Why Agents Love Supabase

What it claims

This is a Latent Space interview with Supabase co-founders Paul Copplestone and Ant Wilson, their first podcast together. The company is described as open-source, fully remote and valued at $10 billion. The host says their own "frontier AEO tracker" found Supabase was the first choice for app backends in an almost complete sweep of frontier models, naming Astra, Fable and Muse. It was also in contention for managed database. The episode covers announcements from the Supabase Select event, which the company grouped into Build, Operate and Scale.

How it works

The Build announcements each remove a setup step that an agent repeats every time it creates and tests an environment.

• Local runtime (alpha): runs Supabase locally without Docker.
• Declarative schemas: the desired database state lives in files in Git, instead of a chain of migrations. The host compares this to Terraform. An agent can read, edit and fork the schema alongside application code without replaying migration history.
• Supabase Compute (private alpha): runs containers next to the database. It covers longer-running work such as chunking documents or generating embeddings, which edge functions do not cover. This means fewer separate services to configure and connect.

The Operate announcements expose logs, checks, connections and context. These tell an agent what the database contains, how the app uses it and why something failed.

The Scale announcements are OrioleDB, Multigres and the acquisition of Turso. Multigres is Supabase's approach to highly available Postgres and is in private alpha. The founders treat two problems as separate: serving many more small databases, which they say is about five million launched a month and could reach billions, and supporting individual customers whose databases grow very large.

The founders explain the better activation and conversion from AI-driven signups this way: an agent takes the user from signup to a working setup in one flow. The user does not have to work through each configuration step first.

Limits

The page is episode notes. It gives no benchmarks, pricing, commands or performance figures for any announced feature. The local runtime is alpha, and Compute and Multigres are private alpha. The conversion claim is the founders' statement and has no figures. The model-preference result comes from the host's own tracker, and its method is not described. Chapter markers place the local development discussion at 58:43, compute at 6:39 and OrioleDB/Multigres at 8:05.