Neon enabled agentic database provisioning
Charles Chretien, co-founder of Prequel, on the modern data stack’s ROI problem
This points to a new database buyer, software that creates infrastructure on its own. Neon won early because its architecture lets an app or coding agent create a fresh Postgres instance or branch with an API call, without copying all the underlying data first. That makes database creation cheap enough to happen inside CI jobs, coding tools, and AI app builders, so usage grows from many small automated actions instead of a single human admin decision.
-
Neon separated compute from storage and made branching a core feature, so a developer or agent can spin up a new database for a pull request, test run, or app session while reusing the same base data. That is the concrete product reason agentic workflows turned into revenue.
-
The scale of the shift was large. Database count grew from 20,000 in early 2023 to more than 700,000 by April 2024, and internal telemetry showed over 80% of databases were being provisioned by AI agents rather than people. That is why Databricks saw Neon as a way to monetize operational infrastructure for AI apps.
-
This also explains the contrast with older database economics. Traditional managed Postgres is sold around long lived instances and human setup. Neon used usage based pricing and scale to zero behavior, so thousands of short lived databases created by agents could still be economical for customers and meaningful for Neon.
Going forward, the winning database products will look less like rented servers and more like callable building blocks inside software workflows. That favors Postgres platforms that are API first, instant to provision, and cheap when idle. It also pushes data platforms like Databricks to own both the analytical layer for models and the transactional layer where AI apps actually run.