.

Our Agentic AI Toolkit includes an MCP server, a RAG server, vector extensions, and a vectorizer that automatically creates and maintains vector embeddings directly inside PostgreSQL.

On the operational side, we offer the AI DBA Workbench, which includes an AI agent we call Ellie. Ellie continuously monitors PostgreSQL databases without requiring predefined thresholds, detects anomalies, diagnoses issues quickly, and recommends fixes typically as executable SQL. Customers review the recommendation, press a play button, and the issue is resolved.

We could automate that process completely, but today most enterprises want to build trust in AI agents before moving to fully autonomous operations.

As enterprises deploy increasingly autonomous AI agents, where do you see the biggest database bottlenecks? Are organizations still overlooking critical gaps in their AI data infrastructure?

Merrick: There are several gaps we’ve consistently seen while working with customers.

The first is how AI agents access enterprise data. Agents should operate from the organization’s source of truth, which is typically its operational databases. PostgreSQL is excellent for that. Even when the underlying systems aren’t PostgreSQL, it serves as an effective gateway through technologies like Foreign Data Wrappers and Apache Iceberg interfaces for data lakes and warehouses.

We recently introduced a technology called Cold Front that makes integrating PostgreSQL databases with Iceberg warehouses and data lakes much easier.

Another area people underestimate is MCP architecture. Connecting an MCP server to PostgreSQL is becoming the standard approach, but if the server isn’t designed properly, token consumption can increase dramatically. Every enterprise is concerned about token costs. Our MCP server is engineered to minimize token usage while providing the governance and security controls enterprises require.

The other common mistake happens during application development. Teams often prototype using developer-friendly cloud platforms or AI coding tools like Lovable. Those are excellent for experimentation, but they’re frequently built around cloud database platforms that aren’t suitable for regulated production environments.

If organizations don’t think about their production data platform early, they fall into the AI production gap, where promising prototypes never make it into production. In many cases, the underlying data infrastructure is the reason.

Finally, AI coding tools such as Claude Code, Lovable, and Replit are generating databases faster than ever. The real question becomes: how do you operate and manage all of those databases once they’re running in production? That’s exactly the operational challenge our AI DBA Workbench addresses.

PostgreSQL has become the default choice for many modern applications. Do you believe it’s also becoming the foundation for enterprise AI, or will AI ultimately require an entirely new database architecture?

Merrick: Across the industry, there’s a growing consensus: just use PostgreSQL.

That has become something of a mantra because PostgreSQL has proven to be an extremely effective solution.

Even before the agentic AI era, PostgreSQL had already become the database of choice for developers. It was also the preferred destination for enterprises migrating away from legacy platforms like Oracle and SQL Server. Naturally, it became the foundation for agentic coding platforms such as Claude Code and Lovable.

That momentum reinforces PostgreSQL as the sensible default data platform for AI applications. Some people still assume they need a specialized vector database like Pinecone. However, PostgreSQL has supported vector capabilities through PGVector for several years. Initially, dedicated vector databases had performance advantages, but today PGVector delivers comparable, and in many cases better, performance.

We’re obviously a PostgreSQL company, so you could say we’re biased. But I think most of the industry has arrived at the same conclusion: just use PostgreSQL.

Sustainability has become a boardroom priority alongside AI. Can distributed database architectures help organizations achieve both AI scalability and lower environmental impact?

Merrick: Sustainability means different things to different organizations. If we’re talking about reducing environmental impact, one practical proxy is token usage. Whether the motivation is controlling cost or reducing energy consumption, enterprises want to minimize token usage while maximizing the value they get from AI applications.

That’s one reason our Agentic AI Toolkit and MCP server focus heavily on token efficiency.

Enterprise architecture discussions increasingly revolve around centralized cloud platforms versus distributed edge deployments. Over the next five years, where do you see that balance shifting?

Merrick: Distributed PostgreSQL has actually been part of our platform from the beginning. We’ve always believed there’s significant value in geographically distributed databases and applications.

One major advantage is resilience. If an entire cloud region goes offline, as we’ve seen happen from time to time, you need the ability to fail over quickly to another region. Distribution is also important for data sovereignty.

Global organizations operate across Europe, Asia, and North America, where regulations often require certain customer data to remain within national or regional borders. A distributed PostgreSQL architecture lets organizations separate globally shared data, such as product catalogs, from customer-specific data that must remain local. That flexibility helps organizations meet residency requirements while maintaining a unified application architecture.

