← Home
The Changelog · August 25, 2026 · 1h 42m

Postgres at PlanetScale (Interview)

Sam Lambert returns to The Changelog after four years to discuss PlanetScale's new Postgres offering, the Neki proxy rewrite, and how database operations are evolving. He shares the reasoning behind PlanetScale's 'do-over' of Vitess for Postgres, the role of AI agents in schema management, and the company's capability to roll back enormous tables instantly. The conversation also touches on the open-source dilemma and Lambert's critique of the major cloud providers as lazy monopolies.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Novel

•
$5 PlanetScale Postgres
PlanetScale launched a new Postgres-compatible serverless database starting at $5, bringing its branching and developer workflow to the Postgres ecosystem.
•
Neki: the Vitess do-over for Postgres
Neki is PlanetScale's from-scratch rewrite of the Vitess proxy, designed specifically for Postgres after discovering that the MySQL‑oriented Vitess couldn't simply be adapted.
•
Agents shipping schema changes through deploy requests
PlanetScale now supports AI agents that autonomously propose and ship database schema changes via the familiar deploy-request workflow.

Highlights

•
Rolling back a 500TB table in seconds
PlanetScale demonstrated the ability to roll back a table as large as 500 terabytes in seconds, defying conventional expectations for big data operations.

Editorial

•
The open source dilemma
Lambert openly debated whether PlanetScale should open source its new infrastructure, weighing community benefits against business defensibility.
•
Break up the sleeping, lazy giants
Lambert argued that the major cloud providers — AWS, Google Cloud, and Azure — have become complacent monopolies that stifle innovation and should be broken up.

Misc

✧Sam Lambert returns after a 4-year absence from the podcast.
✧He refers to AWS, Google Cloud, and Azure as 'sleeping, lazy giants' that should be broken up.
Was this useful?