PostgresWorld Podcast EP02: Giving AI Agents Their Own PostgreSQL Database

How can an AI agent write reliable database code if it knows so little about the database it will run against? In Episode 2 of the Postgres World Podcast, Amir More, Founder and CEO of Baseshift, joins JD to discuss what AI coding agents miss when they work without production data. An agent may know PostgreSQL, but understanding your data and how an application uses it is another matter. Amir explains how Baseshift brings masked samples of production data into separate environments where agents can experiment without connecting to production. They discuss the practical challenge of keeping those copies fresh, and why giving every agent its own database gets expensive as teams adopt more agents. JD also shares an experience with Claude that turned a request to move tests to pytest into a wrapper around the existing shell scripts. That leads to a broader conversation about why AI makes unexpected choices and how much direction it still needs. The episode closes with a look at how Baseshift uses change data capture and lightweight, copy-on-write clones to make those environments available through Docker, using standard PostgreSQL. 💻 Baseshift: baseshift.com 🔗 Amir More: linkedin.com/in/amirmore