Supabase Founders: Agents Now Launch 5M Databases a Month

How Coding Agents Became Supabase's Biggest Customer

Supabase co-founders Paul Copplestone (CEO) and Ant Wilson (CTO) sat down with swyx on Latent Space ahead of Supabase Select, the company's customer conference. swyx was their first angel investor six years ago; swyx believes this is the first interview the two have done together. The conversation is less about Postgres internals than about what happens to a developer-tools business when AI agents, not developers, start creating most of the accounts.

On scale: "At the moment we launch about 5 million databases on our platform every single month." The founders expect that number to reach billions per month, which is why they bought Turso, the SQLite-based database company built for many small databases. Its technology will power a new small instance type called Supabase Spark: one cheap database per prototype, upgradeable to full Postgres and then to Multigres, their open-source sharding layer built with Vitess creator Sugu Sougoumarane.

On agents doing activation: Consumer-style traffic usually lowers activation and paid conversion. At Supabase both went up, and the founders give the reason plainly: "the sign up to activation, the agent does it for you." Agent-built projects convert to paid at a higher rate than their traditional developer users, and some customers are making real money within the first week because payments are added by the agent too.

On sales with a control group: Supabase runs what it calls product sales. Each week roughly a thousand product triggers fire, such as an account scaling up, hitting limits or throwing errors, and each gets an email. Those who don't respond become the control group, and they grow 25% month over month. Those who respond get a call from a salesperson armed with an AI-written script based on their traffic patterns, and their accounts grow 125%. "So that 100% uplift, the incremental uplift is what we compensate our salespeople."

On declarative schemas for agents: Postgres's standard workflow is a growing folder of migration files. Ask an agent the shape of a table and it has to read a thousand of them. Supabase's answer is a fully declarative schemas folder that captures tables, policies and comments, backed by a new open-source library, PG Delta, that works with any Postgres database. The founders also credit Lovable with being first to write skills that stopped agents from doing damaging things like turning off row-level security to make a project work.

On their own AI spend: Asked for a benchmark, the founders checked their own documentation: about $300,000 a month on AI for roughly 450 people. Teams pick their own tools, and a usage leaderboard shows the usual skew toward a few heavy users. Their view is that spend is the wrong metric: "The focus shouldn't be on how much you're spending". What counts is how many end-to-end processes no longer need a human in the loop.

6 Takeaways from Supabase on Building for AI Agents

• Agents are the new top of funnel - Supabase's signups exploded this year with almost no marketing spend, driven by coding agents and app builders like Lovable and Bolt choosing it by default.
• Agent-led activation converts better - Because the agent completes setup in one step, activation is near 100% and paid conversion beats traditional developer signups.
• Pay sales for incremental lift, not gross revenue - A non-responding control cohort sets the baseline (25% monthly growth), and salespeople are paid on the extra 100% they add.
• Enterprises buy greenfield, not migrations - Supabase expects almost no production database migrations for the next five years. It wins by landing new projects, now often started by product leaders building internally with Lovable and Supabase.
• Make infrastructure readable as files - Declarative schemas let an agent read the current state of a database from Git instead of replaying migrations.
• Measure processes removed, not tokens spent - $300K a month for 450 people is a data point, but the founders judge AI by how many human-in-the-loop steps disappear.

What Supabase's Numbers Mean for AI-Native Companies

Supabase is a clear case of a company whose customer is increasingly an agent acting for a human, and it has rebuilt product, sales and infrastructure around that. The lessons travel: make your product the default an agent picks, let the agent finish onboarding, measure what humans add with a real control group, and judge internal AI by the end-to-end work it removes. For the wider shift, see solo AI founders and vibe coding.