The agentic era of Postgres databases
Summary:
David Mytton sorts Postgres products by who creates the database and how long it lives. First came provisioned instances like RDS: sized up front, paid for whether used or not, kept for years. Then Neon made the database an API, with compute that starts in milliseconds and scales to zero, and copy-on-write branches so each pull request can get its own database. Now the customer is an agent. When Databricks bought Neon, it said over 80% of Neon databases were being created by AI agents, up from 0.1% two years earlier, and Supabase reports about 70% of its four million new databases a month come from agents or AI tools. One response is to make databases tiny and disposable, like Cloudflare's D1 or PGlite, Postgres compiled to WebAssembly in under 3 MB. The other is still mostly missing: cloning production, anonymizing the sensitive fields, handing the copy to an agent and throwing it away. Mytton argues whoever makes that routine will matter most in this era. It is the right gap to name. Our own agents work against a separate development database for exactly this reason, and a realistic, safely anonymized copy would make their testing better.
Excerpt:
"David Mytton on how agents became the main creators of Postgres databases, and why safe, anonymized copies of production data are the missing piece."
#Database#AI Agents#Coding Agents#SQLite#Privacy
Read Full Source