As AI consumes enterprise data at unprecedented scale, is proprietary data becoming a greater competitive advantage than the AI models themselves?

Merrick: A company’s competitive advantage increasingly comes from its data. That includes customer information, business processes, and the rules governing that data.

Models themselves aren’t becoming commodities, but organizations have far more choice than they did previously. Simply using the same foundation model as everyone else isn’t going to create differentiation. The real advantage comes from proprietary enterprise data.

Organizations can use that data as the source of truth for agentic applications, or they can go a step further by training or fine-tuning their own models on private data. That allows them to transform their data advantage into a model advantage, although doing that requires a higher level of technical maturity.

If a CTO is building an AI-ready data architecture today, what foundational principles should they prioritize from day one?

Merrick: The first thing to recognize is that every established organization already has data spread across multiple databases and formats. Rather than replacing everything, the goal should be federation.

In the agentic AI era, you want AI agents to interact with a single data interface that provides access to all enterprise information. PostgreSQL is uniquely positioned for that. It doesn’t just connect to PostgreSQL databases. Through Foreign Data Wrappers, it can connect to Oracle, SQL Server, MySQL, and many other systems. Our CTO was instrumental in developing that capability, so it’s an area where we have deep expertise.

Those external tables appear as native PostgreSQL tables, allowing applications and AI agents to access them through a consistent interface. Because PostgreSQL has such a large open-source ecosystem, today’s AI models are already exceptionally well trained to work with it.

By exposing data through PostgreSQL, integrating Iceberg warehouses and data lakes, and using technologies like Cold Front, organizations create a unified data layer for AI without rebuilding everything.

In enterprise database modernization projects, what’s the most overlooked risk you’ve seen organizations underestimate?

Merrick: One of the biggest risks is losing sight of why you’re modernizing in the first place. For most organizations, moving away from legacy platforms like Oracle or SQL Server is largely about escaping vendor lock-in.

PostgreSQL solves that problem, but only if organizations remain committed to genuine open source.

Some vendors market themselves as open source while relying heavily on proprietary extensions that recreate the same dependency customers were trying to escape. Cloud providers often take a similar approach. They’ll offer a standard PostgreSQL service alongside an enterprise version that appears PostgreSQL-compatible but is actually proprietary.

That’s why I’d encourage organizations to pay close attention to one phrase: PostgreSQL compatible. If a platform is merely PostgreSQL compatible, then by definition it isn’t PostgreSQL. That’s often where proprietary lock-in begins.

Looking ahead, will enterprise AI ultimately run on centralized databases, distributed architectures, or a hybrid model?

Merrick: I think the future is unquestionably hybrid. The right architecture depends on the application and the business context. Centralized systems offer simplicity, but they also introduce a single point of failure.

Distributed systems improve resilience, but they create more complex governance challenges. Organizations need the flexibility to use both approaches where they make sense. If you do centralize, make sure your disaster recovery and failover strategy is robust enough that a single failure never becomes a business failure.

If you were designing a database platform from scratch today, knowing everything we’ve learned from AI over the past year, what would you build differently?

Merrick: I may be biased, but I wouldn’t reinvent the wheel. PostgreSQL is already an elegant wheel. There’s no need to reinvent it.

Phillip Merrick Co-founder, Chief Product Officer, pgEdge: Phillip is a technologist, entrepreneur and CEO with over three decades of experience in the enterprise software and Internet industries. Companies he has co-founded and/or run include webMethods, EnterpriseDB (EDB), SparkPost and Fugue, all of whom achieved successful nine or ten figure exits. At webMethods Phillip co-invented web services and web APIs, under his leadership, the company had a highly successful IPO and became the fastest growing US software company. He is currently Co-founder, Chief Product Officer and Chairman of the Board of pgEdge, a Postgres database startup launched in 2023. 

Rajashree Goswami is a professional technology writer with 13+ years of experience covering AI, cybersecurity, cloud computing, SaaS, fintech, regtech, healthtech, sustainable technology, digital transformation, and enterprise innovation. She also specializes in software and app analysis, emerging technologies, and enterprise technology trends. Her work is grounded in research and in-depth conversations with industry leaders, subject matter experts, and technology practitioners, with a focus on the business impact of technology on innovation, operational efficiency, growth, and ROI.