Google is encouraging its database engineers to use AI coding tools as much as they wish when contributing to open source projects like PostgreSQL, but with one strict condition: whoever commits the code is fully responsible for it, regardless of how much was generated by an AI model. Sailesh Krishnamurthy, Vice President of Databases at Google Cloud, explained in an interview that this policy allows Google to pursue productivity gains without compromising on accountability, code quality, or the engineer's duty for what ultimately gets shipped.
AI Tools and Open Source: A Natural Fit
Krishnamurthy noted that open source projects are particularly well-suited for AI-assisted coding because the models have already been trained on publicly available code. For PostgreSQL, which is open source, AI tools can understand the codebase and offer relevant suggestions. In contrast, proprietary code locked behind corporate firewalls lacks this advantage, making AI suggestions less reliable.
PostgreSQL's extensible architecture also makes it ideal for AI assistance. Engineers can work on extensions without touching the core, limiting the "blast radius" of any errors. Krishnamurthy emphasized that his teams use AI tools "quite heavily, but also judiciously."
Accountability Remains with the Engineer
The policy is clear: regardless of whether code is entirely drafted by AI or pasted from a suggestion, the engineer who commits it bears full accountability. "Whether you have a piece of code that is completely drafted by AI, or not even part of what you're pasting into your development environment, the accountability remains on behalf of the person who's done it," Krishnamurthy told The Register.
Google's Growing Reliance on AI-Generated Code
This guidance aligns with a broader trend at Google. In April, the company announced that 75% of new code created internally is now generated by AI and reviewed by human engineers, up from 25% in October 2024 and 50% last fall. CEO Sundar Pichai stated that one complex code migration handled by agents and engineers together finished six times faster than it would have a year earlier. Google is also piloting a new interview process that allows junior and mid-level software engineering candidates to use Gemini during a code comprehension round.
Focus on PostgreSQL Contributions
Google's contributions to PostgreSQL have focused on logical replication, including automatic conflict detection and the logical replication of sequences. The broader industry is moving in the same direction, with Microsoft shipping its own PostgreSQL extensions and building a distributed service called HorizonDB. Krishnamurthy noted that demand is coming from both new applications and migrations off legacy commercial systems like Oracle, SQL Server, Db2, Sybase, and Informix.
