Supabase’s Plan for a Billion AI-Built Apps

Supabase co-founders Paul Copplestone and Ant Wilson appeared on the Latent Space podcast with host swyx to discuss how the backend platform is preparing for a surge of software generated by AI agents. The episode was recorded ahead of Supabase Select 2026, the company's event referenced in the show notes.
The central premise of the discussion, as framed by the episode, is that agents are producing more software, which in turn changes what a backend has to do. Two design directions came up repeatedly: running long-lived workloads alongside the database rather than apart from it, and shaping schemas so that agents can understand them more easily. Supabase Compute is the product tied to the first idea, described in the show notes as a way to run workloads alongside the database.
On reliability, the conversation covers Postgres fault tolerance and scaling. Multigres, described as highly available Postgres infrastructure, is in private alpha. The episode also touches OrioleDB and the Postgres storage engine, suggesting the storage layer is part of the reliability story. Separate from that, Supabase has acquired Turso, which the show notes frame around a scale-out opportunity.
The chapters sketch a broader arc for the company: "Six years in the making," "From prototype to petabyte," "AI and Supabase's growth," and "Agent-driven activation." Others cover community and crowdfunding, open source and enterprise sales, durable workflows, company priorities, and hiring for a remote-first company. Two framing questions stand out: whether the application builder or the database will be stickier, and whether AI-generated software will converge. There is also a segment on how Supabase teams use AI internally.
For developers, the interesting thread is the schema question. A declarative, code-first schema that an agent can read and reason about could let generated applications wire themselves into a database with less hand-holding, which fits the "agent-driven activation" theme. This is my reading rather than a stated claim: the more legible a schema is to a model, the less custom plumbing an agent needs to produce a working app, and the more of the value sits with the data layer. That would support the idea that the database, not the builder, is the stickier piece.
A likely trade-off is that optimizing a backend for agents may pull it away from the ergonomics human developers expect, though the episode's emphasis on code-first infrastructure and declarative schemas suggests Supabase sees the two audiences as aligned. Scale-out via Turso and high availability via Multigres, still in private alpha, are the operational bets.
Why it matters: teams building on Supabase, and anyone shipping agent-generated software, should watch how Compute, Multigres and the Turso acquisition mature, since they affect where workloads run and how applications scale. In practice, schema design may become as much an agent-facing interface as a human one, which is my inference from the topics discussed.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
AI agents are creating more software—and that changes what a backend needs to do. Supabase co-founders Paul Copplestone and Ant Wilson join swyx to discuss running long-lived workloads alongside the database, making schemas easier for agents to underst...