There are still some fundamental truths that won't change. LLMs are making it dramatically easier to build software, so yes, we're going to have more applications, more agents, more workloads, and more data. But here's one thing that won't change: developers want to build on trusted foundations. They don't want to stitch together fragile, Rube Goldberg architectures. They want infrastructure they know, they trust, and they can bet their business on, for the next decade. That's why Postgres continues to win, even as AI reshapes the world.

Despite turning 40 years old this year, and having a name only a parent could love, Postgres is still the database developers love most. It's reliable. It's extensible. It's open. And over four decades it has grown into an incredible platform, spanning time-series, geospatial, vector, analytics, and much more, all backed by the largest developer ecosystem in databases.

The irony is that developers don't want innovation in their database. They want innovation everywhere else. They want their database to be dependable. Predictable.

Boring.

So we asked the people who know Postgres best: with agents and LLMs reshaping the stack at breakneck speed, why is a 40-year-old technology like Postgres still dominating?

Mike Freedman, Co-Founder and CTO of Tiger Data
Postgres scales because it was built on first principles: transactional correctness, query planning that actually works, semantics you can depend on, and a robust ecosystem of open source extensions. The teams seeing 200x performance improvements aren't jumping to new systems. They're understanding their workloads deeply and letting Postgres do what it was built to do. That doesn't change with LLMs. If anything, it matters more.

Postgres actually made a bet 15 years ago for growth and experimentation with its extension model. This fits massively into the concept of open-source development and experimentation: Let a 1000 flowers bloom. Allow people to customize for use cases, and allow open-source ecosystems thrive (including with "coopetition") rather than restricting decisions through a foundation/core developers. The core database which moves a bit slower, more deliberatively, safety and correctness above all. And then the extension framework, which allows rapid experimentation and innovation.

Chris Preimesberger, Freelance Journalist for The New Stack and SDxCentral
PostgreSQL has always been a keeper. This is largely because its architecture has proven remarkably adaptable to each new wave of computing, from client-server applications, to cloud-native platforms, and now AI-driven workloads. While newer databases often optimize for a single use case, PostgreSQL is astute: it has evolved into a highly extensible platform that supports transactional processing, analytics, geospatial data, vector search, and AI applications without forcing enterprises to abandon existing investments. Its open source governance model has encouraged a vast ecosystem of contributors, tools, and cloud services, giving enterprises confidence that they won't be locked into a single vendor. As agents and LLMs accelerate software development, PostgreSQL benefits from being the database developers already know, trust, and can integrate into modern architectures with minimal friction. While so much else is changing around it, PostgreSQL's combination of stability, flexibility, and trustiness has made it the default foundation upon which much of the new AI-powered stack is being built.

Stephen O'Grady, Principal Analyst & Cofounder of RedMonk
No single project characteristic is responsible for it. But PostgreSQL advocates and vendors typically point to the advantages of its broader community, which again is enabled by its liberal license. Multi-entity open source means more commercial options, which large enterprises favor because competition limits vendors' leverage with respect to pricing and it limits their risk to vendor behavior and changes. It can also mean broader project support and faster innovation, as evidenced by the speed at which PostgreSQL has been able to adapt to emerging market demands by adding the ability to handle new workloads from JSON to vector.

Carl Olofson, Independent Analyst (former Research Vice President at IDC)

Relational database systems like PostgreSQL are not competing with agents and LLMs, but rather cooperating. LLMs by themselves are great if all the data you want to access in your AI system is unstructured (text files, email, harvested collaborative system content, etc.), but if you want useful answers to questions involving the structured data that drive applications, and by extension, the enterprise, you must include databases that deliver current and consistent deterministic response with guaranteed 100% accuracy. SQL has been the preferred access language for decades, and the most commonly encountered SQL database system is PostgreSQL, largely because it is fairly easy to manage and most importantly, it's open source. Old does not necessarily mean outdated, and new does not mean better.

Paul Gillin, Enterprise Editor at SiliconANGLE
In an industry that considers "old" a dirty word, Postgres demonstrates that age can also mean reliability, familiarity and adaptability. In fact, AI has amplified the need for databases that adapt to multiple data types and can function as an operational store, metadata store, document store, search substrate and support vector embeddings. Postgres has adapted well to changing needs thanks to its flexible architecture and large, enthusiastic developer community. While it still lags Oracle, MySQL and Microsoft SQL Server in overall popularity, it is a true open-source alternative that appears to have the momentum to be a safe and reliable choice for many years to come.

Conclusion
Different vantage points, but the same conclusion keeps surfacing: Postgres isn't winning the AI era in spite of its age, it's winning because of what that age represents. Decades of battle-tested correctness, a governance model that spreads trust across a whole ecosystem rather than one vendor, and an extension architecture that lets new capabilities (vector search, real-time analytics, AI workloads) bolt onto a stable core instead of requiring a new database altogether.

Agents and LLMs may be changing how fast software gets built, but they haven't changed what developers need it to run on.