Google Tells Its PostgreSQL Engineers To Rely Heavily On AI

Google is urging its database engineers to use AI coding tools extensively as it steps up work on PostgreSQL and other open source database projects, while stressing that human developers remain on the hook for every line of code that ships.

Sailesh Krishnamurthy, VP of Databases at Google Cloud, said the company is seeing “huge” productivity gains internally when developers lean on AI to write and interpret code for open source systems such as PostgreSQL. But he was clear that accountability does not shift to the machine.

“We do encourage folks to use AI heavily,” he said, noting that Google’s engineers are already applying these tools across a spectrum of use cases, from small snippets to entire drafts generated by AI. “Either way, the accountability remains on behalf of the person who’s done it.”

Google announced a wave of new PostgreSQL contributions earlier this year, part of a broader industry trend that has seen the open source database become a default choice for new cloud applications. PostgreSQL was the most popular database among developers in 2023, according to Stack Overflow’s annual survey, buoyed by a growing number of managed PostgreSQL services from the major cloud providers.

Krishnamurthy argued that AI coding tools are particularly well suited to open source work because the models have already been trained on public codebases and “have a better sense of the code” than they do with proprietary, in-house systems that live behind corporate firewalls.

That makes AI a practical assistant for engineers navigating and extending large, mature projects such as PostgreSQL. Still, he said Google expects its engineers to use these tools “heavily, but also judiciously” and to personally stand behind their submissions, regardless of how much of the draft came from an AI system.

PostgreSQL’s extensible architecture also lends itself to what Krishnamurthy described as a good “sweet spot” for AI-assisted work: taking a well-understood academic idea, combining it with a well-understood codebase, and quickly prototyping isolated extensions with a limited “blast radius.” In those cases, AI can help interpret existing code and scaffold new functionality faster, while engineers remain responsible for validation and review.

PostgreSQL’s rise is drawing heavy investment from across the industry. Google, Microsoft, AWS and others all offer PostgreSQL services and are adding new extensions and capabilities on top of the core open source project.

On the Microsoft side, the company has contributed pg_documentdb_core, a PostgreSQL extension that brings support for BSON (Binary JSON) documents, and pg_documentdb_api, a data layer that exposes MongoDB-compatible CRUD operations, queries and index management. These are set to run on Azure Cosmos DB for PostgreSQL, effectively turning it into a document-store-style database positioned as an alternative to MongoDB. Microsoft has also announced a distributed PostgreSQL database service called HorizonDB.

Krishnamurthy said this level of activity reflects a broad “industry at large” shift towards PostgreSQL. He pointed to adoption across customers and digital-native services, and to migrations away from commercial and legacy databases. Organisations are moving from Oracle, Microsoft SQL Server and IBM Db2, as well as older platforms such as Sybase and Informix, he said.

For Google, the current engineering focus in PostgreSQL is on advancing logical replication. Among its recent contributions are:

• Automatic Conflict Detection, designed to allow a replication worker to automatically detect when an incoming insert, update or delete conflicts with the local state.
• Logical replication of sequences, expanding the kinds of database objects that can be reliably replicated.

These improvements target scenarios where data is distributed and constantly changing, an increasingly common pattern for cloud-native architectures.

The competitive landscape around databases is also shifting. Research from Gartner cited by Krishnamurthy shows that of the leading database vendors 15 years ago; Oracle, IBM, Microsoft and SAP—only Microsoft has grown its market share since. Today, AWS leads the database market, offering both its own engines and managed PostgreSQL and MySQL services. Oracle remains third, ahead of Google, and that ranking is unlikely to change quickly.

Even so, the momentum behind open source databases is building as all the major cloud providers contribute to projects such as PostgreSQL while simultaneously selling commercial services built on top of them. Krishnamurthy framed PostgreSQL as an increasingly common “layer” in the data stack, regardless of where the data ultimately resides.

Within that context, Google’s message to its own engineers is that AI will be a core tool for keeping pace but never a shield from responsibility.